https://terms-en.ai-term-hub.com/zh/tags/optimization/Recent content in Optimization on 中文AI术语词典Hugozh-cnSat, 18 Jul 2026 11:44:45 +0000https://terms-en.ai-term-hub.com/zh/terms/curriculum_learning/Sat, 18 Jul 2026 11:38:39 +0000https://terms-en.ai-term-hub.com/zh/terms/curriculum_learning/<h2 id="definition">Definition</h2> <p>课程学习模仿人类教育过程，以结构化的顺序呈现训练数据，通常从简单样本开始，并逐渐增加复杂度。这种方法有助于神经网络更好地收敛并提高泛化能力。</p> <h3 id="summary">Summary</h3> <p>一种训练策略，模型先学习简单的示例，然后再逐步过渡到更难的示例。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>渐进式难度</li> <li>样本排序</li> <li>收敛速度</li> <li>泛化能力</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在复杂图像数据集上训练深度神经网络</li> <li>针对不同句子复杂度的语言建模</li> <li>具有稀疏奖励的强化学习任务</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transfer-learning-%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0/">Transfer Learning (迁移学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hard-negative-mining-%E9%9A%BE%E4%BE%8B%E6%8C%96%E6%8E%98/">Hard Negative Mining (难例挖掘)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/data-augmentation-%E6%95%B0%E6%8D%AE%E5%A2%9E%E5%BC%BA/">Data Augmentation (数据增强)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/self-paced-learning-%E8%87%AA%E6%AD%A5%E5%AD%A6%E4%B9%A0/">Self-Paced Learning (自步学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/voice_activity_detection/Sat, 18 Jul 2026 11:37:50 +0000https://terms-en.ai-term-hub.com/zh/terms/voice_activity_detection/<h2 id="definition">Definition</h2> <p>VAD 算法实时分析音频流，以区分活跃语音时段和非语音间隔（如背景噪声或停顿）。这对于优化带宽至关重要。</p> <h3 id="summary">Summary</h3> <p>语音活动检测（VAD）是一种信号处理技术，用于识别包含人类语音的音频片段与静音或噪声之间的区别。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>静音抑制</li> <li>信号分类</li> <li>实时处理</li> <li>抗噪性</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>高效的 VoIP 通信</li> <li>语音到文本引擎的预处理</li> <li>唤醒词检测系统</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/speech_recognition-%E8%AF%AD%E9%9F%B3%E8%AF%86%E5%88%AB/">speech_recognition (语音识别)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/noise_cancellation-%E9%99%8D%E5%99%AA/">noise_cancellation (降噪)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/audio_segmentation-%E9%9F%B3%E9%A2%91%E5%88%86%E5%89%B2/">audio_segmentation (音频分割)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/wake_word-%E5%94%A4%E9%86%92%E8%AF%8D/">wake_word (唤醒词)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/unsloth/Sat, 18 Jul 2026 11:37:39 +0000https://terms-en.ai-term-hub.com/zh/terms/unsloth/<h2 id="definition">Definition</h2> <p>Unsloth 是一款专为优化大语言模型（LLM）的微调和部署而设计的工具。它通过替换标准的 PyTorch 操作，实现了显著的速度提升和内存占用减少。</p> <h3 id="summary">Summary</h3> <p>Unsloth 是一个开源库，通过优化的内存管理和内核实现，将大语言模型（LLM）的训练和推理速度提升高达 2 倍。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>内存优化</li> <li>自定义内核</li> <li>LLM 微调</li> <li>加速推理</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在有限的 GPU 资源下微调 LLM</li> <li>加速推理流水线</li> <li>降低训练时的云计算成本</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> unsloth <span style="color:#f92672">import</span> FastLanguageModel </span></span><span style="display:flex;"><span>model, tokenizer <span style="color:#f92672">=</span> FastLanguageModel<span style="color:#f92672">.</span>from_pretrained( </span></span><span style="display:flex;"><span> model_name<span style="color:#f92672">=</span><span style="color:#e6db74">&#34;unsloth/Llama-2-7b-bnb-4bit&#34;</span>, </span></span><span style="display:flex;"><span> max_seq_length<span style="color:#f92672">=</span><span style="color:#ae81ff">2048</span>, </span></span><span style="display:flex;"><span> dtype<span style="color:#f92672">=</span><span style="color:#66d9ef">None</span>, </span></span><span style="display:flex;"><span> load_in_4bit<span style="color:#f92672">=</span><span style="color:#66d9ef">True</span> </span></span><span style="display:flex;"><span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora-%E4%BD%8E%E7%A7%A9%E9%80%82%E5%BA%94/">LoRA (低秩适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pytorch-%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0%E6%A1%86%E6%9E%B6/">PyTorch (深度学习框架)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hugging-face-%E6%A8%A1%E5%9E%8B%E7%A4%BE%E5%8C%BA%E5%B9%B3%E5%8F%B0/">Hugging Face (模型社区平台)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/flash-attention-%E9%AB%98%E6%95%88%E6%B3%A8%E6%84%8F%E5%8A%9B%E6%9C%BA%E5%88%B6/">Flash Attention (高效注意力机制)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/vllm/Sat, 18 Jul 2026 11:37:39 +0000https://terms-en.ai-term-hub.com/zh/terms/vllm/<h2 id="definition">Definition</h2> <p>vLLM（Virtual Large Language Model）是一个旨在加速 LLM 服务的开源库。它引入了 PagedAttention，这是一种受操作系统虚拟内存</p> <h3 id="summary">Summary</h3> <p>vLLM 是一个高吞吐量且内存高效的 LLM 推理引擎，利用 PagedAttention 优化 GPU 内存使用。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>PagedAttention</li> <li>KV Cache 管理</li> <li>推理服务</li> <li>吞吐量优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>高并发 API 服务</li> <li>批处理推理处理</li> <li>具有成本效益的 LLM 部署</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> vllm <span style="color:#f92672">import</span> LLM, SamplingParams </span></span><span style="display:flex;"><span>llm <span style="color:#f92672">=</span> LLM(model<span style="color:#f92672">=</span><span style="color:#e6db74">&#34;facebook/opt-125m&#34;</span>) </span></span><span style="display:flex;"><span>prompts <span style="color:#f92672">=</span> [<span style="color:#e6db74">&#34;Hello, my name is&#34;</span>, <span style="color:#e6db74">&#34;The capital of France is&#34;</span>] </span></span><span style="display:flex;"><span>sampling_params <span style="color:#f92672">=</span> SamplingParams(temperature<span style="color:#f92672">=</span><span style="color:#ae81ff">0.8</span>, top_p<span style="color:#f92672">=</span><span style="color:#ae81ff">0.95</span>) </span></span><span style="display:flex;"><span>outputs <span style="color:#f92672">=</span> llm<span style="color:#f92672">.</span>generate(prompts, sampling_params) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/tensorrt-nvidia-%E6%8E%A8%E7%90%86%E4%BC%98%E5%8C%96%E5%BA%93/">TensorRT (NVIDIA 推理优化库)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/tgi-%E6%96%87%E6%9C%AC%E7%94%9F%E6%88%90%E6%8E%A8%E7%90%86%E6%9C%8D%E5%8A%A1%E5%99%A8/">TGI (文本生成推理服务器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pagedattention-%E5%88%86%E9%A1%B5%E6%B3%A8%E6%84%8F%E5%8A%9B%E6%9C%BA%E5%88%B6/">PagedAttention (分页注意力机制)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/llm-serving-llm-%E6%9C%8D%E5%8A%A1%E5%8C%96/">LLM Serving (LLM 服务化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/token_maxxing/Sat, 18 Jul 2026 11:37:02 +0000https://terms-en.ai-term-hub.com/zh/terms/token_maxxing/<h2 id="definition">Definition</h2> <p>Token最大化涉及精心构造输入内容，以充分利用模型的上下文窗口容量，或通过优化Token的语义密度来提升性能。实践者可能会填充无关文本以占满上下文空间（Padding），或者极度压缩信息以提高单位Token的信息量。这种策略常用于在有限的计算资源下获取更精确的回答，或在API调用成本受限的情况下最大化单次请求的输出价值。</p> <h3 id="summary">Summary</h3> <p>一种提示工程优化策略，旨在大型语言模型中最大化特定Token数量的效用或输出质量。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>提示工程</li> <li>上下文窗口</li> <li>语义密度</li> <li>优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>最大化LLM推理能力</li> <li>降低API使用成本</li> <li>测试模型边界</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%B8%8A%E4%B8%8B%E6%96%87%E9%95%BF%E5%BA%A6-context-length/">上下文长度 (Context Length)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%8F%90%E7%A4%BA%E6%B3%A8%E5%85%A5-prompt-injection/">提示注入 (Prompt Injection)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%88%86%E8%AF%8D-tokenization/">分词 (Tokenization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%95%88%E7%8E%87%E8%B0%83%E4%BC%98-efficiency-tuning/">效率调优 (Efficiency Tuning)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/three_factor_learning/Sat, 18 Jul 2026 11:36:49 +0000https://terms-en.ai-term-hub.com/zh/terms/three_factor_learning/<h2 id="definition">Definition</h2> <p>三因子学习是强化学习中的一种特定方法，它将学习过程分解为三个不同的组成部分：奖励信号、价值函数和策略。奖励信号提供即时反馈，价值函数评估长期预期收益，而策略则决定智能体的行动选择。这种分解有助于更精细地调整和优化智能体在复杂环境中的行为表现。</p> <h3 id="summary">Summary</h3> <p>一种利用奖励、价值和策略因子来优化智能体行为的强化学习框架。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>奖励信号</li> <li>价值函数</li> <li>策略优化</li> <li>强化学习</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>游戏智能体</li> <li>机器人控制</li> <li>资源管理</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reinforcement-learning-%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">Reinforcement Learning (强化学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/q-learning-q%E5%AD%A6%E4%B9%A0/">Q-Learning (Q学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/policy-gradient-%E7%AD%96%E7%95%A5%E6%A2%AF%E5%BA%A6/">Policy Gradient (策略梯度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reward-shaping-%E5%A5%96%E5%8A%B1%E5%A1%91%E9%80%A0/">Reward Shaping (奖励塑造)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/tflite/Sat, 18 Jul 2026 11:36:20 +0000https://terms-en.ai-term-hub.com/zh/terms/tflite/<h2 id="definition">Definition</h2> <p>TensorFlow Lite 是一个开源框架，旨在将机器学习模型部署到智能手机、微控制器和物联网设备等资源受限的设备上。它通过优化模型推理过程来实现这一目标。</p> <h3 id="summary">Summary</h3> <p>TensorFlow Lite（TFLite）是一套使机器学习模型能够在移动设备、嵌入式设备和边缘设备上运行的工具集。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>模型量化</li> <li>边缘计算</li> <li>推理优化</li> <li>移动端部署</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>设备端图像识别</li> <li>AR 应用中的实时目标检测</li> <li>智能音箱上的语音指令</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/tensorflow-tensorflow/">tensorflow (TensorFlow)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/edge_ai-%E8%BE%B9%E7%BC%98-ai/">edge_ai (边缘 AI)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model_quantization-%E6%A8%A1%E5%9E%8B%E9%87%8F%E5%8C%96/">model_quantization (模型量化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/text_generation_inference/Sat, 18 Jul 2026 11:36:07 +0000https://terms-en.ai-term-hub.com/zh/terms/text_generation_inference/<h2 id="definition">Definition</h2> <p>文本生成推理（TGI）是一个专用的软件框架，旨在以低延迟和高吞吐量服务大型语言模型（LLM）。它针对文本生成的推理过程进行了深度优化，包括连续批处理和张量并行等技术，从而显著提升生产环境下的响应速度和并发处理能力。</p> <h3 id="summary">Summary</h3> <p>一个高性能的服务引擎，专门针对大规模高效部署大型语言模型以生成文本进行了优化。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>连续批处理</li> <li>张量并行</li> <li>低延迟服务</li> <li>LLM部署</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>生产级聊天机器人API</li> <li>实时内容生成服务</li> <li>高吞吐量文本分析平台</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/llm_serving-llm%E6%9C%8D%E5%8A%A1%E5%8C%96/">llm_serving (LLM服务化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/continuous_batching-%E8%BF%9E%E7%BB%AD%E6%89%B9%E5%A4%84%E7%90%86/">continuous_batching (连续批处理)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/huggingface_tgi-hugging-face-tgi%E5%B7%A5%E5%85%B7/">huggingface_tgi (Hugging Face TGI工具)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model_optimization-%E6%A8%A1%E5%9E%8B%E4%BC%98%E5%8C%96/">model_optimization (模型优化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/symbolic_regression/Sat, 18 Jul 2026 11:35:43 +0000https://terms-en.ai-term-hub.com/zh/terms/symbolic_regression/<h2 id="definition">Definition</h2> <p>符号回归是一种回归分析类型，旨在找到通常表示为树结构的数学表达式，以最优地拟合观测数据。与传统回归不同，它不预设函数形式。</p> <h3 id="summary">Summary</h3> <p>符号回归是一种搜索最佳拟合数据集的数学表达式的技術，旨在发现可解释的公式。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>遗传编程</li> <li>表达式树</li> <li>模型可解释性</li> <li>函数发现</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>物理定律发现</li> <li>化学过程建模</li> <li>金融趋势分析</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/genetic_algorithms-%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">genetic_algorithms (遗传算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/machine_learning-%E6%9C%BA%E5%99%A8%E5%AD%A6%E4%B9%A0/">machine_learning (机器学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/interpretable_ai-%E5%8F%AF%E8%A7%A3%E9%87%8Aai/">interpretable_ai (可解释AI)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/surrogate_model/Sat, 18 Jul 2026 11:35:31 +0000https://terms-en.ai-term-hub.com/zh/terms/surrogate_model/<h2 id="definition">Definition</h2> <p>在机器学习和优化中，代理模型作为难以直接评估的目标函数的代理。它通过原始模型的输入-输出对进行训练，以预测目标函数的行为。</p> <h3 id="summary">Summary</h3> <p>一种简化的数学模型，用于近似更复杂、计算成本高或不可访问的黑盒模型的行为。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>模型近似</li> <li>黑盒优化</li> <li>计算效率</li> <li>代理模型</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>超参数优化</li> <li>工程设计模拟加速</li> <li>复杂系统的敏感性分析</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.gaussian_process <span style="color:#f92672">import</span> GaussianProcessRegressor </span></span><span style="display:flex;"><span><span style="color:#f92672">import</span> numpy <span style="color:#66d9ef">as</span> np </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># Simple surrogate for a noisy function</span> </span></span><span style="display:flex;"><span>X <span style="color:#f92672">=</span> np<span style="color:#f92672">.</span>array([[<span style="color:#ae81ff">1</span>], [<span style="color:#ae81ff">2</span>], [<span style="color:#ae81ff">3</span>], [<span style="color:#ae81ff">4</span>]]) </span></span><span style="display:flex;"><span>y <span style="color:#f92672">=</span> np<span style="color:#f92672">.</span>array([<span style="color:#ae81ff">2.1</span>, <span style="color:#ae81ff">3.9</span>, <span style="color:#ae81ff">6.2</span>, <span style="color:#ae81ff">7.8</span>]) </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span>surrogate <span style="color:#f92672">=</span> GaussianProcessRegressor() </span></span><span style="display:flex;"><span>surrogate<span style="color:#f92672">.</span>fit(X, y) </span></span><span style="display:flex;"><span>prediction <span style="color:#f92672">=</span> surrogate<span style="color:#f92672">.</span>predict(np<span style="color:#f92672">.</span>array([[<span style="color:#ae81ff">2.5</span>]])) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/bayesian-optimization-%E8%B4%9D%E5%8F%B6%E6%96%AF%E4%BC%98%E5%8C%96/">Bayesian Optimization (贝叶斯优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gaussian-process-%E9%AB%98%E6%96%AF%E8%BF%87%E7%A8%8B/">Gaussian Process (高斯过程)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/black-box-function-%E9%BB%91%E7%9B%92%E5%87%BD%E6%95%B0/">Black-Box Function (黑盒函数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/emulator-%E6%A8%A1%E6%8B%9F%E5%99%A8/">Emulator (模拟器)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/structural_risk_minimization/Sat, 18 Jul 2026 11:35:19 +0000https://terms-en.ai-term-hub.com/zh/terms/structural_risk_minimization/<h2 id="definition">Definition</h2> <p>结构风险最小化（SRM）是一种通过控制模型复杂度来防止过拟合、从而最小化期望风险的方法。它在经验风险最小化的基础上增加了正则化项，以惩罚过于复杂的模型。</p> <h3 id="summary">Summary</h3> <p>统计学习中的一项原则，旨在通过平衡模型拟合度和复杂度来最小化泛化误差的上界。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>VC维</li> <li>正则化</li> <li>泛化误差</li> <li>模型复杂度惩罚</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>支持向量机（SVM）训练</li> <li>回归分析中多项式阶数的选择</li> <li>剪枝决策树以避免过拟合</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%BB%8F%E9%AA%8C%E9%A3%8E%E9%99%A9%E6%9C%80%E5%B0%8F%E5%8C%96-empirical-risk-minimization-%E6%9C%80%E5%B0%8F%E5%8C%96%E8%AE%AD%E7%BB%83%E8%AF%AF%E5%B7%AE/">经验风险最小化 (Empirical risk minimization，最小化训练误差)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%A5%A5%E5%8D%A1%E5%A7%86%E5%89%83%E5%88%80-occam-s-razor-%E5%A6%82%E6%97%A0%E5%BF%85%E8%A6%81%E5%8B%BF%E5%A2%9E%E5%AE%9E%E4%BD%93%E7%9A%84%E7%AE%80%E7%BA%A6%E5%8E%9F%E5%88%99/">奥卡姆剃刀 (Occam&rsquo;s razor，如无必要勿增实体的简约原则)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%AD%A3%E5%88%99%E5%8C%96-regularization-%E9%80%9A%E8%BF%87%E6%B7%BB%E5%8A%A0%E7%BA%A6%E6%9D%9F%E9%98%B2%E6%AD%A2%E8%BF%87%E6%8B%9F%E5%90%88/">正则化 (Regularization，通过添加约束防止过拟合)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%81%8F%E5%B7%AE-%E6%96%B9%E5%B7%AE%E6%9D%83%E8%A1%A1-bias-variance-tradeoff-%E5%B9%B3%E8%A1%A1%E6%A8%A1%E5%9E%8B%E8%AF%AF%E5%B7%AE%E6%9D%A5%E6%BA%90/">偏差-方差权衡 (Bias-variance tradeoff，平衡模型误差来源)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/structured_sparsity_regularization/Sat, 18 Jul 2026 11:35:19 +0000https://terms-en.ai-term-hub.com/zh/terms/structured_sparsity_regularization/<h2 id="definition">Definition</h2> <p>结构化稀疏正则化扩展了标准的L1正则化，鼓励在特定模式中产生零值，而不是独立地对待各个系数。它融入了关于特征之间潜在结构或分组的先验知识。</p> <h3 id="summary">Summary</h3> <p>一种正则化技术，基于对数据中特征分组或结构的先验知识，强制实施特定的稀疏模式。