The following are the parts of of a learning agent:
Perceptor: This block receives sensory information from the environment. This information can be visual data, sounds, readings from sensors, etc.
Performance Element: This block decides what action the agent should take in the environment based on the current percept and the agent's internal state.
Critic: This block evaluates the performance of the agent's actions. It compares the achieved outcome with the desired goal and provides feedback to the learning element. This feedback can be a reward signal (positive for good actions, negative for bad actions) or some other form of performance evaluation.
Learning Element: This block is responsible for using the critic's feedback to improve the agent's performance over time. It can update the agent's knowledge, modify the performance element's strategy, or adjust internal parameters based on the learning algorithm used.
This is how a learning agent works:
The agent perceives the environment through the perceptor, gathering information about its current state.
Based on this perception and its internal state (which could include past experiences and knowledge), the performance element selects an action to take in the environment.
The agent takes the chosen action and observes the outcome.
The critic evaluates the outcome based on a pre-defined performance goal. It provides feedback to the learning element in the form of a reward signal or other performance measure.
The learning element utilizes this feedback to improve the agent's future performance. This could involve:
Updating the agent's knowledge about the environment and its dynamics.
Modifying the strategy used by the performance element to select actions.
Adjusting internal parameters of the learning algorithm.