Sparse component analysis: A method that uncovers separable computations within neural population activity.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Neural populations are thought to use shared 'latent factors' for computation.
• Existing methods for uncovering these factors often require supervision, limiting novel discoveries.
• Understanding how neural activity composes actions is crucial for deciphering brain function.

Purpose of the Study:

• To introduce sparse component analysis (SCA), an unsupervised machine learning technique.
• To demonstrate SCA's ability to identify latent factors and compositional structure in neural data.
• To apply SCA across diverse neural systems and behaviors to validate its effectiveness.

Main Methods:

• Developed and applied sparse component analysis (SCA), an unsupervised learning algorithm.
• Utilized linear and nonlinear embedding techniques within SCA.
• Analyzed neural activity data from motor cortex (monkeys), C. elegans, and artificial neural networks.

Main Results:

• SCA successfully parcellated neural activity across various behaviors and datasets.
• The method revealed both expected and novel instances of compositional structure in population responses.
• Identified distinct computational roles for different sets of latent factors.

Conclusions:

• Sparse component analysis (SCA) is a powerful unsupervised tool for understanding neural computation.
• SCA can effectively uncover latent factors and their compositional roles in diverse neural systems.
• This approach facilitates the discovery of novel computational principles in the brain.