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>组Lasso</li> <li>特征分组</li> <li>稀疏恢复</li> <li>先验知识整合</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>具有通路结构的基因表达分析</li> <li>具有块稀疏信号图像处理</li> <li>具有共享特征集的多任务学习</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lasso%E5%9B%9E%E5%BD%92-lasso-regression-%E4%BD%BF%E7%94%A8l1%E6%AD%A3%E5%88%99%E5%8C%96%E7%9A%84%E7%BA%BF%E6%80%A7%E5%9B%9E%E5%BD%92/">Lasso回归 (Lasso regression，使用L1正则化的线性回归)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%BC%B9%E6%80%A7%E7%BD%91%E7%BB%9C-elastic-net-%E7%BB%93%E5%90%88l1%E5%92%8Cl2%E6%AD%A3%E5%88%99%E5%8C%96%E7%9A%84%E6%96%B9%E6%B3%95/">弹性网络 (Elastic net，结合L1和L2正则化的方法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%89%B9%E5%BE%81%E9%80%89%E6%8B%A9-feature-selection-%E4%BB%8E%E5%8E%9F%E5%A7%8B%E7%89%B9%E5%BE%81%E4%B8%AD%E6%8C%91%E9%80%89%E5%AD%90%E9%9B%86/">特征选择 (Feature selection，从原始特征中挑选子集)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8E%8B%E7%BC%A9%E6%84%9F%E7%9F%A5-compressed-sensing-%E5%88%A9%E7%94%A8%E7%A8%80%E7%96%8F%E6%80%A7%E4%BB%8E%E5%B0%91%E9%87%8F%E6%B5%8B%E9%87%8F%E4%B8%AD%E9%87%8D%E5%BB%BA%E4%BF%A1%E5%8F%B7/">压缩感知 (Compressed sensing，利用稀疏性从少量测量中重建信号)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/semantic_folding/Sat, 18 Jul 2026 11:33:07 +0000https://terms-en.ai-term-hub.com/zh/terms/semantic_folding/<h2 id="definition">Definition</h2> <p>语义折叠是指将复杂的高维向量嵌入压缩为更易管理的低维表示的过程，且不会造成语义意义的显著丢失。</p> <h3 id="summary">Summary</h3> <p>一种将高维语义表示映射到低维空间同时保留关系结构的技术。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>降维</li> <li>向量嵌入</li> <li>语义保持</li> <li>压缩</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>高效的大规模向量数据库索引</li> <li>减少自然语言处理模型的内存占用</li> <li>优化实时语义搜索系统</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pca-%E4%B8%BB%E6%88%90%E5%88%86%E5%88%86%E6%9E%90/">PCA (主成分分析)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/autoencoders-%E8%87%AA%E7%BC%96%E7%A0%81%E5%99%A8/">Autoencoders (自编码器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/embedding-compression-%E5%B5%8C%E5%85%A5%E5%8E%8B%E7%BC%A9/">Embedding compression (嵌入压缩)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/latent-space-%E6%BD%9C%E5%9C%A8%E7%A9%BA%E9%97%B4/">Latent space (潜在空间)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/reparameterization_trick/Sat, 18 Jul 2026 11:32:14 +0000https://terms-en.ai-term-hub.com/zh/terms/reparameterization_trick/<h2 id="definition">Definition</h2> <p>重参数化技巧是变分自编码器及其他概率模型中使用的一种基本方法。它通过将随机变量表示为确定性函数和独立噪声变量的组合，使得梯度能够流经随机节点，从而允许使用反向传播算法进行优化。</p> <h3 id="summary">Summary</h3> <p>一种将随机变量与可学习参数分离的技术，旨在变分推断中实现基于梯度的优化。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>变分推断</li> <li>梯度估计</li> <li>随机节点</li> <li>可微模拟</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练变分自编码器 (VAE)</li> <li>贝叶斯神经网络</li> <li>概率图模型</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch </span></span><span style="display:flex;"><span>epsilon <span style="color:#f92672">=</span> torch<span style="color:#f92672">.</span>randn(<span style="color:#ae81ff">100</span>, <span style="color:#ae81ff">10</span>) </span></span><span style="display:flex;"><span>mu <span style="color:#f92672">=</span> torch<span style="color:#f92672">.</span>zeros(<span style="color:#ae81ff">100</span>, <span style="color:#ae81ff">10</span>) </span></span><span style="display:flex;"><span>sigma <span style="color:#f92672">=</span> torch<span style="color:#f92672">.</span>ones(<span style="color:#ae81ff">100</span>, <span style="color:#ae81ff">10</span>) </span></span><span style="display:flex;"><span>z <span style="color:#f92672">=</span> mu <span style="color:#f92672">+</span> sigma <span style="color:#f92672">*</span> epsilon <span style="color:#75715e"># Reparameterized sampling</span> </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/elbo-%E8%AF%81%E6%8D%AE%E4%B8%8B%E7%95%8C/">ELBO (证据下界)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%BD%9C%E5%9C%A8%E5%8F%98%E9%87%8F/">潜在变量</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD/">反向传播</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%92%99%E7%89%B9%E5%8D%A1%E6%B4%9B%E4%BC%B0%E8%AE%A1/">蒙特卡洛估计</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/regularization/Sat, 18 Jul 2026 11:32:02 +0000https://terms-en.ai-term-hub.com/zh/terms/regularization/<h2 id="definition">Definition</h2> <p>正则化是机器学习中一个关键概念，旨在不显著增加训练误差的情况下减少泛化误差。它通过阻止模型学习过于复杂的模式来发挥作用。</p> <h3 id="summary">Summary</h3> <p>一组在训练过程中使用的技术，通过对损失函数添加惩罚项或限制模型复杂度来防止过拟合。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>过拟合</li> <li>偏差-方差权衡</li> <li>L1/L2 惩罚</li> <li>丢弃法 (Dropout)</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>深度神经网络训练</li> <li>线性回归模型</li> <li>防止小数据集上的记忆效应</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.linear_model <span style="color:#f92672">import</span> Ridge </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> Ridge(alpha<span style="color:#f92672">=</span><span style="color:#ae81ff">1.0</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/overfitting-%E8%BF%87%E6%8B%9F%E5%90%88/">Overfitting (过拟合)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/underfitting-%E6%AC%A0%E6%8B%9F%E5%90%88/">Underfitting (欠拟合)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/cross-validation-%E4%BA%A4%E5%8F%89%E9%AA%8C%E8%AF%81/">Cross-validation (交叉验证)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter-tuning-%E8%B6%85%E5%8F%82%E6%95%B0%E8%B0%83%E4%BC%98/">Hyperparameter tuning (超参数调优)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/random_feature/Sat, 18 Jul 2026 11:31:50 +0000https://terms-en.ai-term-hub.com/zh/terms/random_feature/<h2 id="definition">Definition</h2> <p>随机特征映射将输入转换到新空间，使线性模型能够近似非线性核函数。这种方法通常与 Nyström 方法或傅里叶特征相关联，允许在保持计算效率的同时处理复杂的非线性关系。</p> <h3 id="summary">Summary</h3> <p>一种利用随机投影将输入数据映射到高维空间的技术，以高效近似核方法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>核近似</li> <li>特征映射</li> <li>计算效率</li> <li>线性化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>大规模核回归</li> <li>神经切线核 (NTK) 近似</li> <li>可扩展的高斯过程</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> numpy <span style="color:#66d9ef">as</span> np </span></span><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.kernel_approximation <span style="color:#f92672">import</span> RBFSampler </span></span><span style="display:flex;"><span>X <span style="color:#f92672">=</span> np<span style="color:#f92672">.</span>random<span style="color:#f92672">.</span>rand(<span style="color:#ae81ff">100</span>, <span style="color:#ae81ff">5</span>) </span></span><span style="display:flex;"><span>transformer <span style="color:#f92672">=</span> RBFSampler(gamma<span style="color:#f92672">=</span><span style="color:#ae81ff">1</span>, n_components<span style="color:#f92672">=</span><span style="color:#ae81ff">50</span>, random_state<span style="color:#f92672">=</span><span style="color:#ae81ff">42</span>) </span></span><span style="display:flex;"><span>X_transformed <span style="color:#f92672">=</span> transformer<span style="color:#f92672">.</span>fit_transform(X) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A0%B8%E6%8A%80%E5%B7%A7-kernel-trick/">核技巧 (Kernel trick)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%82%85%E9%87%8C%E5%8F%B6%E7%89%B9%E5%BE%81-fourier-features/">傅里叶特征 (Fourier features)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/nystr%C3%B6m-%E6%96%B9%E6%B3%95-nystrom-method/">Nyström 方法 (Nystrom method)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%99%8D%E7%BB%B4-dimensionality-reduction/">降维 (Dimensionality reduction)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/quantized/Sat, 18 Jul 2026 11:31:12 +0000https://terms-en.ai-term-hub.com/zh/terms/quantized/<h2 id="definition">Definition</h2> <p>量化是一种模型优化技术，它降低了机器学习模型参数的数值精度，通常将 32 位浮点数转换为 8 位整数。这种方法可以在保持模型性能基本不变的情况下，显著减少内存占用并加速推理速度。</p> <h3 id="summary">Summary</h3> <p>量化后的指神经网络模型中权重和激活值以较低精度的数字表示，以减少模型大小和推理延迟。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>模型压缩</li> <li>低精度算术</li> <li>边缘部署</li> <li>推理优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>移动 AI 应用</li> <li>物联网设备集成</li> <li>实时视频处理</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%89%AA%E6%9E%9D-pruning/">剪枝 (Pruning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%9F%A5%E8%AF%86%E8%92%B8%E9%A6%8F-distillation/">知识蒸馏 (Distillation)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/tensorflow-lite/">TensorFlow Lite</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/prompt_tuning/Sat, 18 Jul 2026 11:30:52 +0000https://terms-en.ai-term-hub.com/zh/terms/prompt_tuning/<h2 id="definition">Definition</h2> <p>提示微调涉及在预训练语言模型的输入层添加可训练的软提示（连续向量），同时保持底层模型参数冻结。这种方法允许在几乎不增加计算成本的情况下，使模型适应特定任务或领域，特别适用于资源受限的场景。</p> <h3 id="summary">Summary</h3> <p>一种参数高效的微调方法，通过优化连续的输入嵌入（embeddings）而非更新整个模型权重来调整模型。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>软提示 (Soft Prompts)</li> <li>参数效率</li> <li>冻结权重</li> <li>少样本学习 (Few-shot Learning)</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>针对特定领域适配大语言模型</li> <li>低资源环境下的微调</li> <li>多任务学习优化</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/peft-parameter-efficient-fine-tuning-%E5%8F%82%E6%95%B0%E9%AB%98%E6%95%88%E5%BE%AE%E8%B0%83/">PEFT (Parameter-Efficient Fine-Tuning, 参数高效微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora-low-rank-adaptation-%E4%BD%8E%E7%A7%A9%E8%87%AA%E9%80%82%E5%BA%94/">LoRA (Low-Rank Adaptation, 低秩自适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%B8%8A%E4%B8%8B%E6%96%87%E5%AD%A6%E4%B9%A0-in-context-learning/">上下文学习 (In-context Learning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%BE%AE%E8%B0%83-fine-tuning/">微调 (Fine-tuning)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/proximal_gradient_methods_for_learning/Sat, 18 Jul 2026 11:30:52 +0000https://terms-en.ai-term-hub.com/zh/terms/proximal_gradient_methods_for_learning/<h2 id="definition">Definition</h2> <p>近端梯度方法是一种迭代优化技术，当损失函数包含可微分的平滑项和不可微分的正则项（如L1范数）时使用。该算法通过将平滑部分的梯度下降步骤与非平滑部分的近端算子（proximal operator）相结合，有效地处理非光滑优化问题。</p> <h3 id="summary">Summary</h3> <p>旨在最小化包含平滑和非平滑分量的复合目标函数的优化算法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>复合优化</li> <li>近端算子</li> <li>L1正则化</li> <li>非光滑凸性</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>稀疏特征选择</li> <li>Lasso回归</li> <li>结构化预测模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D-gradient-descent/">梯度下降 (Gradient Descent)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lasso-least-absolute-shrinkage-and-selection-operator/">Lasso (Least Absolute Shrinkage and Selection Operator)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%87%B8%E4%BC%98%E5%8C%96-convex-optimization/">凸优化 (Convex Optimization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%AD%A3%E5%88%99%E5%8C%96-regularization/">正则化 (Regularization)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/pagedattention/Sat, 18 Jul 2026 11:29:14 +0000https://terms-en.ai-term-hub.com/zh/terms/pagedattention/<h2 id="definition">Definition</h2> <p>PagedAttention 是由 vLLM 项目引入的一项技术，旨在提高大语言模型推理的效率。它解决了管理 KV 缓存时的碎片化和开销问题，通过将 KV 缓存视为非连续的内存块进行分配和管理，从而显著减少了内存浪费，提高了吞吐量并降低了显存占用。</p> <h3 id="summary">Summary</h3> <p>PagedAttention 是一种内存管理算法，它将虚拟内存分页概念应用于优化 Transformer 模型中键值（KV）缓存的存储和访问。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>KV 缓存管理</li> <li>内存碎片化</li> <li>推理优化</li> <li>虚拟内存分页</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>高吞吐量大语言模型服务</li> <li>减少 GPU 内存使用</li> <li>生产环境中的批处理优化</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/vllm-%E9%AB%98%E6%80%A7%E8%83%BDllm%E6%8E%A8%E7%90%86%E5%BC%95%E6%93%8E/">vLLM (高性能LLM推理引擎)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/key-value-cache-%E9%94%AE%E5%80%BC%E7%BC%93%E5%AD%98/">Key-Value Cache (键值缓存)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transformer-architecture-transformer%E6%9E%B6%E6%9E%84/">Transformer Architecture (Transformer架构)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gpu-memory-optimization-gpu%E5%86%85%E5%AD%98%E4%BC%98%E5%8C%96/">GPU Memory Optimization (GPU内存优化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/openvino/Sat, 18 Jul 2026 11:28:49 +0000https://terms-en.ai-term-hub.com/zh/terms/openvino/<h2 id="definition">Definition</h2> <p>由英特尔开发，OpenVINO（开放视觉推理和神经网络优化）允许开发人员将经过训练的深度学习模型高效地部署到英特尔硬件上。它包含一个推理引擎和优化模型的工具链。</p> <h3 id="summary">Summary</h3> <p>OpenVINO 是英特尔推出的开源工具包，用于在各种硬件平台上高效地优化和部署深度学习模型。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>模型优化</li> <li>硬件加速</li> <li>推理引擎</li> <li>边缘计算</li> <li>跨框架支持</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在英特尔 CPU/GPU 上部署目标检测</li> <li>优化视频分析流水线</li> <li>在边缘设备上运行实时推理</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/intel_hardware-%E8%8B%B1%E7%89%B9%E5%B0%94%E7%A1%AC%E4%BB%B6/">intel_hardware (英特尔硬件)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model_deployment-%E6%A8%A1%E5%9E%8B%E9%83%A8%E7%BD%B2/">model_deployment (模型部署)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/inference_optimization-%E6%8E%A8%E7%90%86%E4%BC%98%E5%8C%96/">inference_optimization (推理优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/edge_ai-%E8%BE%B9%E7%BC%98%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD/">edge_ai (边缘人工智能)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/mxfp4/Sat, 18 Jul 2026 11:28:00 +0000https://terms-en.ai-term-hub.com/zh/terms/mxfp4/<h2 id="definition">Definition</h2> <p>MXFP4（混合扩展浮点 4 位）是一种专门的数据类型格式，旨在优化 AI 工作负载的性能并减少内存带宽使用。通过允许混合精度操作，它在保持精度的同时显著降低了计算资源消耗。</p> <h3 id="summary">Summary</h3> <p>MXFP4 是一种混合精度浮点格式，专为在 AI 硬件加速器中高效进行矩阵乘法而优化。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>量化</li> <li>混合精度</li> <li>硬件加速</li> <li>矩阵乘法</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>优化边缘设备上的大语言模型推理</li> <li>减少大规模训练中的内存占用</li> <li>加速基于 Transformer 模型的部署</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/int8-8%E4%BD%8D%E6%95%B4%E6%95%B0%E6%A0%BC%E5%BC%8F/">INT8 (8位整数格式)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fp16-%E5%8D%8A%E7%B2%BE%E5%BA%A6%E6%B5%AE%E7%82%B9%E6%A0%BC%E5%BC%8F/">FP16 (半精度浮点格式)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%87%8F%E5%8C%96-quantization/">量化 (Quantization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gpu-%E4%BC%98%E5%8C%96-gpu-optimization/">GPU 优化 (GPU Optimization)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/multiplicative_weight_update_method/Sat, 18 Jul 2026 11:27:47 +0000https://terms-en.ai-term-hub.com/zh/terms/multiplicative_weight_update_method/<h2 id="definition">Definition</h2> <p>乘法权重更新法是一种基础的在线学习算法，用于在不确定环境中做出决策。它为不同的策略或专家维护一组权重，并根据其表现动态调整这些权重，从而优化长期决策效果。</p> <h3 id="summary">Summary</h3> <p>一种迭代算法，根据性能反馈以乘法方式更新权重，以最小化遗憾值。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>在线学习</li> <li>加权多数算法</li> <li>遗憾最小化</li> <li>指数加权</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>投资组合优化</li> <li>专家建议聚合</li> <li>对抗性老虎机问题</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient_descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">gradient_descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/online_learning-%E5%9C%A8%E7%BA%BF%E5%AD%A6%E4%B9%A0/">online_learning (在线学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/regret_bound-%E9%81%97%E6%86%BE%E7%95%8C/">regret_bound (遗憾界)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/boosting-%E6%8F%90%E5%8D%87%E7%AE%97%E6%B3%95/">boosting (提升算法)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/multi_task_learning/Sat, 18 Jul 2026 11:26:49 +0000https://terms-en.ai-term-hub.com/zh/terms/multi_task_learning/<h2 id="definition">Definition</h2> <p>该技术利用相关任务之间共享的归纳偏置来提高学习效率 and 性能。通过训练单个模型同时执行多项任务，模型能够学习到更具通用性的特征表示，从而提升整体表现。</p> <h3 id="summary">Summary</h3> <p>多任务学习是一种机器学习范式，其中模型同时训练多个相关任务以提高泛化能力。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>共享表示</li> <li>归纳偏置</li> <li>任务干扰</li> <li>泛化能力</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自然语言处理（如命名实体识别和词性标注）</li> <li>计算机视觉（如目标检测和分割）</li> <li>语音识别与合成</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transfer-learning-%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0/">Transfer Learning (迁移学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/joint-training-%E8%81%94%E5%90%88%E8%AE%AD%E7%BB%83/">Joint Training (联合训练)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/neural-architecture-search-%E7%A5%9E%E7%BB%8F%E6%9E%B6%E6%9E%84%E6%90%9C%E7%B4%A2/">Neural Architecture Search (神经架构搜索)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/regularization-%E6%AD%A3%E5%88%99%E5%8C%96/">Regularization (正则化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/mobilenet/Sat, 18 Jul 2026 11:26:24 +0000https://terms-en.ai-term-hub.com/zh/terms/mobilenet/<h2 id="definition">Definition</h2> <p>与标准卷积相比，MobileNet 利用深度可分离卷积大幅降低了计算成本和模型体积。这种架构使得在资源受限设备上高效提取特征成为可能。</p> <h3 id="summary">Summary</h3> <p>MobileNet 是一系列专为移动端和嵌入式视觉应用设计的轻量级深度神经网络。