Representational spaces in orbitofrontal and ventromedial prefrontal cortex: task states, values, and beyond

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The orbitofrontal cortex (OFC) and ventromedial-prefrontal cortex (vmPFC) are crucial for decision-making.
• These brain regions encode not only expected value but also task states.

Purpose of the Study:

• To review evidence connecting value and state representations in the OFC/vmPFC.
• To propose that OFC/vmPFC integrate stimulus, context, and outcome information.
• To suggest that outcome maximization leads to complex representational spaces.

Main Methods:

• Review of existing neuroscientific and computational evidence.
• Analysis of representational principles in deep reinforcement learning (RL) models.
• Comparison of encoding principles between biological systems and artificial neural networks.

Main Results:

• OFC/vmPFC integrate stimulus, context, and outcome information.
• Deep RL models show comparable encoding principles, with mixed-selectivity nodes.
• Complex representational spaces are shaped by outcome maximization, challenging linear value signal assumptions.

Conclusions:

• OFC/vmPFC representations are more complex than previously characterized by linear value signals.
• Deep RL models offer insights into neural computation in decision-making circuits.
• Outstanding questions remain regarding learning across tasks, task-irrelevant information encoding, and hippocampus-PFC interactions.