Geometry of neural computation unifies working memory and planning

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Cognitive functions like working memory, decision-making, and planning are crucial for real-world tasks.
• These functions have traditionally been studied in isolation, limiting a unified understanding of brain processes.

Purpose of the Study:

• To propose and investigate contingency representations as a unifying mechanism for working memory and planning.
• To differentiate contingency representations from traditional sensory models of working memory.

Main Methods:

• Designed a novel task to distinguish between different representational types.
• Utilized task-optimized recurrent neural networks to explore circuit mechanisms.
• Conducted behavioral experiments with human participants.
• Generated predictions for neural data analysis.

Main Results:

• Contingency representations provide a unified framework for working memory and planning.
• Neural network models demonstrated how contingency representations can explain prefrontal cortex activity.
• Human behavior aligned with contingency representations, not traditional sensory models.

Conclusions:

• Contingency representations offer a novel neural strategy unifying working memory, planning, and decision-making.
• This framework can explain observed neurophysiological and behavioral data.
• Provides testable predictions for future neural investigations.