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>深度可分离卷积</li> <li>模型效率</li> <li>边缘计算</li> <li>迁移学习</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>智能手机上的实时目标检测</li> <li>物联网设备上的图像分类</li> <li>移动应用中的面部识别</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> tensorflow.keras.applications <span style="color:#f92672">import</span> MobileNetV2 </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> MobileNetV2(weights<span style="color:#f92672">=</span><span style="color:#e6db74">&#39;imagenet&#39;</span>, input_shape<span style="color:#f92672">=</span>(<span style="color:#ae81ff">224</span>, <span style="color:#ae81ff">224</span>, <span style="color:#ae81ff">3</span>)) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/shufflenet-%E6%B4%97%E7%89%8C%E7%BD%91%E7%BB%9C/">ShuffleNet (洗牌网络)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/squeezenet-%E6%8C%A4%E5%8E%8B%E7%BD%91%E7%BB%9C/">SqueezeNet (挤压网络)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/efficientnet-%E9%AB%98%E6%95%88%E7%BD%91%E7%BB%9C/">EfficientNet (高效网络)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/convolutional-neural-network-%E5%8D%B7%E7%A7%AF%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C/">Convolutional Neural Network (卷积神经网络)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/model_compression/Sat, 18 Jul 2026 11:26:24 +0000https://terms-en.ai-term-hub.com/zh/terms/model_compression/<h2 id="definition">Definition</h2> <p>该类别包括剪枝、量化和知识蒸馏等方法，旨在缩小模型规模的同时保持性能。这对于部署复杂的人工智能模型至关重要，尤其是在资源受限的环境中。</p> <h3 id="summary">Summary</h3> <p>模型压缩是指减少机器学习模型体积和计算需求的技术。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>量化</li> <li>剪枝</li> <li>知识蒸馏</li> <li>推理速度</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在移动设备上部署模型</li> <li>降低云端推理成本</li> <li>加速实时视频处理</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.quantization <span style="color:#66d9ef">as</span> quant </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> quant<span style="color:#f92672">.</span>quantize_dynamic(model, {torch<span style="color:#f92672">.</span>nn<span style="color:#f92672">.</span>Linear}, dtype<span style="color:#f92672">=</span>torch<span style="color:#f92672">.</span>qint8) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">Quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pruning-%E5%89%AA%E6%9E%9D/">Pruning (剪枝)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/distillation-%E8%92%B8%E9%A6%8F/">Distillation (蒸馏)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/edge-ai-%E8%BE%B9%E7%BC%98%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD/">Edge AI (边缘人工智能)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/mixed_precision_training/Sat, 18 Jul 2026 11:26:13 +0000https://terms-en.ai-term-hub.com/zh/terms/mixed_precision_training/<h2 id="definition">Definition</h2> <p>混合精度训练（MPT）在神经网络训练过程中结合使用半精度（FP16）和全精度（FP32）数据类型。通过使用FP16处理大多数操作，MPT减少了内存占用并提</p> <h3 id="summary">Summary</h3> <p>一种使用16位和32位浮点数进行训练的技術，旨在加速计算并减少内存使用。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>FP16</li> <li>FP32</li> <li>张量核心</li> <li>数值稳定性</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>大模型训练</li> <li>GPU加速</li> <li>内存受限环境</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch </span></span><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.cuda.amp <span style="color:#66d9ef">as</span> amp </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># Example snippet showing automatic mixed precision context</span> </span></span><span style="display:flex;"><span><span style="color:#66d9ef">with</span> amp<span style="color:#f92672">.</span>autocast(): </span></span><span style="display:flex;"><span> output <span style="color:#f92672">=</span> model(input) </span></span><span style="display:flex;"><span> loss <span style="color:#f92672">=</span> criterion(output, target) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-scaling-%E6%A2%AF%E5%BA%A6%E7%BC%A9%E6%94%BE/">gradient scaling (梯度缩放)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/amp-%E8%87%AA%E5%8A%A8%E6%B7%B7%E5%90%88%E7%B2%BE%E5%BA%A6/">AMP (自动混合精度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/half-precision-%E5%8D%8A%E7%B2%BE%E5%BA%A6/">half-precision (半精度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimization-%E4%BC%98%E5%8C%96/">optimization (优化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/meta_learning/Sat, 18 Jul 2026 11:26:00 +0000https://terms-en.ai-term-hub.com/zh/terms/meta_learning/<h2 id="definition">Definition</h2> <p>元学习专注于设计能够从先前任务中学习以改善在新颖、未见任务上表现的算法。它不是针对每个问题从头开始训练模型，而是优化模型的学习机制，使其具备快速适应新环境的能力。</p> <h3 id="summary">Summary</h3> <p>元学习，或称“学会学习”，是一种机器学习方法，使模型能够利用以往经验，以最少数据快速适应新任务。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>少样本学习</li> <li>迁移学习</li> <li>模型无关元学习</li> <li>任务分布</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>快速适应新客户偏好</li> <li>多变环境下的机器人控制</li> <li>有限患者数据的医疗诊断</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transfer-learning-%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0/">Transfer Learning (迁移学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/few-shot-learning-%E5%B0%91%E6%A0%B7%E6%9C%AC%E5%AD%A6%E4%B9%A0/">Few-Shot Learning (少样本学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reinforcement-learning-%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">Reinforcement Learning (强化学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/matrix_regularization/Sat, 18 Jul 2026 11:25:36 +0000https://terms-en.ai-term-hub.com/zh/terms/matrix_regularization/<h2 id="definition">Definition</h2> <p>矩阵正则化将标量正则化的概念扩展到矩阵，常用于多任务学习或推荐系统。它对权重矩阵的范数施加约束，例如弗罗贝尼乌斯范数或核范数，以鼓励低秩近似或稀疏解，从而提高模型的泛化能力。</p> <h3 id="summary">Summary</h3> <p>一种将惩罚项应用于矩阵值参数的技术，以防止过拟合并强制执行稀疏性等结构属性。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>弗罗贝尼乌斯范数</li> <li>核范数</li> <li>防止过拟合</li> <li>低秩近似</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>协同过滤</li> <li>多任务学习</li> <li>降维</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/ridge-regression-%E5%B2%AD%E5%9B%9E%E5%BD%92/">Ridge Regression (岭回归)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lasso-%E5%A5%97%E7%B4%A2%E5%9B%9E%E5%BD%92/">Lasso (套索回归)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/nuclear-norm-minimization-%E6%A0%B8%E8%8C%83%E6%95%B0%E6%9C%80%E5%B0%8F%E5%8C%96/">Nuclear Norm Minimization (核范数最小化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sparse-learning-%E7%A8%80%E7%96%8F%E5%AD%A6%E4%B9%A0/">Sparse Learning (稀疏学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/lottery_ticket_hypothesis/Sat, 18 Jul 2026 11:24:46 +0000https://terms-en.ai-term-hub.com/zh/terms/lottery_ticket_hypothesis/<h2 id="definition">Definition</h2> <p>彩票假设指出，在一个大型随机初始化的神经网络中，存在一个稀疏的子网络（即“中奖彩票”），其初始化状态非常适合训练。通过剪枝掉不重要的权重，仅保留并训练这个子网络，可以在不损失性能的前提下大幅减少模型规模。</p> <h3 id="summary">Summary</h3> <p>该理论认为密集神经网络中包含较小的子网络，若从初始状态单独训练，其准确率可与原始网络相匹配。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>权重剪枝</li> <li>稀疏网络</li> <li>模型压缩</li> <li>初始化敏感性</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在边缘设备上部署轻量级模型</li> <li>降低训练期间的计算成本</li> <li>加速推理速度</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/network-pruning-%E7%BD%91%E7%BB%9C%E5%89%AA%E6%9E%9D/">Network Pruning (网络剪枝)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-distillation-%E6%A8%A1%E5%9E%8B%E8%92%B8%E9%A6%8F/">Model Distillation (模型蒸馏)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sparse-training-%E7%A8%80%E7%96%8F%E8%AE%AD%E7%BB%83/">Sparse Training (稀疏训练)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/efficient-ai-%E9%AB%98%E6%95%88%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD/">Efficient AI (高效人工智能)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/local_case_control_sampling/Sat, 18 Jul 2026 11:24:46 +0000https://terms-en.ai-term-hub.com/zh/terms/local_case_control_sampling/<h2 id="definition">Definition</h2> <p>局部病例对照采样是一种主要用于训练对比学习模型或推荐系统的策略。它不是随机选择负样本，而是识别出与正例在嵌入空间中距离较近、难以区分的“困难负例”，从而提升模型的判别能力。</p> <h3 id="summary">Summary</h3> <p>一种负采样技术，从嵌入空间中正例的邻近区域选择困难负例。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>困难负例</li> <li>对比学习</li> <li>嵌入空间</li> <li>负采样</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练图像检索系统</li> <li>提高推荐引擎的准确性</li> <li>针对特定任务微调大型语言模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/triplet-loss-%E4%B8%89%E5%85%83%E7%BB%84%E6%8D%9F%E5%A4%B1/">Triplet Loss (三元组损失)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/infonce-%E4%BF%A1%E6%81%AF%E5%99%AA%E5%A3%B0%E5%AF%B9%E6%AF%94%E4%BC%B0%E8%AE%A1%E6%8D%9F%E5%A4%B1/">InfoNCE (信息噪声对比估计损失)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hard-negative-mining-%E5%9B%B0%E9%9A%BE%E8%B4%9F%E4%BE%8B%E6%8C%96%E6%8E%98/">Hard Negative Mining (困难负例挖掘)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/contrastive-divergence-%E5%AF%B9%E6%AF%94%E6%95%A3%E5%BA%A6/">Contrastive Divergence (对比散度)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/learning_automaton/Sat, 18 Jul 2026 11:23:53 +0000https://terms-en.ai-term-hub.com/zh/terms/learning_automaton/<h2 id="definition">Definition</h2> <p>这一概念源于强化学习，涉及智能体与未知环境的交互。自动机从有限集合中选择动作，并接收表示成功或失败的惩罚或奖励信号，从而调整其策略。</p> <h3 id="summary">Summary</h3> <p>学习自动机是一种简单的随机决策单元，它根据环境反馈迭代更新动作概率，以最大化奖励。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>动作概率向量</li> <li>奖励/惩罚信号</li> <li>随机优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>资源分配问题</li> <li>网络路由优化</li> <li>简单控制系统</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reinforcement-learning-%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">Reinforcement learning (强化学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/markov-decision-process-%E9%A9%AC%E5%B0%94%E5%8F%AF%E5%A4%AB%E5%86%B3%E7%AD%96%E8%BF%87%E7%A8%8B/">Markov decision process (马尔可夫决策过程)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/q-learning-q%E5%AD%A6%E4%B9%A0/">Q-learning (Q学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/layer_normalization/Sat, 18 Jul 2026 11:23:41 +0000https://terms-en.ai-term-hub.com/zh/terms/layer_normalization/<h2 id="definition">Definition</h2> <p>层归一化通过减少内部协变量偏移来稳定训练过程，尤其在循环神经网络和Transformer架构中非常有效。与依赖于批次统计信息的批归一化不同，层归一化不依赖于批次大小。</p> <h3 id="summary">Summary</h3> <p>一种技术，针对每个单独样本，对神经网络层的激活值在特征维度上进行归一化处理。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>归一化</li> <li>内部协变量偏移</li> <li>Transformer模型</li> <li>循环神经网络</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练BERT等Transformer模型</li> <li>稳定RNN/LSTM的训练</li> <li>小批量大小的深度学习</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.nn <span style="color:#66d9ef">as</span> nn </span></span><span style="display:flex;"><span>norm_layer <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>LayerNorm(normalized_shape<span style="color:#f92672">=</span>[<span style="color:#ae81ff">768</span>]) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/batch_normalization-%E6%89%B9%E5%BD%92%E4%B8%80%E5%8C%96/">batch_normalization (批归一化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transformer-transformer/">transformer (Transformer)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/normalization-%E5%BD%92%E4%B8%80%E5%8C%96/">normalization (归一化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/deep_learning-%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/">deep_learning (深度学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/knowledge_compilation/Sat, 18 Jul 2026 11:23:05 +0000https://terms-en.ai-term-hub.com/zh/terms/knowledge_compilation/<h2 id="definition">Definition</h2> <p>知识编译是指人工智能中的一类技术，它将知识库或逻辑理论转换为不同的表示形式，从而促进更快的操作，如可满足性检查。</p> <h3 id="summary">Summary</h3> <p>知识编译是将知识表示转换为更高效形式的过程，以便更快地回答问题或进行推理。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>逻辑表示</li> <li>查询优化</li> <li>可满足性</li> <li>预处理</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自动规划系统</li> <li>约束满足问题</li> <li>实时诊断系统</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/logic-programming-%E9%80%BB%E8%BE%91%E7%BC%96%E7%A8%8B/">Logic Programming (逻辑编程)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sat-solvers-%E5%B8%83%E5%B0%94%E5%8F%AF%E6%BB%A1%E8%B6%B3%E6%80%A7%E9%97%AE%E9%A2%98%E6%B1%82%E8%A7%A3%E5%99%A8/">SAT Solvers (布尔可满足性问题求解器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/ontology-%E6%9C%AC%E4%BD%93%E8%AE%BA/">Ontology (本体论)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reasoning-%E6%8E%A8%E7%90%86/">Reasoning (推理)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/knowledge_distillation/Sat, 18 Jul 2026 11:23:05 +0000https://terms-en.ai-term-hub.com/zh/terms/knowledge_distillation/<h2 id="definition">Definition</h2> <p>知识蒸馏是一种机器学习方法，用于将庞大复杂的神经网络（教师模型）压缩为更小、更高效的网络（学生模型）。学生模型经过训练以模仿教师模型的行为。</p> <h3 id="summary">Summary</h3> <p>知识蒸馏是一种模型压缩技术，其中较小的学生模型通过学习模仿较大的教师模型的行为来工作。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>师生模型</li> <li>模型压缩</li> <li>软标签</li> <li>效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在边缘设备上部署模型</li> <li>降低推理延迟</li> <li>降低云计算成本</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch </span></span><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.nn <span style="color:#66d9ef">as</span> nn </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#66d9ef">def</span> <span style="color:#a6e22e">distillation_loss</span>(student_logits, teacher_logits, temperature<span style="color:#f92672">=</span><span style="color:#ae81ff">2.0</span>): </span></span><span style="display:flex;"><span> T <span style="color:#f92672">=</span> temperature </span></span><span style="display:flex;"><span> student_probs <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>functional<span style="color:#f92672">.</span>softmax(student_logits <span style="color:#f92672">/</span> T, dim<span style="color:#f92672">=</span><span style="color:#ae81ff">1</span>) </span></span><span style="display:flex;"><span> teacher_probs <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>functional<span style="color:#f92672">.</span>softmax(teacher_logits <span style="color:#f92672">/</span> T, dim<span style="color:#f92672">=</span><span style="color:#ae81ff">1</span>) </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">return</span> nn<span style="color:#f92672">.</span>functional<span style="color:#f92672">.</span>kl_div( </span></span><span style="display:flex;"><span> nn<span style="color:#f92672">.</span>functional<span style="color:#f92672">.</span>log_softmax(student_logits <span style="color:#f92672">/</span> T, dim<span style="color:#f92672">=</span><span style="color:#ae81ff">1</span>), </span></span><span style="display:flex;"><span> teacher_probs, </span></span><span style="display:flex;"><span> reduction<span style="color:#f92672">=</span><span style="color:#e6db74">&#39;batchmean&#39;</span> </span></span><span style="display:flex;"><span> ) <span style="color:#f92672">*</span> (T <span style="color:#f92672">*</span> T) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-compression-%E6%A8%A1%E5%9E%8B%E5%8E%8B%E7%BC%A9/">Model Compression (模型压缩)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pruning-%E5%89%AA%E6%9E%9D/">Pruning (剪枝)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">Quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/neural-networks-%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C/">Neural Networks (神经网络)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/instance_selection/Sat, 18 Jul 2026 11:22:25 +0000https://terms-en.ai-term-hub.com/zh/terms/instance_selection/<h2 id="definition">Definition</h2> <p>实例选择旨在通过移除冗余或噪声数据点来提高计算效率和模型性能。与特征选择不同，它作用于数据集的行。其目标。</p> <h3 id="summary">Summary</h3> <p>一种预处理技术，通过选择代表性实例的子集来减小数据集的大小。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>数据缩减</li> <li>去噪</li> <li>代表性子集</li> <li>计算效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>大规模数据集预处理</li> <li>加速最近邻搜索</li> <li>清理不平衡数据集</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%95%B0%E6%8D%AE%E9%87%87%E6%A0%B7-data-sampling/">数据采样 (Data sampling)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%AC%A0%E9%87%87%E6%A0%B7-under-sampling/">欠采样 (Under-sampling)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8E%8B%E7%BC%A9%E6%9C%80%E8%BF%91%E9%82%BB-condensed-nearest-neighbor/">压缩最近邻 (Condensed Nearest Neighbor)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/incremental_heuristic_search/Sat, 18 Jul 2026 11:22:14 +0000https://terms-en.ai-term-hub.com/zh/terms/incremental_heuristic_search/<h2 id="definition">Definition</h2> <p>增量启发式搜索指的是通过启发式算法引导，逐步细化候选解的算法，这些启发式算法估计到达目标所需的成本。与穷举搜索不同，这些方法&hellip;</p> <h3 id="summary">Summary</h3> <p>一种搜索策略，利用启发式指导逐步改进解的估计，而不是探索整个状态空间。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>启发式方法</li> <li>局部搜索</li> <li>优化</li> <li>状态空间</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>物流中的路线规划</li> <li>调度问题</li> <li>约束满足任务</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hill-climbing-%E7%88%AC%E5%B1%B1%E7%AE%97%E6%B3%95/">Hill Climbing (爬山算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/simulated-annealing-%E6%A8%A1%E6%8B%9F%E9%80%80%E7%81%AB/">Simulated Annealing (模拟退火)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/a-search-a-%E6%90%9C%E7%B4%A2/">A* Search (A* 搜索)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/metaheuristics-%E5%85%83%E5%90%AF%E5%8F%91%E5%BC%8F%E7%AE%97%E6%B3%95/">Metaheuristics (元启发式算法)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/imatrix/Sat, 18 Jul 2026 11:21:59 +0000https://terms-en.ai-term-hub.com/zh/terms/imatrix/<h2 id="definition">Definition</h2> <p>Imatrix（Importance Matrix，重要性矩阵）是一种主要与基于GGML的大语言模型（LLM）训练和量化相关的技术。