Principles of Temporal Processing Across the Cortical Hierarchy

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• The external world possesses intricate structures across various spatiotemporal scales.
• Machine learning models utilize hierarchical architectures with repeated spatial operations (pooling, normalization, pattern completion) to interpret spatial data.
• The brain also processes rich temporal information across multiple timescales.

Purpose of the Study:

• To investigate if the brain employs operations analogous to machine learning's spatial processing for temporal information.
• To define a candidate set of temporal operations.
• To review evidence for their hierarchical implementation in the mammalian cerebral cortex.

Main Methods:

• Literature review and theoretical analysis.
• Defining a set of temporal operations analogous to spatial operations in machine learning.
• Examining evidence for hierarchical implementation in the cerebral cortex.

Main Results:

• A candidate set of temporal operations, including temporal pooling, temporal normalization, and temporal pattern completion, is proposed.
• Evidence suggests these operations are implemented hierarchically in the mammalian cerebral cortex.
• Cortical processing stages can be understood through these temporal operations.

Conclusions:

• The mammalian cerebral cortex likely employs a hierarchical set of temporal operations for information processing.
• These operations (temporal pooling, normalization, pattern completion) are analogous to spatial operations used in machine learning.
• This framework offers a new perspective on how the brain processes complex temporal information.