Associative learning in random environments using neural networks

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Associative learning is fundamental to decision-making in dynamic environments.
• Neural networks and learning automata provide frameworks for modeling learning processes.
• Reinforcement learning is crucial for optimizing actions based on environmental feedback.

Purpose of the Study:

• To investigate associative learning using neural networks within a reinforcement learning paradigm.
• To determine optimal actions for a single decision-maker in a stochastic environment.
• To compare different neural network-based methods for generating discriminant functions.

Main Methods:

• Utilized neural networks and learning automata concepts for associative learning.
• Employed reinforcement learning to train a single decision-maker in a random environment.
• Developed and compared three distinct neural network approaches for action selection and weight updates.

Main Results:

• The study presents simulation results demonstrating the feasibility of the proposed methods.
• The most general method uses network output to determine action probabilities, with weights updated via environmental response.
• Modifications were introduced to enhance the practical viability of the general method.

Conclusions:

• All proposed neural network methods are feasible for associative learning and decision-making.
• The choice of method depends on desired accuracy and available computational resources.
• The research extends to decentralized decision-making frameworks within a context space.