它计算损失函数关于模型参数的二阶导数（即海森矩阵的近似值），从而评估每个参数对整体模型性能的重要性。这种方法有助于更精确地进行模型量化，减少精度损失，并优化微调过程中的计算效率。</p> <h3 id="summary">Summary</h3> <p>一种用于大语言模型训练的具体算法，用于计算重要性矩阵以实现高效的参数优化。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>海森矩阵</li> <li>参数重要性</li> <li>模型量化</li> <li>微调优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>高效的大语言模型微调</li> <li>面向边缘设备的模型量化</li> <li>降低训练过程中的计算开销</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/ggml-ggml%E5%BA%93/">GGML (GGML库)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora-%E4%BD%8E%E7%A7%A9%E9%80%82%E5%BA%94/">LoRA (低秩适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">Quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/second-order-optimization-%E4%BA%8C%E9%98%B6%E4%BC%98%E5%8C%96/">Second-Order Optimization (二阶优化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/hyperparameter_tuning/Sat, 18 Jul 2026 11:21:34 +0000https://terms-en.ai-term-hub.com/zh/terms/hyperparameter_tuning/<h2 id="definition">Definition</h2> <p>超参数调优涉及评估不同的超参数集，以找到能产生最高模型准确率或最低错误率的配置。常见的策略包括网格搜索、随机搜索以及基于贝叶斯优化的方法，旨在通过迭代改进来寻找最优解。</p> <h3 id="summary">Summary</h3> <p>系统地搜索最佳超参数组合以优化模型性能的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>网格搜索</li> <li>随机搜索</li> <li>交叉验证</li> <li>性能指标</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>优化深度学习模型</li> <li>提高分类器准确率</li> <li>减少模型偏差和方差</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter_optimization-%E8%B6%85%E5%8F%82%E6%95%B0%E4%BC%98%E5%8C%96/">hyperparameter_optimization (超参数优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/grid_search-%E7%BD%91%E6%A0%BC%E6%90%9C%E7%B4%A2/">grid_search (网格搜索)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/random_search-%E9%9A%8F%E6%9C%BA%E6%90%9C%E7%B4%A2/">random_search (随机搜索)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/cross_validation-%E4%BA%A4%E5%8F%89%E9%AA%8C%E8%AF%81/">cross_validation (交叉验证)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/hyperparameter_optimization/Sat, 18 Jul 2026 11:21:34 +0000https://terms-en.ai-term-hub.com/zh/terms/hyperparameter_optimization/<h2 id="definition">Definition</h2> <p>超参数优化（HPO）指的是自动化选择超参数的更广泛领域。虽然“调优”是一个通用术语，但“优化”通常暗示使用复杂的算法（如贝叶斯优化）来高效地探索超参数空间，从而在有限的计算资源下找到最优配置。</p> <h3 id="summary">Summary</h3> <p>一种自动寻找最佳超参数配置的方法，通常使用概率模型来指导搜索过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>贝叶斯优化</li> <li>代理模型</li> <li>自动化机器学习</li> <li>搜索空间</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>AutoML 管道</li> <li>大规模模型训练</li> <li>资源受限环境下的优化</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter_tuning-%E8%B6%85%E5%8F%82%E6%95%B0%E8%B0%83%E4%BC%98/">hyperparameter_tuning (超参数调优)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/bayesian_optimization-%E8%B4%9D%E5%8F%B6%E6%96%AF%E4%BC%98%E5%8C%96/">bayesian_optimization (贝叶斯优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/automl-%E8%87%AA%E5%8A%A8%E5%8C%96%E6%9C%BA%E5%99%A8%E5%AD%A6%E4%B9%A0/">automl (自动化机器学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/neural_architecture_search-%E7%A5%9E%E7%BB%8F%E6%9E%B6%E6%9E%84%E6%90%9C%E7%B4%A2/">neural_architecture_search (神经架构搜索)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/hierarchical_risk_parity/Sat, 18 Jul 2026 11:20:58 +0000https://terms-en.ai-term-hub.com/zh/terms/hierarchical_risk_parity/<h2 id="definition">Definition</h2> <p>分层风险平价（HRP）是一种投资组合构建方法，通过纳入相关性结构来解决传统均值-方差优化的局限性。它利用层次聚类算法对资产进行分组，并基于聚类树状图在层级结构中分配风险，从而避免了对协方差矩阵求逆的数值不稳定问题。</p> <h3 id="summary">Summary</h3> <p>一种利用聚类处理资产相关性并在分层集群间分配风险的投资组合优化技术。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>聚类</li> <li>树状图</li> <li>风险分配</li> <li>协方差矩阵</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>投资组合管理</li> <li>资产配置</li> <li>风险缓解</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/mean-variance-optimization-%E5%9D%87%E5%80%BC-%E6%96%B9%E5%B7%AE%E4%BC%98%E5%8C%96/">Mean-Variance Optimization (均值-方差优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/black-litterman-model-%E5%B8%83%E8%8E%B1%E5%85%8B-%E5%88%A9%E7%89%B9%E6%9B%BC%E6%A8%A1%E5%9E%8B/">Black-Litterman Model (布莱克-利特曼模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/cluster-analysis-%E8%81%9A%E7%B1%BB%E5%88%86%E6%9E%90/">Cluster Analysis (聚类分析)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/gradient_accumulation/Sat, 18 Jul 2026 11:19:45 +0000https://terms-en.ai-term-hub.com/zh/terms/gradient_accumulation/<h2 id="definition">Definition</h2> <p>这种优化策略允许深度学习模型使用超出 GPU 显存容量的有效批量大小进行训练。通过从多个小批量中累积梯度并执行权重更新，可以实现这一目标。</p> <h3 id="summary">Summary</h3> <p>梯度累积是一种通过在前向/反向传播过程中累加多个步骤的梯度，从而模拟更大批量大小的技术，之后才更新权重。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>批量大小模拟</li> <li>内存优化</li> <li>随机梯度下降</li> <li>权重更新</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>微调大模型</li> <li>在有限显存下训练</li> <li>稳定损失收敛</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%89%B9%E5%BD%92%E4%B8%80%E5%8C%96/">批归一化</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%AD%A6%E4%B9%A0%E7%8E%87%E7%BC%A9%E6%94%BE/">学习率缩放</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%BC%98%E5%8C%96%E5%99%A8/">优化器</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD/">反向传播</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/glm_moe_dsa/Sat, 18 Jul 2026 11:19:07 +0000https://terms-en.ai-term-hub.com/zh/terms/glm_moe_dsa/<h2 id="definition">Definition</h2> <p>不存在单一的标准术语“GLM MoE DSA”。然而，它很可能结合了 GLM（一种特定的大型语言模型架构）、MoE（混合专家，一种通过仅激活部分专家来高效扩展模型规模的技术）以及 DSA（可能指动态稀疏注意力或其他类似机制）。</p> <h3 id="summary">Summary</h3> <p>这似乎是不同概念的混合：GLM（通用语言模型）、MoE（混合专家）以及可能的 DSA（动态稀疏注意力等）。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>混合专家</li> <li>稀疏注意力</li> <li>模型扩展</li> <li>高效推理</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>扩展大型语言模型</li> <li>降低推理成本</li> <li>高性能自然语言处理</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/moe-%E6%B7%B7%E5%90%88%E4%B8%93%E5%AE%B6/">MoE (混合专家)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sparse-transformers-%E7%A8%80%E7%96%8Ftransformer/">Sparse Transformers (稀疏Transformer)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/glm-%E9%80%9A%E7%94%A8%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B/">GLM (通用语言模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/switch-transformer-%E5%BC%80%E5%85%B3transformer/">Switch Transformer (开关Transformer)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/gguf/Sat, 18 Jul 2026 11:17:53 +0000https://terms-en.ai-term-hub.com/zh/terms/gguf/<h2 id="definition">Definition</h2> <p>GGUF（GPT-Generated Unified Format，GPT生成统一格式）是一种专为在消费级硬件上运行大型语言模型而设计的二进制文件格式。它支持各种量化技术，允许模型在资源受限的设备上更高效地运行。</p> <h3 id="summary">Summary</h3> <p>由gger.ai开发的一种文件格式，用于在本地硬件上高效存储和加载量化大型语言模型。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>量化</li> <li>模型序列化</li> <li>本地推理</li> <li>内存优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在笔记本电脑或台式机上本地运行LLM</li> <li>在资源受限的边缘设备上部署模型</li> <li>在开源社区中分享优化的模型权重</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/llama-cpp-llama-cpp%E5%BA%93/">LLAMA.cpp (LLAMA.cpp库)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">Quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/onnx-%E5%BC%80%E6%94%BE%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C%E4%BA%A4%E6%8D%A2%E6%A0%BC%E5%BC%8F/">ONNX (开放神经网络交换格式)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-weights-%E6%A8%A1%E5%9E%8B%E6%9D%83%E9%87%8D/">Model Weights (模型权重)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/fp8/Sat, 18 Jul 2026 11:17:40 +0000https://terms-en.ai-term-hub.com/zh/terms/fp8/<h2 id="definition">Definition</h2> <p>8位浮点数（FP8）是一种数值数据类型，在现代AI硬件上进行了专门优化，以在计算效率和准确性之间取得平衡。它减少了内存带宽需求并提高了吞吐量，特别适用于大规模语言模型训练和边缘设备上的实时推理。</p> <h3 id="summary">Summary</h3> <p>FP8是一种专为加速深度学习训练和推理而设计的8位浮点格式，同时保持可接受的精度水平。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>量化</li> <li>精度</li> <li>内存带宽</li> <li>硬件加速</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>大语言模型训练</li> <li>边缘AI推理</li> <li>实时视频处理</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fp16-%E5%8D%8A%E7%B2%BE%E5%BA%A6%E6%B5%AE%E7%82%B9/">fp16 (半精度浮点)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fp32-%E5%8D%95%E7%B2%BE%E5%BA%A6%E6%B5%AE%E7%82%B9/">fp32 (单精度浮点)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/mixed_precision-%E6%B7%B7%E5%90%88%E7%B2%BE%E5%BA%A6/">mixed_precision (混合精度)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/fitness_approximation/Sat, 18 Jul 2026 11:17:25 +0000https://terms-en.ai-term-hub.com/zh/terms/fitness_approximation/<h2 id="definition">Definition</h2> <p>当评估真实适应度函数在计算上昂贵或耗时较长时，适应度近似被用于进化计算中。与其计算精确值，不如使用代理模型来估算适应度，从而加速搜索过程并减少资源消耗。</p> <h3 id="summary">Summary</h3> <p>进化算法中的一种技术，通过估计解的质量来降低优化过程中的计算成本。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>代理建模</li> <li>计算效率</li> <li>进化算法</li> <li>选择压力</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>工程设计优化</li> <li>复杂的基于仿真的问题</li> <li>大规模参数调优</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/genetic-algorithm-%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">Genetic Algorithm (遗传算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/surrogate-model-%E4%BB%A3%E7%90%86%E6%A8%A1%E5%9E%8B/">Surrogate Model (代理模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/bayesian-optimization-%E8%B4%9D%E5%8F%B6%E6%96%AF%E4%BC%98%E5%8C%96/">Bayesian Optimization (贝叶斯优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/evolutionary-computation-%E8%BF%9B%E5%8C%96%E8%AE%A1%E7%AE%97/">Evolutionary Computation (进化计算)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/finetuned/Sat, 18 Jul 2026 11:17:25 +0000https://terms-en.ai-term-hub.com/zh/terms/finetuned/<h2 id="definition">Definition</h2> <p>微调是指采用已在大型通用数据集上训练好的模型，并在较小、特定领域的数据集上继续训练的技术。这使得模型能够保留通用知识的同时，针对特定任务进行优化和适配。</p> <h3 id="summary">Summary</h3> <p>在特定数据集上进一步训练预训练模型的过程，以使其适应特定任务。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>迁移学习</li> <li>参数调整</li> <li>领域自适应</li> <li>预训练模型</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自定义图像分类</li> <li>专用情感分析</li> <li>医疗诊断辅助</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transfer-learning-%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0/">Transfer Learning (迁移学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pre-training-%E9%A2%84%E8%AE%AD%E7%BB%83/">Pre-training (预训练)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter-tuning-%E8%B6%85%E5%8F%82%E6%95%B0%E8%B0%83%E4%BC%98/">Hyperparameter Tuning (超参数调优)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/overfitting-%E8%BF%87%E6%8B%9F%E5%90%88/">Overfitting (过拟合)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/feature_hashing/Sat, 18 Jul 2026 11:17:13 +0000https://terms-en.ai-term-hub.com/zh/terms/feature_hashing/<h2 id="definition">Definition</h2> <p>特征哈希，也称为哈希技巧（hashing trick），允许机器学习模型处理大型稀疏特征空间，而无需维护特征与索引之间的显式映射。通过应用哈希函数，模型可以直接将任意特征映射到固定的向量维度中，从而节省内存并简化特征工程流程。</p> <h3 id="summary">Summary</h3> <p>一种利用哈希函数将高维稀疏特征映射到固定大小向量的技术。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>哈希函数</li> <li>稀疏向量</li> <li>降维</li> <li>内存效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>具有大型词汇表的文本分类</li> <li>拥有海量物品集的推荐系统</li> <li>实时流数据处理</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.feature_extraction <span style="color:#f92672">import</span> FeatureHasher </span></span><span style="display:flex;"><span><span style="color:#f92672">import</span> numpy <span style="color:#66d9ef">as</span> np </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># Example: Hashing text features</span> </span></span><span style="display:flex;"><span>hasher <span style="color:#f92672">=</span> FeatureHasher(n_features<span style="color:#f92672">=</span><span style="color:#ae81ff">10</span>, input_type<span style="color:#f92672">=</span><span style="color:#e6db74">&#39;string&#39;</span>) </span></span><span style="display:flex;"><span>docs <span style="color:#f92672">=</span> [<span style="color:#e6db74">&#39;hello world&#39;</span>, <span style="color:#e6db74">&#39;hello python&#39;</span>, <span style="color:#e6db74">&#39;world python&#39;</span>] </span></span><span style="display:flex;"><span>hashed <span style="color:#f92672">=</span> hasher<span style="color:#f92672">.</span>transform(docs) </span></span><span style="display:flex;"><span>print(hashed<span style="color:#f92672">.</span>toarray()) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/one-hot-encoding-%E7%8B%AC%E7%83%AD%E7%BC%96%E7%A0%81/">One-hot encoding (独热编码)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/bag-of-words-%E8%AF%8D%E8%A2%8B%E6%A8%A1%E5%9E%8B/">Bag of Words (词袋模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/dimensionality-reduction-%E9%99%8D%E7%BB%B4/">Dimensionality reduction (降维)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sparse-matrix-%E7%A8%80%E7%96%8F%E7%9F%A9%E9%98%B5/">Sparse matrix (稀疏矩阵)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/feature_scaling/Sat, 18 Jul 2026 11:17:13 +0000https://terms-en.ai-term-hub.com/zh/terms/feature_scaling/<h2 id="definition">Definition</h2> <p>特征缩放通过标准化输入变量的范围，防止量级较大的特征主导学习过程。常见方法包括归一化（最小-最大缩放）和标准化（Z-score标准化）。这一预处理步骤对于基于距离的算法和梯度下降优化至关重要，能加速收敛并提高模型性能。</p> <h3 id="summary">Summary</h3> <p>将数据的独立变量或特征的范围进行归一化，以确保量级一致性的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>归一化</li> <li>标准化</li> <li>梯度下降</li> <li>数据预处理</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>神经网络训练</li> <li>K-means聚类</li> <li>支持向量机 (SVM)</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.preprocessing <span style="color:#f92672">import</span> StandardScaler </span></span><span style="display:flex;"><span><span style="color:#f92672">import</span> numpy <span style="color:#66d9ef">as</span> np </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span>X <span style="color:#f92672">=</span> np<span style="color:#f92672">.</span>array([[<span style="color:#ae81ff">1</span>, <span style="color:#ae81ff">2</span>], [<span style="color:#ae81ff">3</span>, <span style="color:#ae81ff">4</span>], [<span style="color:#ae81ff">5</span>, <span style="color:#ae81ff">6</span>]]) </span></span><span style="display:flex;"><span>scaler <span style="color:#f92672">=</span> StandardScaler() </span></span><span style="display:flex;"><span>X_scaled <span style="color:#f92672">=</span> scaler<span style="color:#f92672">.</span>fit_transform(X) </span></span><span style="display:flex;"><span>print(X_scaled) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/min-max-scaling-%E6%9C%80%E5%B0%8F-%E6%9C%80%E5%A4%A7%E7%BC%A9%E6%94%BE/">Min-Max Scaling (最小-最大缩放)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/z-score-normalization-z-score%E6%A0%87%E5%87%86%E5%8C%96/">Z-score Normalization (Z-score标准化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/data-preprocessing-%E6%95%B0%E6%8D%AE%E9%A2%84%E5%A4%84%E7%90%86/">Data preprocessing (数据预处理)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">Gradient Descent (梯度下降)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/feature_engineering/Sat, 18 Jul 2026 11:17:01 +0000https://terms-en.ai-term-hub.com/zh/terms/feature_engineering/<h2 id="definition">Definition</h2> <p>特征工程是利用领域专业知识，将原始数据转换为更能代表底层模式的特征的艺术。该过程包括创建新变量、转换现有数据以及选择最具预测力的特征，从而提升算法效果。</p> <h3 id="summary">Summary</h3> <p>利用领域知识创建新特征或修改现有特征，以增强机器学习模型性能的做法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>领域知识</li> <li>数据转换</li> <li>模型性能</li> <li>变量创建</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>提高回归模型准确性</li> <li>增强分类边界</li> <li>优化时间序列预测</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span>df[<span style="color:#e6db74">&#39;new_feature&#39;</span>] <span style="color:#f92672">=</span> df[<span style="color:#e6db74">&#39;feature_a&#39;</span>] <span style="color:#f92672">*</span> df[<span style="color:#e6db74">&#39;feature_b&#39;</span>] </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%89%B9%E5%BE%81%E6%8F%90%E5%8F%96/">特征提取</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%95%B0%E6%8D%AE%E9%A2%84%E5%A4%84%E7%90%86/">数据预处理</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A8%A1%E5%9E%8B%E8%B0%83%E4%BC%98/">模型调优</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%A2%86%E5%9F%9F%E4%B8%93%E9%95%BF/">领域专长</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/extremal_optimization/Sat, 18 Jul 2026 11:16:49 +0000https://terms-en.ai-term-hub.com/zh/terms/extremal_optimization/<h2 id="definition">Definition</h2> <p>与维持种群的遗传算法不同，EO作用于单个解。它识别对整体适应度贡献最小的组件，并用随机替代方案替换它。</p> <h3 id="summary">Summary</h3> <p>极值优化是一种受自组织临界性启发的启发式搜索算法，旨在通过迭代移除表现最差的组件来解决组合优化问题。