Simple artificial neural networks that match probability and exploit and explore when confronting a multiarmed bandit

Area of Science:

• Cognitive Science
• Neuroscience
• Machine Learning

Background:

• The matching law posits response rates are proportional to reinforcement rates.
• Probability matching is an empirical observation related to the matching law.
• Artificial neural networks offer a model for understanding learning and choice behavior.

Purpose of the Study:

• To demonstrate that a simple artificial neural network can exhibit probability matching.
• To develop an operant learning procedure for artificial neural networks.
• To explore the application of perceptrons in unifying diverse learning theories.

Main Methods:

• Simulated a simple artificial neural network.
• Implemented an operant procedure using the multiarmed bandit problem.
• Utilized perceptrons as a framework for analysis.

Main Results:

• The artificial neural network's responses were consistent with probability matching.
• Operant training successfully balanced exploitation and exploration in the network.
• Perceptrons provided a unifying framework for various learning paradigms.

Conclusions:

• Artificial neural networks can replicate complex animal learning phenomena like probability matching.
• Operant training is a viable method for network learning and optimizing choice behavior.
• Perceptrons serve as a valuable tool for integrating insights from machine learning, animal behavior, and reinforcement learning theories.