A generalized reinforcement learning based deep neural network agent model for diverse cognitive constructs.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Human cognition involves diverse functions like attention, decision-making, and working memory, often studied in isolation.
• Cognitive impairments frequently affect multiple functions simultaneously, necessitating integrated research approaches.
• Existing models often focus on single cognitive components, limiting their application in complex conditions.

Purpose of the Study:

• To develop a unified computational model capable of simulating multiple human cognitive functions.
• To demonstrate the model's ability to represent and integrate various cognitive tasks within a single framework.
• To establish a computational testbench for simulating cognitive impairment and therapeutic interventions.

Main Methods:

• Proposed a unified agent model based on reinforcement learning principles.
• Incorporated distinct systems for representation, memory, value computation, and exploration.
• Modeled key cognitive tasks such as selective attention, distractor suppression, decision-making, response inhibition, and working memory.

Main Results:

• Successfully modeled diverse cognitive tasks using the unified reinforcement learning agent.
• Demonstrated that individual cognitive performance can be mapped to specific model meta-parameters.
• Validated the model's potential as a proxy for studying cognitive impairments.

Conclusions:

• The unified reinforcement learning model provides a cohesive framework for understanding interconnected cognitive functions.
• This model offers a valuable tool for simulating cognitive impairments and evaluating potential therapeutic strategies.
• The research facilitates a more integrated approach to cognitive neuroscience and clinical applications.