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>启发式搜索</li> <li>组合优化</li> <li>自组织临界性</li> <li>局部改进</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>图划分问题</li> <li>计算集群中的任务调度</li> <li>优化网络路由路径</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">遗传算法</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A8%A1%E6%8B%9F%E9%80%80%E7%81%AB/">模拟退火</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%B2%92%E5%AD%90%E7%BE%A4%E4%BC%98%E5%8C%96/">粒子群优化</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%85%83%E5%90%AF%E5%8F%91%E5%BC%8F%E7%AE%97%E6%B3%95/">元启发式算法</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/explorationexploitation_dilemma/Sat, 18 Jul 2026 11:16:49 +0000https://terms-en.ai-term-hub.com/zh/terms/explorationexploitation_dilemma/<h2 id="definition">Definition</h2> <p>在决策过程中，智能体面临权衡：它们可以利用当前知识获得最佳即时奖励，或者探索未知选项以潜在地找到更好的长期策略。</p> <h3 id="summary">Summary</h3> <p>探索-利用困境是强化学习中的一个基本问题，智能体必须在探索新动作以收集信息和利用已知动作以最大化奖励之间做出选择。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>奖励最大化</li> <li>ε-贪婪算法</li> <li>上置信界</li> <li>强化学习</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>推荐系统决定是否建议热门商品还是新商品</li> <li>营销活动中的A/B测试</li> <li>动态环境中的资源分配</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">强化学习</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%A4%9A%E8%87%82%E8%80%81%E8%99%8E%E6%9C%BA/">多臂老虎机</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/q%E5%AD%A6%E4%B9%A0/">Q学习</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%AD%96%E7%95%A5%E6%A2%AF%E5%BA%A6/">策略梯度</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/evolvability/Sat, 18 Jul 2026 11:16:37 +0000https://terms-en.ai-term-hub.com/zh/terms/evolvability/<h2 id="definition">Definition</h2> <p>在计算语境中，可进化性指算法或神经网络架构在代际或训练步骤中提高其适应度的难易程度。高可进化性意味着微小的变化就能导致显著的适应性改进，这对于进化算法和神经架构搜索至关重要。</p> <h3 id="summary">Summary</h3> <p>基因型或系统产生可遗传的表型变异以供选择适应的能力。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>遗传变异</li> <li>适应速率</li> <li>搜索空间结构</li> <li>鲁棒性</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>遗传算法优化</li> <li>神经架构搜索</li> <li>研究生物进化模拟</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/genetic_algorithms-%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">genetic_algorithms (遗传算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fitness_landscape-%E9%80%82%E5%BA%94%E5%BA%A6%E6%99%AF%E8%A7%82/">fitness_landscape (适应度景观)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/mutation_rate-%E7%AA%81%E5%8F%98%E7%8E%87/">mutation_rate (突变率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/neuroevolution-%E7%A5%9E%E7%BB%8F%E8%BF%9B%E5%8C%96/">neuroevolution (神经进化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/empirical_risk_minimization/Sat, 18 Jul 2026 11:16:12 +0000https://terms-en.ai-term-hub.com/zh/terms/empirical_risk_minimization/<h2 id="definition">Definition</h2> <p>经验风险最小化（ERM）是训练监督学习模型的标准目标函数。它涉及从一组函数中选择假设，以最小化训练数据上的平均误差（即经验风险）。其核心思想是通过优化训练集上的表现来逼近真实的风险分布。</p> <h3 id="summary">Summary</h3> <p>机器学习中的一个基本原则，旨在最小化观测训练数据上的平均损失。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>损失函数</li> <li>训练误差</li> <li>泛化能力</li> <li>过拟合</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练逻辑回归模型</li> <li>优化神经网络权重</li> <li>支持向量机构建</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/structural-risk-minimization-%E7%BB%93%E6%9E%84%E9%A3%8E%E9%99%A9%E6%9C%80%E5%B0%8F%E5%8C%96/">Structural risk minimization (结构风险最小化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/regularization-%E6%AD%A3%E5%88%99%E5%8C%96/">Regularization (正则化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/loss-function-%E6%8D%9F%E5%A4%B1%E5%87%BD%E6%95%B0/">Loss function (损失函数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/bias-variance-tradeoff-%E5%81%8F%E5%B7%AE-%E6%96%B9%E5%B7%AE%E6%9D%83%E8%A1%A1/">Bias-variance tradeoff (偏差-方差权衡)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/efficientnet/Sat, 18 Jul 2026 11:15:59 +0000https://terms-en.ai-term-hub.com/zh/terms/efficientnet/<h2 id="definition">Definition</h2> <p>由Google开发，EfficientNet使用复合缩放方法来平衡网络深度、宽度和输入图像分辨率。这种方法使模型能够在保持参数效率的同时达到最先进的准确率水平。</p> <h3 id="summary">Summary</h3> <p>EfficientNet是一系列卷积神经网络架构，通过均匀扩展深度、宽度和分辨率，以更少的参数实现更高的准确率。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>复合缩放 (Compound scaling)</li> <li>MobileNetV2块</li> <li>SE注意力机制 (SE attention)</li> <li>参数效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>移动设备上的图像分类</li> <li>资源受限环境中的物体检测</li> <li>大规模视觉数据集的特征提取</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/resnet/">ResNet</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/mobilenet/">MobileNet</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8D%B7%E7%A7%AF%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C-convolutional-neural-network/">卷积神经网络 (Convolutional Neural Network)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%A5%9E%E7%BB%8F%E6%9E%B6%E6%9E%84%E6%90%9C%E7%B4%A2-neural-architecture-search/">神经架构搜索 (Neural Architecture Search)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/edge_inference/Sat, 18 Jul 2026 11:15:59 +0000https://terms-en.ai-term-hub.com/zh/terms/edge_inference/<h2 id="definition">Definition</h2> <p>这种做法涉及将训练好的AI模型直接部署到智能手机、物联网传感器或嵌入式系统等硬件上。通过在本地处理数据，边缘推理显著降低了延迟，并提高了响应速度。</p> <h3 id="summary">Summary</h3> <p>边缘推理是指将机器学习模型在终端用户设备上本地执行，而不是在集中式云服务器上运行的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>降低延迟</li> <li>隐私保护</li> <li>模型量化</li> <li>本地处理</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自主无人机中的实时物体检测</li> <li>智能音箱上的语音识别</li> <li>工业物联网传感器上的预测性维护</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%BA%91%E8%AE%A1%E7%AE%97-cloud-computing/">云计算 (Cloud computing)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%BE%AE%E5%9E%8B%E6%9C%BA%E5%99%A8%E5%AD%A6%E4%B9%A0-tinyml/">微型机器学习 (TinyML)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%81%94%E9%82%A6%E5%AD%A6%E4%B9%A0-federated-learning/">联邦学习 (Federated learning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%8E%A8%E7%90%86%E5%BC%95%E6%93%8E-inference-engine/">推理引擎 (Inference engine)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/em_algorithm_and_gmm_model/Sat, 18 Jul 2026 11:15:46 +0000https://terms-en.ai-term-hub.com/zh/terms/em_algorithm_and_gmm_model/<h2 id="definition">Definition</h2> <p>该术语指期望最大化（EM）算法与高斯混合模型（GMM）之间的协同关系。GMM假设所有数据点都是由多个分布混合生成的，而EM算法则通过迭代方式优化这些混合成分的参数（如均值、协方差和权重），从而实现对数据的最佳拟合。</p> <h3 id="summary">Summary</h3> <p>期望最大化（EM）算法是一种用于在包含隐变量的统计模型中寻找最大似然估计的迭代方法，常用于拟合高斯混合模型。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>隐变量</li> <li>最大似然估计</li> <li>迭代优化</li> <li>高斯分布</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>语音识别中的音素建模</li> <li>图像分割</li> <li>客户细分分析</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/k-means%E8%81%9A%E7%B1%BB-%E4%B8%80%E7%A7%8D%E5%9F%BA%E4%BA%8E%E8%B7%9D%E7%A6%BB%E7%9A%84%E5%88%92%E5%88%86%E8%81%9A%E7%B1%BB%E7%AE%97%E6%B3%95/">K-means聚类 (一种基于距离的划分聚类算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%9A%90%E9%A9%AC%E5%B0%94%E5%8F%AF%E5%A4%AB%E6%A8%A1%E5%9E%8B-%E7%94%A8%E4%BA%8E%E5%BA%8F%E5%88%97%E6%95%B0%E6%8D%AE%E7%9A%84%E7%BB%9F%E8%AE%A1%E6%A8%A1%E5%9E%8B/">隐马尔可夫模型 (用于序列数据的统计模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%B4%9D%E5%8F%B6%E6%96%AF%E6%8E%A8%E6%96%AD-%E5%9F%BA%E4%BA%8E%E8%B4%9D%E5%8F%B6%E6%96%AF%E5%AE%9A%E7%90%86%E7%9A%84%E6%A6%82%E7%8E%87%E6%8E%A8%E7%90%86%E6%96%B9%E6%B3%95/">贝叶斯推断 (基于贝叶斯定理的概率推理方法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%AF%86%E5%BA%A6%E4%BC%B0%E8%AE%A1-%E4%BC%B0%E8%AE%A1%E9%9A%8F%E6%9C%BA%E5%8F%98%E9%87%8F%E6%A6%82%E7%8E%87%E5%AF%86%E5%BA%A6%E5%87%BD%E6%95%B0%E7%9A%84%E8%BF%87%E7%A8%8B/">密度估计 (估计随机变量概率密度函数的过程)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/early_stopping/Sat, 18 Jul 2026 11:15:46 +0000https://terms-en.ai-term-hub.com/zh/terms/early_stopping/<h2 id="definition">Definition</h2> <p>早停法是一种主要用于梯度下降等迭代训练过程中的正则化形式。在训练期间，模型在训练数据上的表现通常会持续改善，但在验证集上的表现可能会在某个点后开始恶化（表明出现过拟合）。早停法通过监控验证误差并在连续多个epoch未改善时停止训练来解决这一问题。</p> <h3 id="summary">Summary</h3> <p>早停法是一种正则化技术，当模型在验证集上的性能开始下降时停止训练过程，以防止过拟合。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>正则化</li> <li>验证集</li> <li>防止过拟合</li> <li>耐心参数</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>神经网络训练</li> <li>梯度提升算法</li> <li>时间序列预测模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/l2%E6%AD%A3%E5%88%99%E5%8C%96-%E6%9D%83%E9%87%8D%E8%A1%B0%E5%87%8F%E6%8A%80%E6%9C%AF/">L2正则化 (权重衰减技术)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/dropout-%E9%9A%8F%E6%9C%BA%E4%B8%A2%E5%BC%83%E7%A5%9E%E7%BB%8F%E5%85%83%E7%9A%84%E6%8A%80%E6%9C%AF/">Dropout (随机丢弃神经元的技术)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%BA%A4%E5%8F%89%E9%AA%8C%E8%AF%81-%E8%AF%84%E4%BC%B0%E6%A8%A1%E5%9E%8B%E6%B3%9B%E5%8C%96%E8%83%BD%E5%8A%9B%E7%9A%84%E6%8A%80%E6%9C%AF/">交叉验证 (评估模型泛化能力的技术)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%B3%9B%E5%8C%96%E8%AF%AF%E5%B7%AE-%E6%A8%A1%E5%9E%8B%E5%AF%B9%E6%9C%AA%E8%A7%81%E6%95%B0%E6%8D%AE%E9%A2%84%E6%B5%8B%E8%AF%AF%E5%B7%AE%E7%9A%84%E4%BC%B0%E8%AE%A1/">泛化误差 (模型对未见数据预测误差的估计)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/diffusersltxpipeline/Sat, 18 Jul 2026 11:14:55 +0000https://terms-en.ai-term-hub.com/zh/terms/diffusersltxpipeline/<h2 id="definition">Definition</h2> <p>LTX 管道专为优先考虑生成任务速度和效率的模型量身定制，通常利用蒸馏或加速采样方法。它与 Diffusers 无缝集成。</p> <h3 id="summary">Summary</h3> <p>Diffusers 中针对 LTX（闪电文本转视频或类似高速生成）模型优化的管道实现，侧重于快速推理。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>快速推理</li> <li>模型蒸馏</li> <li>实时生成</li> <li>优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>实时交互式艺术装置</li> <li>视频概念的快速原型开发</li> <li>低延迟内容创作工具</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/ltx-%E6%A8%A1%E5%9E%8B-ltx-model/">LTX 模型 (LTX Model)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%8E%A8%E7%90%86%E9%80%9F%E5%BA%A6-inference-speed/">推理速度 (Inference Speed)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A8%A1%E5%9E%8B%E8%92%B8%E9%A6%8F-model-distillation/">模型蒸馏 (Model Distillation)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/diffusers/">Diffusers</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/differentially_private_stochastic_gradient_descent/Sat, 18 Jul 2026 11:14:55 +0000https://terms-en.ai-term-hub.com/zh/terms/differentially_private_stochastic_gradient_descent/<h2 id="definition">Definition</h2> <p>DP-SGD 是随机梯度下降的一种变体，旨在保护训练数据的隐私。它通过裁剪每个样本梯度的贡献来限制敏感度，然后添加高斯噪声来实现隐私保护。</p> <h3 id="summary">Summary</h3> <p>一种优化算法，通过对标准随机梯度下降（SGD）进行梯度裁剪和添加噪声的修改，确保训练后的模型满足差分隐私约束。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度裁剪</li> <li>高斯噪声注入</li> <li>样本子采样</li> <li>隐私会计</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在私有用户数据上训练深度神经网络</li> <li>医疗保健预测建模</li> <li>使用受监管数据进行金融欺诈检测</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%B7%AE%E5%88%86%E9%9A%90%E7%A7%81-differential-privacy/">差分隐私 (Differential Privacy)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%9A%8F%E6%9C%BA%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D-sgd/">随机梯度下降 (SGD)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A8%A1%E5%9E%8B%E5%8F%8D%E8%BD%AC%E6%94%BB%E5%87%BB-model-inversion-attacks/">模型反转攻击 (Model Inversion Attacks)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%9A%90%E7%A7%81%E9%A2%84%E7%AE%97-privacy-budget/">隐私预算 (Privacy Budget)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/decision_tree_pruning/Sat, 18 Jul 2026 11:13:51 +0000https://terms-en.ai-term-hub.com/zh/terms/decision_tree_pruning/<h2 id="definition">Definition</h2> <p>剪枝是一种用于防止决策树模型过拟合的方法，通过移除具有弱预测能力的分支来实现。它可以以预剪枝（提前停止树的生长）或后剪枝的方式执行&hellip;</p> <h3 id="summary">Summary</h3> <p>一种通过移除对分类实例贡献较小的部分来减小决策树规模的技术。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>防止过拟合</li> <li>预剪枝</li> <li>后剪枝</li> <li>模型复杂度</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>提高模型泛化能力</li> <li>降低推理延迟</li> <li>简化规则提取</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%AD%A3%E5%88%99%E5%8C%96-regularization/">正则化 (Regularization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%BA%A4%E5%8F%89%E9%AA%8C%E8%AF%81-cross-validation/">交叉验证 (Cross-validation)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%86%B5-entropy/">熵 (Entropy)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%BF%A1%E6%81%AF%E5%A2%9E%E7%9B%8A-information-gain/">信息增益 (Information gain)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/cross_entropy_method/Sat, 18 Jul 2026 11:12:12 +0000https://terms-en.ai-term-hub.com/zh/terms/cross_entropy_method/<h2 id="definition">Definition</h2> <p>交叉熵法（CEM）是一种强大的通用优化算法，用于解决离散和连续问题。它通过维护搜索空间上的概率分布来工作，并通过迭代更新该分布以聚焦于高性能样本，从而逐步优化解决方案。</p> <h3 id="summary">Summary</h3> <p>一种随机优化技术，利用蒙特卡洛模拟迭代改进对罕见事件概率的估计。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>蒙特卡洛模拟</li> <li>迭代优化</li> <li>概率分布更新</li> <li>精英样本</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>机器人路径规划</li> <li>游戏AI策略优化</li> <li>风险分析中的罕见事件估计</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/simulated-annealing-%E6%A8%A1%E6%8B%9F%E9%80%80%E7%81%AB/">Simulated Annealing (模拟退火)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/genetic-algorithms-%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">Genetic Algorithms (遗传算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reinforcement-learning-%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">Reinforcement Learning (强化学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/cost_sensitive_machine_learning/Sat, 18 Jul 2026 11:11:39 +0000https://terms-en.ai-term-hub.com/zh/terms/cost_sensitive_machine_learning/<h2 id="definition">Definition</h2> <p>代价敏感机器学习通过为不同类型的错误分配不同的惩罚来扩展传统的监督学习。在现实场景中，假阳性和假阴性往往具有不同的后果&hellip;</p> <h3 id="summary">Summary</h3> <p>一种将误分类代价纳入训练过程的机器学习范式，旨在优化经济影响而非仅仅追求准确率。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>损失函数修改</li> <li>类别不平衡</li> <li>误分类代价</li> <li>优化目标</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>银行欺诈检测</li> <li>医学疾病筛查</li> <li>高假阳性代价的垃圾邮件过滤</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E4%B8%8D%E5%B9%B3%E8%A1%A1%E5%AD%A6%E4%B9%A0-imbalanced-learning/">不平衡学习 (Imbalanced Learning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%B2%BE%E7%A1%AE%E7%8E%87-%E5%8F%AC%E5%9B%9E%E7%8E%87%E6%9D%83%E8%A1%A1-precision-recall-tradeoff/">精确率-召回率权衡 (Precision-Recall Tradeoff)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/roc-%E6%9B%B2%E7%BA%BF-receiver-operating-characteristic-curve/">ROC 曲线 (Receiver Operating Characteristic Curve)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8A%A0%E6%9D%83%E6%8D%9F%E5%A4%B1-weighted-loss/">加权损失 (Weighted Loss)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/contrastive_learning/Sat, 18 Jul 2026 11:11:27 +0000https://terms-en.ai-term-hub.com/zh/terms/contrastive_learning/<h2 id="definition">Definition</h2> <p>对比学习是一种不需要标记数据的表示学习方法。它通过对同一输入创建增强视图（正样本对）并将它们与不同的（负样本）进行对比来工作。</p> <h3 id="summary">Summary</h3> <p>一种自监督学习技术，通过将正样本对拉近并将负样本对推开来学习表示。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>自监督</li> <li>正/负样本对</li> <li>嵌入空间</li> <li>增强策略</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>无标签图像分类</li> <li>语义搜索索引</li> <li>时间序列异常检测</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/simclr/">SimCLR</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/moco/">MoCo</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%87%AA%E7%9B%91%E7%9D%A3%E5%AD%A6%E4%B9%A0-self-supervised-learning/">自监督学习 (Self-Supervised Learning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%A1%A8%E7%A4%BA%E5%AD%A6%E4%B9%A0-representation-learning/">表示学习 (Representation Learning)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/computational_heuristic_intelligence/Sat, 18 Jul 2026 11:10:50 +0000https://terms-en.ai-term-hub.com/zh/terms/computational_heuristic_intelligence/<h2 id="definition">Definition</h2> <p>计算启发式智能涉及采用经验法则、近似值或合理猜测，在合理时间内找到令人满意的解决方案的算法。与穷举搜索不同，它侧重于效率而非绝对最优解。</p> <h3 id="summary">Summary</h3> <p>当精确方法过于缓慢时，使用基于经验和实用技巧来高效解决问题的AI方法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>近似</li> <li>经验法则</li> <li>搜索优化</li> <li>时间复杂度</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>物流路线规划</li> <li>国际象棋和围棋引擎走法</li> <li>云计算中的资源分配</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/genetic-algorithms-%E9%81%97%E4%BC%A0%E7%AE%97%E6%B3%95/">Genetic Algorithms (遗传算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/simulated-annealing-%E6%A8%A1%E6%8B%9F%E9%80%80%E7%81%AB/">Simulated Annealing (模拟退火)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/a-search-a-%E6%90%9C%E7%B4%A2%E7%AE%97%E6%B3%95/">A* Search (A*搜索算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/metaheuristics-%E5%85%83%E5%90%AF%E5%8F%91%E5%BC%8F%E7%AE%97%E6%B3%95/">Metaheuristics (元启发式算法)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/compressed_tensors/Sat, 18 Jul 2026 11:10:50 +0000https://terms-en.ai-term-hub.com/zh/terms/compressed_tensors/<h2 id="definition">Definition</h2> <p>压缩张量是深度学习中使用的多维数组，其数值精度（例如从float32降至int8）或稀疏性已降低。