Building compositional tasks with shared neural subspaces.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Cognitive flexibility allows humans to perform diverse tasks and adapt to changing demands.
• Artificial neural networks exhibit task compositionality by reusing representations and components.
• It remains unclear if the brain utilizes similar compositional strategies for flexible task performance.

Purpose of the Study:

• To investigate whether the brain employs compositional mechanisms for flexible multi-task performance.
• To determine if neural activity subspaces are shared and flexibly engaged across different tasks.

Main Methods:

• Monkeys were trained to switch between three compositionally related tasks.
• Neural recordings were used to analyze task-relevant information representation.
• Investigated how neural representations are transformed between sensory and motor subspaces during task switching.

Main Results:

• Shared subspaces of neural activity were identified representing task-relevant information across tasks.
• Neural representations transformed from sensory to motor subspaces during task performance.
• Monkeys adapted to task changes by updating beliefs and flexibly engaging relevant neural subspaces.

Conclusions:

• The brain flexibly performs multiple tasks by compositionally combining task-relevant neural representations.
• Shared neural subspaces facilitate cognitive flexibility and efficient task switching.
• Neural compositionality is a key mechanism for adaptive behavior in the brain.