Theoretical Neuroscience Podcast On low-dimensional manifolds in motor cortex - with Sara Solla - #36
12 snips
Jan 3, 2026 Sara Solla, a theoretical neuroscientist with a physics background, shares her journey from neural networks to pioneering manifold analyses of motor cortex activity. She explains how modern multi-electrode technology allows for population-level insights, revealing low-dimensional structures in motor tasks. Solla discusses the implications of these findings for brain-machine interfaces and how structured sensory inputs contribute to low-dimensional coding. Her experiences at Bell Labs and collaborations with notable figures add depth to the conversation, making complex concepts accessible.
AI Snips
Chapters
Books
Transcript
Episode notes
Hopfield Talk Sparked A Career Shift
- A John Hopfield talk at a physics meeting inspired Sara Solla to switch toward neural networks research.
- That led her to IBM and to work on simulated annealing and networks at Bell Labs.
Origins Of 'Optimal Brain Damage' Pruning
- Sara Solla, Jan LeCun and John Denker developed the Optimal Brain Damage pruning idea in 1989.
- The method prunes weights by saliency (second derivative times weight squared) and became influential decades later.
Population PCA Reveals Task Geometry
- Projecting population activity with PCA revealed eight clusters matching target directions in the center-out task.
- The neural clusters preserved the spatial ordering of physical targets, showing abstract population encoding.
