The 'neat' and 'messy' in task-dependent neural geometry and computation

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain must flexibly switch between task rules to solve real-world problems.
• Neural population activity encodes task states and variables, facilitating task switching.
• However, task-irrelevant variables and behavioral suboptimalities complicate neural computations.

Purpose of the Study:

• To review the dual facets of task-dependent computation in the brain.
• To highlight both the 'neat' and 'messy' features of neural representations.
• To propose future research directions for understanding brain computation.

Main Methods:

• Review of experimental observations of task-dependent computation in primates.
• Analysis of neural data and neural network modeling.
• Examination of behavioral data and switch costs.

Main Results:

• Neuronal population activity exhibits structured representations enabling efficient task switching ('neat' features).
• Task-irrelevant variables and entangled representations are observed, alongside behavioral switch costs ('messy' features).
• A duality exists between idealized and complex aspects of neural computation.

Conclusions:

• Understanding task-dependent computation requires acknowledging both structured and unstructured neural dynamics.
• Future theories should incorporate brain-specific constraints.
• Network models fitted to neural and behavioral data are crucial for advancing computational neuroscience.