这种技术被称为量化或剪枝。</p> <h3 id="summary">Summary</h3> <p>通过降低数据精度或大小以优化存储和计算效率的张量。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>量化</li> <li>稀疏性</li> <li>内存优化</li> <li>推理速度</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>移动AI应用部署</li> <li>边缘设备处理</li> <li>大型语言模型服务优化</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch </span></span><span style="display:flex;"><span><span style="color:#75715e"># Example of converting a tensor to half precision</span> </span></span><span style="display:flex;"><span>x <span style="color:#f92672">=</span> torch<span style="color:#f92672">.</span>randn(<span style="color:#ae81ff">10</span>, <span style="color:#ae81ff">10</span>) </span></span><span style="display:flex;"><span>x_compressed <span style="color:#f92672">=</span> x<span style="color:#f92672">.</span>half() </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/quantization-%E9%87%8F%E5%8C%96/">Quantization (量化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pruning-%E5%89%AA%E6%9E%9D/">Pruning (剪枝)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-distillation-%E6%A8%A1%E5%9E%8B%E8%92%B8%E9%A6%8F/">Model Distillation (模型蒸馏)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/float16-%E5%8D%8A%E7%B2%BE%E5%BA%A6%E6%B5%AE%E7%82%B9%E6%95%B0/">Float16 (半精度浮点数)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/clip/Sat, 18 Jul 2026 11:10:07 +0000https://terms-en.ai-term-hub.com/zh/terms/clip/<h2 id="definition">Definition</h2> <p>在深度学习工程中，裁剪通常应用于梯度以缓解梯度爆炸问题，确保反向传播的稳定性。它也可以指在应用激活函数之前限制输出 logits 的范围。</p> <h3 id="summary">Summary</h3> <p>裁剪是一种限制数值（如梯度或输出概率）幅度的技术，旨在防止训练过程中的数值不稳定。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度裁剪</li> <li>数值稳定性</li> <li>梯度爆炸</li> <li>正则化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>循环神经网络训练</li> <li>稳定 Transformer 训练</li> <li>防止损失发散</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/learning-rate-%E5%AD%A6%E4%B9%A0%E7%8E%87/">Learning Rate (学习率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/backpropagation-%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD/">Backpropagation (反向传播)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/vanishing-gradient-%E6%A2%AF%E5%BA%A6%E6%B6%88%E5%A4%B1/">Vanishing Gradient (梯度消失)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/normalization-%E5%BD%92%E4%B8%80%E5%8C%96/">Normalization (归一化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/caching/Sat, 18 Jul 2026 11:09:42 +0000https://terms-en.ai-term-hub.com/zh/terms/caching/<h2 id="definition">Definition</h2> <p>在AI工程中，缓存通过将最近或频繁的查询结果、模型预测或中间计算保留在快速内存（如RAM）中来优化性能。这减少了昂贵的主数据存储访问需求，从而显著提升系统响应速度。</p> <h3 id="summary">Summary</h3> <p>缓存是一种将频繁访问的数据存储在临时高速存储层中的技术，旨在降低延迟并减少主数据源的负载。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>延迟降低</li> <li>内存优化</li> <li>驱逐策略</li> <li>命中率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>存储模型推理结果</li> <li>缓存数据库查询输出</li> <li>预计算特征嵌入</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> redis </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># Simple caching example</span> </span></span><span style="display:flex;"><span>r <span style="color:#f92672">=</span> redis<span style="color:#f92672">.</span>Redis(host<span style="color:#f92672">=</span><span style="color:#e6db74">&#39;localhost&#39;</span>, port<span style="color:#f92672">=</span><span style="color:#ae81ff">6379</span>, db<span style="color:#f92672">=</span><span style="color:#ae81ff">0</span>) </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#66d9ef">def</span> <span style="color:#a6e22e">get_prediction</span>(model_id, input_data): </span></span><span style="display:flex;"><span> cache_key <span style="color:#f92672">=</span> <span style="color:#e6db74">f</span><span style="color:#e6db74">&#34;pred_</span><span style="color:#e6db74">{</span>model_id<span style="color:#e6db74">}</span><span style="color:#e6db74">_</span><span style="color:#e6db74">{</span>hash(str(input_data))<span style="color:#e6db74">}</span><span style="color:#e6db74">&#34;</span> </span></span><span style="display:flex;"><span> result <span style="color:#f92672">=</span> r<span style="color:#f92672">.</span>get(cache_key) </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">if</span> result: </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">return</span> result<span style="color:#f92672">.</span>decode(<span style="color:#e6db74">&#39;utf-8&#39;</span>) </span></span><span style="display:flex;"><span> <span style="color:#75715e"># Compute if not cached</span> </span></span><span style="display:flex;"><span> prediction <span style="color:#f92672">=</span> model<span style="color:#f92672">.</span>predict(input_data) </span></span><span style="display:flex;"><span> r<span style="color:#f92672">.</span>setex(cache_key, <span style="color:#ae81ff">3600</span>, str(prediction)) </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">return</span> prediction </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/redis-%E5%86%85%E5%AD%98%E6%95%B0%E6%8D%AE%E7%BB%93%E6%9E%84%E5%AD%98%E5%82%A8/">Redis (内存数据结构存储)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/memcached-%E5%88%86%E5%B8%83%E5%BC%8F%E5%86%85%E5%AD%98%E7%BC%93%E5%AD%98%E7%B3%BB%E7%BB%9F/">memcached (分布式内存缓存系统)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/performance-tuning-%E6%80%A7%E8%83%BD%E8%B0%83%E4%BC%98/">performance tuning (性能调优)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/database-indexing-%E6%95%B0%E6%8D%AE%E5%BA%93%E7%B4%A2%E5%BC%95/">database indexing (数据库索引)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/bayesian_optimization/Sat, 18 Jul 2026 11:08:51 +0000https://terms-en.ai-term-hub.com/zh/terms/bayesian_optimization/<h2 id="definition">Definition</h2> <p>贝叶斯优化使用概率代理模型（通常为高斯过程）来建模目标函数。它采用采集函数来平衡探索与利用，从而高效地找到最优参数配置。</p> <h3 id="summary">Summary</h3> <p>一种用于全局优化昂贵评估黑盒函数的顺序设计策略。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>代理模型</li> <li>采集函数</li> <li>探索与利用权衡</li> <li>黑盒优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>深度学习模型的超参数调优</li> <li>科学实验设计的优化</li> <li>机器人控制参数调整</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter_tuning-%E8%B6%85%E5%8F%82%E6%95%B0%E8%B0%83%E4%BC%98/">hyperparameter_tuning (超参数调优)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gaussian_processes-%E9%AB%98%E6%96%AF%E8%BF%87%E7%A8%8B/">gaussian_processes (高斯过程)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/acquisition_function-%E9%87%87%E9%9B%86%E5%87%BD%E6%95%B0/">acquisition_function (采集函数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/grid_search-%E7%BD%91%E6%A0%BC%E6%90%9C%E7%B4%A2/">grid_search (网格搜索)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/batch_size/Sat, 18 Jul 2026 11:08:51 +0000https://terms-en.ai-term-hub.com/zh/terms/batch_size/<h2 id="definition">Definition</h2> <p>批大小是一个关键的超参数，决定了在更新模型内部参数之前处理多少个样本。较大的批大小能提供更准确的梯度估计，但可能增加内存需求并影响收敛特性。</p> <h3 id="summary">Summary</h3> <p>在随机梯度下降算法的一次迭代中使用的训练样本数量。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度估计</li> <li>内存限制</li> <li>收敛稳定性</li> <li>超参数调整</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>调整模型收敛速度</li> <li>训练期间管理GPU内存限制</li> <li>通过噪声梯度提高泛化能力</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/learning_rate-%E5%AD%A6%E4%B9%A0%E7%8E%87/">learning_rate (学习率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/stochastic_gradient_descent-%E9%9A%8F%E6%9C%BA%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">stochastic_gradient_descent (随机梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/mini_batch-%E5%B0%8F%E6%89%B9%E9%87%8F/">mini_batch (小批量)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/memory_usage-%E5%86%85%E5%AD%98%E4%BD%BF%E7%94%A8/">memory_usage (内存使用)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/batch_normalization/Sat, 18 Jul 2026 11:08:27 +0000https://terms-en.ai-term-hub.com/zh/terms/batch_normalization/<h2 id="definition">Definition</h2> <p>该方法在训练期间调整并缩放激活值，使其在每个小批量中具有零均值和单位方差。它减少了内部协变量偏移，允许使用更高的学习率和更快的收敛速度。</p> <h3 id="summary">Summary</h3> <p>批归一化是一种技术，通过对小批量数据上的层输入进行归一化，以稳定并加速神经网络的训练。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>内部协变量偏移</li> <li>小批量统计量</li> <li>梯度稳定</li> <li>正则化效应</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>深度神经网络</li> <li>卷积神经网络</li> <li>训练优化</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.nn <span style="color:#66d9ef">as</span> nn </span></span><span style="display:flex;"><span>layer <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>Sequential( </span></span><span style="display:flex;"><span> nn<span style="color:#f92672">.</span>Linear(<span style="color:#ae81ff">10</span>, <span style="color:#ae81ff">20</span>), </span></span><span style="display:flex;"><span> nn<span style="color:#f92672">.</span>BatchNorm1d(<span style="color:#ae81ff">20</span>), </span></span><span style="display:flex;"><span> nn<span style="color:#f92672">.</span>ReLU() </span></span><span style="display:flex;"><span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/layer-normalization-%E5%B1%82%E5%BD%92%E4%B8%80%E5%8C%96/">Layer Normalization (层归一化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">Gradient Descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/overfitting-%E8%BF%87%E6%8B%9F%E5%90%88/">Overfitting (过拟合)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/alphachip/Sat, 18 Jul 2026 11:04:49 +0000https://terms-en.ai-term-hub.com/zh/terms/alphachip/<h2 id="definition">Definition</h2> <p>AlphaChip是一个专门的AI系统，旨在自动化并增强微芯片上组件的放置和布线过程。通过采用深度强化学习，它显著缩短了设计周期。</p> <h3 id="summary">Summary</h3> <p>由Google DeepMind开发的AI驱动芯片设计工具，利用强化学习优化集成电路布局。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>强化学习</li> <li>芯片放置</li> <li>硬件优化</li> <li>DeepMind</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>微芯片布局优化</li> <li>缩短设计周期时间</li> <li>提高处理器的能效</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/eda-tools-%E7%94%B5%E5%AD%90%E8%AE%BE%E8%AE%A1%E8%87%AA%E5%8A%A8%E5%8C%96%E5%B7%A5%E5%85%B7/">EDA tools (电子设计自动化工具)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/reinforcement-learning-%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/">Reinforcement learning (强化学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/integrated-circuits-%E9%9B%86%E6%88%90%E7%94%B5%E8%B7%AF/">Integrated circuits (集成电路)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/google-deepmind-%E8%B0%B7%E6%AD%8Cdeepmind/">Google DeepMind (谷歌DeepMind)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/algorithm_selection/Sat, 18 Jul 2026 11:04:37 +0000https://terms-en.ai-term-hub.com/zh/terms/algorithm_selection/<h2 id="definition">Definition</h2> <p>算法选择涉及评估不同的计算方法，以确定哪一种能最高效地解决给定任务。此过程考虑的因素包括时间复杂度、空间复杂</p> <h3 id="summary">Summary</h3> <p>算法选择是根据性能指标和约束条件，为特定问题选择最合适的计算方法的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>性能指标</li> <li>复杂度分析</li> <li>基准测试</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>为大型数据集选择排序算法</li> <li>为机器学习流水线选择分类器</li> <li>优化数据库查询执行计划</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/big-o-notation-%E5%A4%A7o%E7%AC%A6%E5%8F%B7/">Big O Notation (大O符号)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter-tuning-%E8%B6%85%E5%8F%82%E6%95%B0%E8%B0%83%E4%BC%98/">Hyperparameter Tuning (超参数调优)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-evaluation-%E6%A8%A1%E5%9E%8B%E8%AF%84%E4%BC%B0/">Model Evaluation (模型评估)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/admissible_heuristic/Sat, 18 Jul 2026 11:04:13 +0000https://terms-en.ai-term-hub.com/zh/terms/admissible_heuristic/<h2 id="definition">Definition</h2> <p>在路径查找和搜索问题中，可容许启发式函数提供了到达目标节点实际成本的下界。通过保证估计成本始终小于或等于实际成本，它确保了搜索算法（如A*）能找到最优解。</p> <h3 id="summary">Summary</h3> <p>搜索算法中的一种启发式函数，它从不高估到达目标的真实成本，从而保证最优性。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>下界</li> <li>最优性保证</li> <li>A*搜索</li> <li>成本估计</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>GPS导航路线规划</li> <li>解谜游戏（如八数码问题）</li> <li>障碍物丰富环境中的机器人运动规划</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/consistent-heuristic-%E4%B8%80%E8%87%B4%E5%90%AF%E5%8F%91%E5%BC%8F%E5%87%BD%E6%95%B0/">consistent heuristic (一致启发式函数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/a-algorithm-a-%E7%AE%97%E6%B3%95/">A* algorithm (A*算法)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/search-optimization-%E6%90%9C%E7%B4%A2%E4%BC%98%E5%8C%96/">search optimization (搜索优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pathfinding-%E8%B7%AF%E5%BE%84%E6%9F%A5%E6%89%BE/">pathfinding (路径查找)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/accelerated_linear_algebra/Sat, 18 Jul 2026 11:04:01 +0000https://terms-en.ai-term-hub.com/zh/terms/accelerated_linear_algebra/<h2 id="definition">Definition</h2> <p>该领域专注于加速基本的线性代数计算，这些计算是机器学习和科学模拟的核心。通过利用 GPU、TPU 和其他并行处理能力的优势，显著提高了大规模矩阵运算的速度和效率。</p> <h3 id="summary">Summary</h3> <p>加速线性代数涉及使用 GPU 和 TPU 等硬件加速器优化矩阵运算，以实现高性能计算。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>GPU 计算</li> <li>矩阵乘法</li> <li>并行处理</li> <li>CUDA</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>深度学习模型训练</li> <li>科学模拟</li> <li>实时图形渲染</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/blas-%E5%9F%BA%E7%A1%80%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0%E5%AD%90%E7%A8%8B%E5%BA%8F/">BLAS (基础线性代数子程序)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/cudnn-cuda-%E6%B7%B1%E5%BA%A6%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C%E5%BA%93/">cuDNN (CUDA 深度神经网络库)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/tensor-cores-%E5%BC%A0%E9%87%8F%E6%A0%B8%E5%BF%83/">Tensor Cores (张量核心)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/ab_testing/Sat, 18 Jul 2026 11:02:40 +0000https://terms-en.ai-term-hub.com/zh/terms/ab_testing/<h2 id="definition">Definition</h2> <p>A/B 测试是一种随机对照实验，通过比较两个变体（A 和 B）来评估哪一个在特定指标上能产生更好的结果。在人工智能工程中，它对优化至关重要。</p> <h3 id="summary">Summary</h3> <p>一种统计方法，通过比较变量的两个版本来确定哪个表现更好。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>对照组</li> <li>统计显著性</li> <li>假设检验</li> <li>随机化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>优化推荐引擎的点击率</li> <li>比较不同模型架构的准确性</li> <li>测试 AI 驱动应用中的用户界面更改</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/multivariate-testing-%E5%A4%9A%E5%8F%98%E9%87%8F%E6%B5%8B%E8%AF%95/">Multivariate Testing (多变量测试)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/statistical-power-%E7%BB%9F%E8%AE%A1%E5%8A%9F%E6%95%88/">Statistical Power (统计功效)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/conversion-rate-optimization-%E8%BD%AC%E5%8C%96%E7%8E%87%E4%BC%98%E5%8C%96/">Conversion Rate Optimization (转化率优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/randomized-controlled-trial-%E9%9A%8F%E6%9C%BA%E5%AF%B9%E7%85%A7%E8%AF%95%E9%AA%8C/">Randomized Controlled Trial (随机对照试验)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/residual_connection/Sat, 18 Jul 2026 11:01:52 +0000https://terms-en.ai-term-hub.com/zh/terms/residual_connection/<h2 id="definition">Definition</h2> <p>残差连接（也称为跳跃连接）通过将输入直接添加到后续层的输出来允许梯度在网络中流动。这种架构解决了深层网络中的梯度消失问题。</p> <h3 id="summary">Summary</h3> <p>一种将输入直接加到层输出上的机制，以促进深层网络中的梯度流动。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>跳跃连接</li> <li>梯度消失问题</li> <li>深度残差学习</li> <li>梯度流</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练深层卷积神经网络</li> <li>Transformer 架构</li> <li>图像分类模型</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.nn <span style="color:#66d9ef">as</span> nn </span></span><span style="display:flex;"><span><span style="color:#66d9ef">class</span> <span style="color:#a6e22e">ResidualBlock</span>(nn<span style="color:#f92672">.</span>Module): </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">def</span> __init__(self, channels): </span></span><span style="display:flex;"><span> super()<span style="color:#f92672">.</span>__init__() </span></span><span style="display:flex;"><span> self<span style="color:#f92672">.</span>conv <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>Conv2d(channels, channels, <span style="color:#ae81ff">3</span>, padding<span style="color:#f92672">=</span><span style="color:#ae81ff">1</span>) </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">def</span> <span style="color:#a6e22e">forward</span>(self, x): </span></span><span style="display:flex;"><span> <span style="color:#66d9ef">return</span> x <span style="color:#f92672">+</span> self<span style="color:#f92672">.