The geometry of context-dependent biased decisions during learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Adaptive behavior relies on inferring environmental context and adjusting decisions.
• Understanding how reward context is learned, represented, and updated is crucial for decision-making research.

Purpose of the Study:

• Investigate the neural mechanisms underlying reward context inference and adaptation in decision-making.
• Examine how the lateral prefrontal cortex (LPFC) represents and updates changing reward contingencies.

Main Methods:

• Recorded neural activity from large populations of neurons in the LPFC of macaque monkeys during a direction-discrimination task.
• Task involved unpredictable alternations in reward contingencies, requiring context inference.
• Analyzed neural population activity using representational geometry and compared with recurrent neural network models.

Main Results:

• Monkeys demonstrated one-trial inference of context switches, adjusting choices and reaction times accordingly.
• Neural representations showed systematic geometric shifts across contexts, primarily along the decision-variable axis.
• Learning accelerated the speed of these representational transitions, correlating with faster behavioral generalization.

Conclusions:

• LPFC circuits support hierarchical inference by encoding reward context through structured shifts in representational geometry.
• This geometric encoding allows for context-dependent biases and flexible control of decision policies.
• Findings provide a mechanism for rapid generalization and adaptation in decision-making under changing environments.