Reinforcement Learning on English AI Terms Dictionary

Matchbox Educable Noughts and Crosses Engine

Sat, 18 Jul 2026 10:06:42 +0000

Definition

The ME-Noughts-and-Crosses Engine was an early demonstration of machine learning, specifically reinforcement learning. Constructed from 304 matchboxes, each representing a unique board state, the system used colored beads to represent possible moves. After playing against a human opponent, the operator would reinforce successful moves by adding beads and remove beads from losing paths. Over time, the machine learned optimal strategies through trial and error, serving as a tangible precursor to modern AI algorithms like Q-learning.

Learning automaton

Sat, 18 Jul 2026 10:04:43 +0000

Definition

This concept originates from reinforcement learning and involves an agent interacting with an unknown environment. The automaton selects actions from a finite set and receives a penalty or reward signal. Based on this feedback, it adjusts the probability distribution over its actions using a learning algorithm, gradually converging toward the optimal action that yields the highest expected reward. It serves as a foundational block for more complex multi-agent systems.

Empowerment

Sat, 18 Jul 2026 09:56:53 +0000

Definition

In reinforcement learning and artificial intelligence, empowerment is a intrinsic motivation metric that quantifies the amount of control an agent has over its environment. It is defined as the mutual information between the agent’s actions and the resulting future states. By maximizing empowerment, agents are driven to explore environments where they can effect meaningful change, leading to more robust and adaptive behaviors without relying solely on external reward signals.

Bayesian regret

Sat, 18 Jul 2026 09:48:06 +0000

Definition

Bayesian regret quantifies the difference between the optimal reward achievable with perfect information and the expected reward obtained by an agent acting under uncertainty. It is calculated by integrating the regret over all possible states of the world weighted by their prior probabilities. This concept is crucial in reinforcement learning and game theory, helping to evaluate how well an algorithm performs when it must make decisions without knowing the true underlying parameters or environment dynamics.

Apprenticeship learning

Sat, 18 Jul 2026 09:45:50 +0000

Definition

Apprenticeship learning, also known as inverse reinforcement learning from demonstrations, enables agents to acquire skills by observing expert behavior rather than relying solely on reward functions. The agent infers the underlying reward structure that explains the expert’s actions and then optimizes its own policy to match or exceed that performance. This technique is particularly useful in complex environments where defining explicit rewards is difficult or ambiguous.

Summary

A reinforcement learning method where an agent learns a policy by imitating an expert’s demonstrations.

Action model learning

Sat, 18 Jul 2026 09:44:54 +0000

Definition

Action model learning involves an agent constructing an internal representation of how its actions transition the environment from one state to another. Unlike passive observation, this method leverages the agent’s agency to gather data, allowing it to predict outcomes and plan future moves. It is crucial in environments where the underlying physics or rules are unknown, enabling the agent to build a predictive model through trial and error, thereby improving decision-making efficiency over time without requiring pre-labeled datasets.

Actor-critic algorithm

Sat, 18 Jul 2026 09:44:54 +0000

Definition

The actor-critic algorithm employs two components: the actor, which updates the policy to select actions, and the critic, which evaluates the quality of those actions by estimating the value function. The critic provides feedback to the actor, guiding policy improvements based on temporal difference errors. This hybrid approach leverages the low variance of value-based methods and the high bias but potentially lower variance of policy gradient methods, resulting in more stable and efficient learning in complex continuous control tasks.

Guided

Sat, 18 Jul 2026 09:33:06 +0000

Definition

The term ‘guided’ in AI typically refers to techniques where the model’s behavior is steered by additional information beyond the primary input. Common examples include guided diffusion, where a classifier or text prompt directs image generation, or guided policy search in reinforcement learning, where high-level plans guide low-level control actions. This approach helps mitigate issues like mode collapse or aimless exploration by providing a structured path toward the desired outcome, improving both the quality and controllability of the AI’s output.