</span>conv(x) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/skip_connection-%E8%B7%B3%E8%B7%83%E8%BF%9E%E6%8E%A5/">skip_connection (跳跃连接)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/vanishing_gradient-%E6%A2%AF%E5%BA%A6%E6%B6%88%E5%A4%B1/">vanishing_gradient (梯度消失)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/deep_learning-%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/">deep_learning (深度学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/resnet-%E6%AE%8B%E5%B7%AE%E7%BD%91%E7%BB%9C/">resnet (残差网络)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/qlora/Sat, 18 Jul 2026 11:01:29 +0000https://terms-en.ai-term-hub.com/zh/terms/qlora/<h2 id="definition">Definition</h2> <p>QLoRA 将低秩适应（LoRA）与 4 位量化相结合，显著减少了微调大规模模型所需的内存占用。通过将权重存储为 4 位格式并添加训练好的低秩适配器，该方法在保持性能的同时降低了资源需求。</p> <h3 id="summary">Summary</h3> <p>量化低秩适应，一种使用 4 位量化和低秩适配器高效微调大型语言模型的方法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>低秩适应</li> <li>4 位量化</li> <li>内存效率</li> <li>微调</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>消费级 GPU 微调</li> <li>资源受限环境</li> <li>快速模型迭代</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> peft <span style="color:#f92672">import</span> LoraConfig, get_peft_model </span></span><span style="display:flex;"><span>config <span style="color:#f92672">=</span> LoraConfig(r<span style="color:#f92672">=</span><span style="color:#ae81ff">8</span>, lora_alpha<span style="color:#f92672">=</span><span style="color:#ae81ff">32</span>, target_modules<span style="color:#f92672">=</span>[<span style="color:#e6db74">&#34;q_proj&#34;</span>, <span style="color:#e6db74">&#34;v_proj&#34;</span>]) </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> get_peft_model(base_model, config) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora/">LoRA</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8F%82%E6%95%B0%E9%AB%98%E6%95%88%E5%BE%AE%E8%B0%83-peft/">参数高效微调 (PEFT)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%87%8F%E5%8C%96-quantization/">量化 (Quantization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%8F%82%E6%95%B0%E9%AB%98%E6%95%88%E5%BE%AE%E8%B0%83-parameter-efficient-fine-tuning/">参数高效微调 (Parameter-Efficient Fine-Tuning)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/quantization/Sat, 18 Jul 2026 11:01:29 +0000https://terms-en.ai-term-hub.com/zh/terms/quantization/<h2 id="definition">Definition</h2> <p>量化将高精度浮点数（如 FP32）转换为低精度格式（如 INT8 或 FP16）。这种转换减少了模型的内存使用和计算需求，从而加速推理过程并降低硬件要求。</p> <h3 id="summary">Summary</h3> <p>一种模型优化技术，通过降低神经网络计算中数字的精度来减小模型体积并提高速度。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>精度降低</li> <li>推理速度</li> <li>内存优化</li> <li>INT8/FP16</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>边缘设备部署</li> <li>移动 AI 应用</li> <li>实时推理</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.quantization <span style="color:#66d9ef">as</span> quant </span></span><span style="display:flex;"><span><span style="color:#75715e"># Example of converting a model to quantized format</span> </span></span><span style="display:flex;"><span>model<span style="color:#f92672">.</span>eval() </span></span><span style="display:flex;"><span>model<span style="color:#f92672">.</span>qconfig <span style="color:#f92672">=</span> quant<span style="color:#f92672">.</span>get_default_qconfig(<span style="color:#e6db74">&#39;fbgemm&#39;</span>) </span></span><span style="display:flex;"><span>quantized_model <span style="color:#f92672">=</span> quant<span style="color:#f92672">.</span>prepare(model, inplace<span style="color:#f92672">=</span><span style="color:#66d9ef">False</span>) </span></span><span style="display:flex;"><span>quantized_model <span style="color:#f92672">=</span> quant<span style="color:#f92672">.</span>convert(quantized_model, inplace<span style="color:#f92672">=</span><span style="color:#66d9ef">False</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%89%AA%E6%9E%9D-pruning/">剪枝 (Pruning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%9F%A5%E8%AF%86%E8%92%B8%E9%A6%8F-knowledge-distillation/">知识蒸馏 (Knowledge Distillation)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%B7%B7%E5%90%88%E7%B2%BE%E5%BA%A6%E8%AE%AD%E7%BB%83-mixed-precision-training/">混合精度训练 (Mixed Precision Training)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/onnx/">ONNX</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/learning_rate/Sat, 18 Jul 2026 11:00:35 +0000https://terms-en.ai-term-hub.com/zh/terms/learning_rate/<h2 id="definition">Definition</h2> <p>学习率决定了在每次训练迭代中，模型权重相对于计算出的梯度更新了多少。如果学习率过高，可能导致模型在优化过程中越过最优解；如果过低，则可能导致训练收敛缓慢。</p> <h3 id="summary">Summary</h3> <p>控制模型优化过程中步长的超参数，旨在最小化损失函数。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度下降</li> <li>超参数调优</li> <li>收敛</li> <li>优化器</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>神经网络训练</li> <li>深度学习模型优化</li> <li>强化学习策略更新</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.optim <span style="color:#66d9ef">as</span> optim </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> MyModel() </span></span><span style="display:flex;"><span>optimizer <span style="color:#f92672">=</span> optim<span style="color:#f92672">.</span>SGD(model<span style="color:#f92672">.</span>parameters(), lr<span style="color:#f92672">=</span><span style="color:#ae81ff">0.01</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient_descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">gradient_descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimizer-%E4%BC%98%E5%8C%96%E5%99%A8/">optimizer (优化器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter-%E8%B6%85%E5%8F%82%E6%95%B0/">hyperparameter (超参数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/convergence-%E6%94%B6%E6%95%9B/">convergence (收敛)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/gradient_descent/Sat, 18 Jul 2026 11:00:15 +0000https://terms-en.ai-term-hub.com/zh/terms/gradient_descent/<h2 id="definition">Definition</h2> <p>梯度下降是一种用于寻找可微函数局部最小值的一阶迭代优化算法。在机器学习中，它沿梯度的反方向更新模型权重，以逐步降低损失。</p> <h3 id="summary">Summary</h3> <p>一种迭代优化算法，通过调整模型参数来最小化损失函数。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>损失函数</li> <li>学习率</li> <li>优化</li> <li>反向传播</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练深度神经网络</li> <li>线性回归参数调优</li> <li>图像识别模型优化</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/backpropagation-%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD/">Backpropagation (反向传播)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/learning-rate-%E5%AD%A6%E4%B9%A0%E7%8E%87/">Learning Rate (学习率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimizer-%E4%BC%98%E5%8C%96%E5%99%A8/">Optimizer (优化器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/loss-function-%E6%8D%9F%E5%A4%B1%E5%87%BD%E6%95%B0/">Loss Function (损失函数)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/distributed_training/Sat, 18 Jul 2026 10:59:40 +0000https://terms-en.ai-term-hub.com/zh/terms/distributed_training/<h2 id="definition">Definition</h2> <p>分布式训练通过在多个 GPU 或节点上并行化计算来加速模型收敛。主要技术包括数据并行（每个工作进程处理数据子集）和模型并行（将模型的不同部分分布在不同设备上），以处理大规模数据集和巨型模型。</p> <h3 id="summary">Summary</h3> <p>通过将数据或计算任务拆分到多个设备或服务器上，从而训练机器学习模型的方法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>数据并行</li> <li>模型并行</li> <li>GPU 集群</li> <li>梯度同步</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练大型语言模型</li> <li>加速计算机视觉数据集的处理</li> <li>减少复杂神经网络的训练时间</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/parallel-computing-%E5%B9%B6%E8%A1%8C%E8%AE%A1%E7%AE%97/">Parallel Computing (并行计算)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gpu-%E5%9B%BE%E5%BD%A2%E5%A4%84%E7%90%86%E5%99%A8/">GPU (图形处理器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/horovod-%E5%88%86%E5%B8%83%E5%BC%8F%E8%AE%AD%E7%BB%83%E6%A1%86%E6%9E%B6/">Horovod (分布式训练框架)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pytorch-ddp-pytorch-%E5%88%86%E5%B8%83%E5%BC%8F%E6%95%B0%E6%8D%AE%E5%B9%B6%E8%A1%8C/">PyTorch DDP (PyTorch 分布式数据并行)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/adapter/Sat, 18 Jul 2026 10:59:15 +0000https://terms-en.ai-term-hub.com/zh/terms/adapter/<h2 id="definition">Definition</h2> <p>适配器是一种参数高效的微调技术，主要用于大型语言模型和Transformer架构中。与计算成本高昂的更新所有模型权重不同，适配器仅训练少量新增参数，从而保留预训练知识并降低资源消耗。</p> <h3 id="summary">Summary</h3> <p>插入预训练模型中的轻量级模块，用于针对特定下游任务进行高效微调。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>参数高效微调</li> <li>迁移学习</li> <li>模块化架构</li> <li>灾难性遗忘</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>为客服聊天机器人微调大型语言模型</li> <li>适配视觉模型以进行医学图像分析</li> <li>高效部署多个领域特定的模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora-%E4%BD%8E%E7%A7%A9%E9%80%82%E5%BA%94/">LoRA (低秩适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/prompt-tuning-%E6%8F%90%E7%A4%BA%E5%BE%AE%E8%B0%83/">Prompt Tuning (提示微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fine-tuning-%E5%BE%AE%E8%B0%83/">Fine-Tuning (微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transformer-transformer%E6%9E%B6%E6%9E%84/">Transformer (Transformer架构)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/task_specific/Sat, 18 Jul 2026 10:57:22 +0000https://terms-en.ai-term-hub.com/zh/terms/task_specific/<h2 id="definition">Definition</h2> <p>任务特定指的是专为在狭窄的目标集合上表现出色而定制的AI模型或组件，例如检测图像中的物体或翻译语言。与通用基础模型不同，它们在特定领域往往具有更高的效率和精度。</p> <h3 id="summary">Summary</h3> <p>任务特定模型是为执行单一、明确定义的功能而设计和优化的AI系统，而非用于通用推理。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>专业化</li> <li>效率</li> <li>微调</li> <li>领域适配</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>垃圾邮件过滤</li> <li>人脸识别系统</li> <li>医学图像分割</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/general_purpose-%E9%80%9A%E7%94%A8%E7%9B%AE%E7%9A%84/">general_purpose (通用目的)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fine_tuning-%E5%BE%AE%E8%B0%83/">fine_tuning (微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/specialized_model-%E4%B8%93%E7%94%A8%E6%A8%A1%E5%9E%8B/">specialized_model (专用模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/domain_adaptation-%E9%A2%86%E5%9F%9F%E9%80%82%E9%85%8D/">domain_adaptation (领域适配)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/one_step/Sat, 18 Jul 2026 10:57:10 +0000https://terms-en.ai-term-hub.com/zh/terms/one_step/<h2 id="definition">Definition</h2> <p>在机器学习和优化领域，一步法直接解决问题，不需要多次迭代或更新即可收敛。与需要多步才能最小化</p> <h3 id="summary">Summary</h3> <p>指在单次迭代中完成任务或决策过程，无需进行迭代优化的算法或流程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>闭式解</li> <li>计算效率</li> <li>非迭代性</li> <li>直接映射</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>通过正规方程进行的线性回归</li> <li>实时信号处理</li> <li>简单的分类规则</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/iterative-%E8%BF%AD%E4%BB%A3%E7%9A%84/">iterative (迭代的)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">gradient descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/closed-form-%E9%97%AD%E5%BC%8F%E7%9A%84/">closed-form (闭式的)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/direct-method-%E7%9B%B4%E6%8E%A5%E6%B3%95/">direct method (直接法)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/first_order/Sat, 18 Jul 2026 10:56:36 +0000https://terms-en.ai-term-hub.com/zh/terms/first_order/<h2 id="definition">Definition</h2> <p>在人工智能和数学中，“一阶”通常描述不涉及高阶交互的直接、线性关系系统或操作。在优化领域，它指的是仅利用目标函数的一阶导数（梯度）信息来寻找极值的方法，例如梯度下降法。与二阶方法相比，一阶方法计算成本较低，但收敛速度可能较慢。</p> <h3 id="summary">Summary</h3> <p>指涉及直接关系或线性近似概念，如一阶逻辑或一阶导数。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度下降</li> <li>一阶逻辑</li> <li>线性近似</li> <li>导数</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>通过反向传播训练神经网络</li> <li>符号AI中的知识表示</li> <li>如随机梯度下降（SGD）等优化算法</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/second-order-%E4%BA%8C%E9%98%B6/">second-order (二阶)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-%E6%A2%AF%E5%BA%A6/">gradient (梯度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/logic-%E9%80%BB%E8%BE%91/">logic (逻辑)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimization-%E4%BC%98%E5%8C%96/">optimization (优化)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/fine_tuned/Sat, 18 Jul 2026 10:56:23 +0000https://terms-en.ai-term-hub.com/zh/terms/fine_tuned/<h2 id="definition">Definition</h2> <p>微调涉及使用较小且针对特定任务的数据集，继续训练已在大型通用数据集上训练好的模型。该技术利用了预训练模型中习得的通用特征表示，使其能够以较低的计算成本和较少的数据量，高效地适配到具体的应用场景中。</p> <h3 id="summary">Summary</h3> <p>在特定数据集上进一步训练预训练模型，使其适应特定下游任务的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>迁移学习</li> <li>权重更新</li> <li>任务特定</li> <li>预训练模型</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>调整大语言模型以审阅法律文档</li> <li>定制视觉模型用于工业缺陷检测</li> <li>专门化语音识别以适应特定口音</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pre_training-%E9%A2%84%E8%AE%AD%E7%BB%83/">pre_training (预训练)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/transfer_learning-%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0/">transfer_learning (迁移学习)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/supervised_fine_tuning-%E7%9B%91%E7%9D%A3%E5%BE%AE%E8%B0%83/">supervised_fine_tuning (监督微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/parameter_efficient-%E5%8F%82%E6%95%B0%E9%AB%98%E6%95%88/">parameter_efficient (参数高效)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/tuning/Sat, 18 Jul 2026 10:55:40 +0000https://terms-en.ai-term-hub.com/zh/terms/tuning/<h2 id="definition">Definition</h2> <p>调优涉及改进机器学习模型以实现更好的准确性或效率。它可以指超参数调优，即优化学习率或批次大小等设置；也可以指模型权重的微调。</p> <h3 id="summary">Summary</h3> <p>调整超参数或模型权重以优化特定数据集或任务性能的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>超参数</li> <li>网格搜索</li> <li>随机搜索</li> <li>防止过拟合</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>优化模型准确性</li> <li>降低推理延迟</li> <li>使模型适应特定领域</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> sklearn.model_selection <span style="color:#f92672">import</span> GridSearchCV </span></span><span style="display:flex;"><span>param_grid <span style="color:#f92672">=</span> {<span style="color:#e6db74">&#39;C&#39;</span>: [<span style="color:#ae81ff">0.1</span>, <span style="color:#ae81ff">1</span>, <span style="color:#ae81ff">10</span>]} </span></span><span style="display:flex;"><span>search <span style="color:#f92672">=</span> GridSearchCV(svm, param_grid, cv<span style="color:#f92672">=</span><span style="color:#ae81ff">5</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/hyperparameter_optimization-%E8%B6%85%E5%8F%82%E6%95%B0%E4%BC%98%E5%8C%96/">hyperparameter_optimization (超参数优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/grid_search-%E7%BD%91%E6%A0%BC%E6%90%9C%E7%B4%A2/">grid_search (网格搜索)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fine_tuning-%E5%BE%AE%E8%B0%83/">fine_tuning (微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/validation-%E9%AA%8C%E8%AF%81/">validation (验证)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/transfer_learning/Sat, 18 Jul 2026 10:55:40 +0000https://terms-en.ai-term-hub.com/zh/terms/transfer_learning/<h2 id="definition">Definition</h2> <p>迁移学习利用预训练模型来提高在新且相关任务上的性能并减少训练时间。开发人员无需从头开始训练，而是对现有权重进行微调，从而利用先前学到的知识。</p> <h3 id="summary">Summary</h3> <p>一种机器学习技术，将在一个任务中开发的模型作为第二个任务的模型的起点进行复用。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>预训练模型</li> <li>微调</li> <li>领域自适应</li> <li>特征提取</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>数据有限的图像分类</li> <li>针对小众主题的情感分析</li> <li>医疗诊断辅助</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> transformers <span style="color:#f92672">import</span> AutoModelForSequenceClassification </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> AutoModelForSequenceClassification<span style="color:#f92672">.</span>from_pretrained(<span style="color:#e6db74">&#39;bert-base-uncased&#39;</span>, num_labels<span style="color:#f92672">=</span><span style="color:#ae81ff">2</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fine_tuning-%E5%BE%AE%E8%B0%83/">fine_tuning (微调)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pre_training-%E9%A2%84%E8%AE%AD%E7%BB%83/">pre_training (预训练)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/domain_adaptation-%E9%A2%86%E5%9F%9F%E8%87%AA%E9%80%82%E5%BA%94/">domain_adaptation (领域自适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/few_shot_learning-%E5%B0%91%E6%A0%B7%E6%9C%AC%E5%AD%A6%E4%B9%A0/">few_shot_learning (少样本学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/towards/Sat, 18 Jul 2026 10:55:25 +0000https://terms-en.ai-term-hub.com/zh/terms/towards/<h2 id="definition">Definition</h2> <p>在AI开发中，“趋向”通常描述优化过程的轨迹，例如梯度下降使权重向最小损失值移动。它还表示研究方向，即研究重点逐渐集中在提高效率或对齐人类价值观上。</p> <h3 id="summary">Summary</h3> <p>“趋向”指AI中的方向性进展或优化目标，表示向性能提升或特定目标移动的方向。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>优化方向</li> <li>梯度下降</li> <li>目标对齐</li> <li>迭代改进</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>监控训练期间的损失减少</li> <li>使AI输出与人类价值观对齐</li> <li>将研究重点导向效率提升</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimization-%E4%BC%98%E5%8C%96/">Optimization (优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-%E6%A2%AF%E5%BA%A6/">Gradient (梯度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/alignment-%E5%AF%B9%E9%BD%90/">Alignment (对齐)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/objective-%E7%9B%AE%E6%A0%87/">Objective (目标)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/scaling/Sat, 18 Jul 2026 10:54:38 +0000https://terms-en.ai-term-hub.com/zh/terms/scaling/<h2 id="definition">Definition</h2> <p>扩展是扩展AI系统的主动方法论，通过增加更多层、神经元或训练样本来实现。