Holistic Reinforcement Learning: The Role of Structure and Attention

Area of Science:

• Cognitive Science
• Neuroscience
• Decision Science

Background:

• Reinforcement learning models explain behavior but not environmental structure's influence.
• Bayesian cognitive models explain structured knowledge but lack neurobiological basis.

Purpose of the Study:

• Integrate Bayesian and reinforcement learning models.
• Explain how structured knowledge influences learning and decision-making.
• Propose a neurobiologically plausible model of structure learning.

Main Methods:

• Approximate Bayesian inference for structure learning.
• Selective attention as a mediator between structure learning and reinforcement learning.
• Computational modeling of human decision-making.

Main Results:

• Structured knowledge learned via Bayesian inference biases attention.
• Attention selectively guides reinforcement learning towards relevant environmental dimensions.
• The integrated model accounts for structure-dependent learning effects.

Conclusions:

• Integrating Bayesian inference and reinforcement learning provides a unified framework for decision science.
• Understanding structure learning is key to explaining human decision-making.
• This approach offers a neurobiologically plausible account of learning and attention.