它包括分布式训练等技术，以处理增加的计算需求</p> <h3 id="summary">Summary</h3> <p>扩展是通过调整模型大小或数据量来增强学习能力和性能的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>分布式训练</li> <li>模型复杂度</li> <li>过拟合</li> <li>硬件加速</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>分布式深度学习</li> <li>增加神经网络深度</li> <li>扩大训练语料库</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/scale-%E8%A7%84%E6%A8%A1/">Scale (规模)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/distributed-computing-%E5%88%86%E5%B8%83%E5%BC%8F%E8%AE%A1%E7%AE%97/">Distributed Computing (分布式计算)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/model-architecture-%E6%A8%A1%E5%9E%8B%E6%9E%B6%E6%9E%84/">Model Architecture (模型架构)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/performance-%E6%80%A7%E8%83%BD/">Performance (性能)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/rate/Sat, 18 Jul 2026 10:54:02 +0000https://terms-en.ai-term-hub.com/zh/terms/rate/<h2 id="definition">Definition</h2> <p>在AI中，“率”最常指学习率，这是一个超参数，控制每次更新模型权重时，根据估计误差对模型进行调整的幅度。一个适……</p> <h3 id="summary">Summary</h3> <p>频率或速度的度量，通常指优化中的学习率或令牌生成速度。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>学习率</li> <li>优化</li> <li>吞吐量</li> <li>超参数</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>调整梯度下降优化</li> <li>监控API使用限制</li> <li>测量推理延迟</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span>optimizer <span style="color:#f92672">=</span> torch<span style="color:#f92672">.</span>optim<span style="color:#f92672">.</span>SGD(model<span style="color:#f92672">.</span>parameters(), lr<span style="color:#f92672">=</span><span style="color:#ae81ff">0.01</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimizer-%E4%BC%98%E5%8C%96%E5%99%A8/">Optimizer (优化器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/convergence-%E6%94%B6%E6%95%9B/">Convergence (收敛)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/speed-%E9%80%9F%E5%BA%A6/">Speed (速度)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/latency-%E5%BB%B6%E8%BF%9F/">Latency (延迟)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/optimal/Sat, 18 Jul 2026 10:53:38 +0000https://terms-en.ai-term-hub.com/zh/terms/optimal/<h2 id="definition">Definition</h2> <p>在人工智能和优化理论中，最优解是指达到最高可能性能指标的解决方案，例如强化学习中的最大奖励或回归分析中的最小误差。寻找最优解通常涉及在复杂的目标函数空间中进行搜索。</p> <h3 id="summary">Summary</h3> <p>最优是指在给定约束条件下可能的最佳解决方案或行动，旨在最大化奖励或最小化成本。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>全局与局部最优</li> <li>目标函数</li> <li>约束满足</li> <li>效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>路径查找算法（如 A*）</li> <li>超参数调优</li> <li>云计算中的资源分配</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/heuristic-%E5%90%AF%E5%8F%91%E5%BC%8F/">Heuristic (启发式)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">Gradient Descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/loss-function-%E6%8D%9F%E5%A4%B1%E5%87%BD%E6%95%B0/">Loss Function (损失函数)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/pareto-frontier-%E5%B8%95%E7%B4%AF%E6%89%98%E5%89%8D%E6%B2%BF/">Pareto Frontier (帕累托前沿)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/loss/Sat, 18 Jul 2026 10:52:50 +0000https://terms-en.ai-term-hub.com/zh/terms/loss/<h2 id="definition">Definition</h2> <p>损失函数（也称为成本函数）衡量机器学习模型的预测结果与真实标签在训练期间的匹配程度。优化算法的目标是最小化这个损失值，从而提高模型的准确性。</p> <h3 id="summary">Summary</h3> <p>一个数值，用于量化模型预测值与实际目标值之间的误差。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>成本函数</li> <li>优化</li> <li>梯度下降</li> <li>误差指标</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>训练图像分类器</li> <li>优化回归模型</li> <li>评估模型收敛性</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.nn <span style="color:#66d9ef">as</span> nn </span></span><span style="display:flex;"><span>criterion <span style="color:#f92672">=</span> nn<span style="color:#f92672">.</span>CrossEntropyLoss() </span></span><span style="display:flex;"><span>loss <span style="color:#f92672">=</span> criterion(outputs, targets) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/accuracy-%E5%87%86%E7%A1%AE%E7%8E%87/">Accuracy (准确率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/gradient-descent-%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">Gradient Descent (梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/cross-entropy-%E4%BA%A4%E5%8F%89%E7%86%B5/">Cross-Entropy (交叉熵)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/overfitting-%E8%BF%87%E6%8B%9F%E5%90%88/">Overfitting (过拟合)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/lora/Sat, 18 Jul 2026 10:52:16 +0000https://terms-en.ai-term-hub.com/zh/terms/lora/<h2 id="definition">Definition</h2> <p>LoRA冻结预训练模型的权重，并在Transformer架构的每一层中插入可训练的分解矩阵。通过仅优化这些低秩矩阵，LoRA显著减少</p> <h3 id="summary">Summary</h3> <p>低秩自适应是一种参数高效的微调方法，它将可训练的秩分解矩阵注入现有模型权重中。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>参数高效微调</li> <li>秩分解</li> <li>权重冻结</li> <li>适配器模块</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>定制大语言模型</li> <li>领域特定适配</li> <li>资源受限的训练环境</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> peft <span style="color:#f92672">import</span> LoraConfig, get_peft_model </span></span><span style="display:flex;"><span>config <span style="color:#f92672">=</span> LoraConfig(r<span style="color:#f92672">=</span><span style="color:#ae81ff">8</span>, lora_alpha<span style="color:#f92672">=</span><span style="color:#ae81ff">32</span>) </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> get_peft_model(base_model, config) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/peft-%E5%8F%82%E6%95%B0%E9%AB%98%E6%95%88%E5%BE%AE%E8%B0%83%E6%8A%80%E6%9C%AF/">PEFT (参数高效微调技术)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%BE%AE%E8%B0%83-fine-tuning/">微调 (Fine-Tuning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%87%8F%E5%8C%96-quantization/">量化 (Quantization)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/global/Sat, 18 Jul 2026 10:51:41 +0000https://terms-en.ai-term-hub.com/zh/terms/global/<h2 id="definition">Definition</h2> <p>在人工智能中，“全局”一词通常与“局部”相对，指涵盖整个系统的方面。在优化领域，全局最小值代表在整个损失景观中可能的最佳解决方案。</p> <h3 id="summary">Summary</h3> <p>描述适用于整个模型或数据集而非局部子集的属性、优化或范围。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>全局最小值</li> <li>完整上下文</li> <li>系统级</li> <li>聚合</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>损失景观分析</li> <li>全局注意力机制</li> <li>批归一化统计量</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/local-%E5%B1%80%E9%83%A8/">Local (局部)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimization-%E4%BC%98%E5%8C%96/">Optimization (优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/attention-%E6%B3%A8%E6%84%8F%E5%8A%9B/">Attention (注意力)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/convergence-%E6%94%B6%E6%95%9B/">Convergence (收敛)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/fast/Sat, 18 Jul 2026 10:51:19 +0000https://terms-en.ai-term-hub.com/zh/terms/fast/<h2 id="definition">Definition</h2> <p>“快速”一词描述了人工智能模型中的计算效率，强调快速的推理时间和高效的数据处理能力。这对于实时应用至关重要。</p> <h3 id="summary">Summary</h3> <p>在人工智能领域，“快速”指为低延迟和高吞吐量处理任务而优化的系统或算法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>延迟</li> <li>吞吐量</li> <li>实时处理</li> <li>优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自动驾驶导航</li> <li>直播视频流分析</li> <li>高频交易算法</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/efficiency-%E6%95%88%E7%8E%87/">Efficiency (效率)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/latency-%E5%BB%B6%E8%BF%9F/">Latency (延迟)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/performance-%E6%80%A7%E8%83%BD/">Performance (性能)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/efficient/Sat, 18 Jul 2026 10:50:55 +0000https://terms-en.ai-term-hub.com/zh/terms/efficient/<h2 id="definition">Definition</h2> <p>效率是人工智能中的一项关键指标，用于衡量模型或算法利用可用资源的程度。它包括计算效率（推理和训练的速度）以及资源利用率。高效的AI系统能够在保持高精度的同时，显著降低硬件需求和能耗，这对于大规模部署和实时应用至关重要。</p> <h3 id="summary">Summary</h3> <p>在人工智能领域，效率是指以最少的时间、内存或计算资源消耗来实现最佳性能。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>计算复杂度</li> <li>资源优化</li> <li>延迟降低</li> <li>模型压缩</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>为移动端部署优化大型语言模型</li> <li>降低深度学习网络的训练成本</li> <li>提高自主系统中的实时推理速度</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimization-%E4%BC%98%E5%8C%96/">Optimization (优化)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/scalability-%E5%8F%AF%E6%89%A9%E5%B1%95%E6%80%A7/">Scalability (可扩展性)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/latency-%E5%BB%B6%E8%BF%9F/">Latency (延迟)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/throughput-%E5%90%9E%E5%90%90%E9%87%8F/">Throughput (吞吐量)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/divergence/Sat, 18 Jul 2026 10:50:42 +0000https://terms-en.ai-term-hub.com/zh/terms/divergence/<h2 id="definition">Definition</h2> <p>在优化的背景下，当模型参数的更新方式导致损失增加而不是减少时，就会发生发散，这通常会导致NaN值或无限梯度。这通常是由于学习率过高或数值不稳定性引起的。</p> <h3 id="summary">Summary</h3> <p>发散是指机器学习算法的损失函数在训练过程中未能下降，导致性能不稳定或恶化的现象。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>损失爆炸</li> <li>学习率敏感性</li> <li>梯度不稳定</li> <li>数值精度</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>调试深度学习框架中的训练循环</li> <li>调整超参数以实现稳定收敛</li> <li>实施梯度裁剪策略以防止发散</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A2%AF%E5%BA%A6%E6%B6%88%E5%A4%B1-vanishing-gradient/">梯度消失 (Vanishing Gradient)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%A2%AF%E5%BA%A6%E7%88%86%E7%82%B8-exploding-gradient/">梯度爆炸 (Exploding Gradient)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%94%B6%E6%95%9B-convergence/">收敛 (Convergence)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%A8%B3%E5%AE%9A%E6%80%A7-stability/">稳定性 (Stability)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/distillation/Sat, 18 Jul 2026 10:50:42 +0000https://terms-en.ai-term-hub.com/zh/terms/distillation/<h2 id="definition">Definition</h2> <p>这一过程涉及将知识从复杂的高性能“教师”神经网络转移到更简单、高效的“学生”网络中。学生不仅从硬标签中学习，还从软标签（即教师模型的输出概率分布）中学习，从而捕捉更丰富的信息。</p> <h3 id="summary">Summary</h3> <p>知识蒸馏是一种模型压缩技术，其中较小的学生模型通过学习模仿较大的教师模型的行为来工作。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>师生架构</li> <li>软目标</li> <li>模型压缩</li> <li>推理效率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>在移动设备上部署大型语言模型</li> <li>降低实时计算机视觉应用中的延迟</li> <li>优化边缘计算环境中的深度学习模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%87%8F%E5%8C%96-quantization/">量化 (Quantization)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E5%89%AA%E6%9E%9D-pruning/">剪枝 (Pruning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0-transfer-learning/">迁移学习 (Transfer Learning)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%A5%9E%E7%BB%8F%E6%9E%B6%E6%9E%84%E6%90%9C%E7%B4%A2-neural-architecture-search/">神经架构搜索 (Neural Architecture Search)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/combining/Sat, 18 Jul 2026 10:49:42 +0000https://terms-en.ai-term-hub.com/zh/terms/combining/<h2 id="definition">Definition</h2> <p>这一概念包括集成学习方法，其中聚合多个模型的预测结果以减少方差或偏差。它还涵盖多模态融合，即结合不同类型的数据（如文本和图像）以提供更丰富的上下文信息。</p> <h3 id="summary">Summary</h3> <p>AI中的组合指整合多个模型、数据源或技术，以提高整体性能和鲁棒性。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>集成学习</li> <li>模型融合</li> <li>多模态集成</li> <li>聚合</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>用于分类的随机森林</li> <li>多模态情感分析</li> <li>用于回归任务的堆叠模型</li> </ul> <h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/ensemble-%E9%9B%86%E6%88%90/">Ensemble (集成)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fusion-%E8%9E%8D%E5%90%88/">Fusion (融合)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/integration-%E9%9B%86%E6%88%90/">Integration (集成)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/meta-learning-%E5%85%83%E5%AD%A6%E4%B9%A0/">Meta-learning (元学习)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/adam/Sat, 18 Jul 2026 10:49:06 +0000https://terms-en.ai-term-hub.com/zh/terms/adam/<h2 id="definition">Definition</h2> <p>Adam（自适应矩估计）是一种流行的基于一阶梯度的优化算法，用于训练深度神经网络。它结合了两种其他随机梯度下降扩展的优势。</p> <h3 id="summary">Summary</h3> <p>一种为每个参数计算自适应学习率的优化算法。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>梯度下降</li> <li>学习率</li> <li>动量</li> <li>方差估计</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>深度学习训练</li> <li>计算机视觉模型</li> <li>自然语言处理</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">import</span> torch.optim <span style="color:#66d9ef">as</span> optim </span></span><span style="display:flex;"><span>optimizer <span style="color:#f92672">=</span> optim<span style="color:#f92672">.</span>Adam(model<span style="color:#f92672">.</span>parameters(), lr<span style="color:#f92672">=</span><span style="color:#ae81ff">0.001</span>) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/sgd-%E9%9A%8F%E6%9C%BA%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D/">SGD (随机梯度下降)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/rmsprop-%E5%9D%87%E6%96%B9%E6%A0%B9%E4%BC%A0%E6%92%AD/">RMSProp (均方根传播)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/optimizer-%E4%BC%98%E5%8C%96%E5%99%A8/">Optimizer (优化器)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/backpropagation-%E5%8F%8D%E5%90%91%E4%BC%A0%E6%92%AD/">Backpropagation (反向传播)</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/fine_tuning/Sat, 18 Jul 2026 07:44:46 +0000https://terms-en.ai-term-hub.com/zh/terms/fine_tuning/<h2 id="definition">Definition</h2> <p>微调涉及在一个已在大而通用数据集上训练好的模型基础上，继续在专业化数据集上进行训练。这使得模型在保留通用知识的同时，能够习得特定任务的技能。</p> <h3 id="summary">Summary</h3> <p>使用较小的数据集将预训练模型适配到特定下游任务的过程。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>迁移学习</li> <li>预训练模型</li> <li>任务特定适配</li> <li>学习率</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>适配大语言模型以构建客户服务聊天机器人</li> <li>专门化图像分类器用于医疗诊断</li> <li>定制语音识别以针对特定口音</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span><span style="color:#f92672">from</span> transformers <span style="color:#f92672">import</span> AutoModelForSequenceClassification </span></span><span style="display:flex;"><span>model <span style="color:#f92672">=</span> AutoModelForSequenceClassification<span style="color:#f92672">.</span>from_pretrained(<span style="color:#e6db74">&#39;bert-base-uncased&#39;</span>) </span></span><span style="display:flex;"><span><span style="color:#75715e"># Freeze base layers</span> </span></span><span style="display:flex;"><span><span style="color:#66d9ef">for</span> param <span style="color:#f92672">in</span> model<span style="color:#f92672">.</span>bert<span style="color:#f92672">.</span>parameters(): </span></span><span style="display:flex;"><span> param<span style="color:#f92672">.</span>requires_grad <span style="color:#f92672">=</span> <span style="color:#66d9ef">False</span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># Train only classification head</span> </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E9%A2%84%E8%AE%AD%E7%BB%83/">预训练</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E6%8F%90%E7%A4%BA%E5%B7%A5%E7%A8%8B/">提示工程</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/lora-%E4%BD%8E%E7%A7%A9%E9%80%82%E5%BA%94/">LoRA (低秩适应)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/%E7%9B%91%E7%9D%A3%E5%AD%A6%E4%B9%A0/">监督学习</a></li> </ul>https://terms-en.ai-term-hub.com/zh/terms/prompt_engineering/Sat, 18 Jul 2026 07:43:58 +0000https://terms-en.ai-term-hub.com/zh/terms/prompt_engineering/<h2 id="definition">Definition</h2> <p>提示工程涉及制作特定的输入（即提示），以从生成式 AI 模型中获取准确、相关且高质量的响应。它要求理解模型如何解释指令、上下文及约束条件，从而通过调整提示词的结构和措辞来最大化模型的表现。</p> <h3 id="summary">Summary</h3> <p>设计和优化输入文本，以引导大型语言模型生成预期输出的实践。</p> <h2 id="key-concepts">Key Concepts</h2> <ul> <li>语境框架</li> <li>少样本学习</li> <li>指令微调</li> <li>令牌优化</li> </ul> <h2 id="use-cases">Use Cases</h2> <ul> <li>自动化客户支持聊天机器人</li> <li>代码生成助手</li> <li>创意写作辅助工具</li> </ul> <h2 id="code-example">Code Example</h2> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;"><code class="language-python" data-lang="python"><span style="display:flex;"><span>prompt <span style="color:#f92672">=</span> <span style="color:#e6db74">&#34;Translate the following English text to French: &#39;Hello world&#39;&#34;</span> </span></span><span style="display:flex;"><span>response <span style="color:#f92672">=</span> llm<span style="color:#f92672">.</span>generate(prompt) </span></span></code></pre></div><h2 id="related-terms">Related Terms</h2> <ul> <li><a href="https://terms-en.ai-term-hub.com/en/terms/llm-%E5%A4%A7%E5%9E%8B%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B/">LLM (大型语言模型)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/chain-of-thought-%E6%80%9D%E7%BB%B4%E9%93%BE/">Chain-of-Thought (思维链)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/rag-%E6%A3%80%E7%B4%A2%E5%A2%9E%E5%BC%BA%E7%94%9F%E6%88%90/">RAG (检索增强生成)</a></li> <li><a href="https://terms-en.ai-term-hub.com/en/terms/fine-tuning-%E5%BE%AE%E8%B0%83/">Fine-tuning (微调)</a></li> </ul>