Neural Noise

Neural population-based approaches to neural computation and behavior

2 min read

  • Neural population-based approaches have opened new windows into neural computations and behavior (The Transmitter).1
  • These approaches are relevant to decoding and BCI design.1 1

Gardner updates

  • Neural population-based approaches have opened new windows into neural computations and behavior, with relevance to decoding and BCI design (The Transmitter). 1

Weekly enrichment (2026-07-20)

  • The source is a perspective essay in The Transmitter by Matthew Perich arguing that neural manifolds—the assumption that geometric structure constrains neural population activity—offer a compact, population-centric lens for describing computation and its links to behavior.2
  • Perich highlights that manifold geometry underlies flexible behavior: the same neurons can support different behaviors via functionally separate (orthogonal) dimensions along the manifold, allowing animals to plan and execute movements simultaneously, and his own work shows that modifying activity in the planning dimensions accelerates motor learning.2
  • Causal evidence comes from BCI learning paradigms: Sadtler and colleagues showed that within-manifold neural activity patterns are learned easily whereas outside-manifold patterns are much harder, demonstrating that manifolds constrain how monkeys can modulate population activity.3 4
  • The manifold framework generalizes across species and systems, explaining features of grid cells in mouse entorhinal cortex and head-direction cells in flies, zebrafish, and mice, with reported consistency spanning C. elegans to mammals.3
  • The underlying formalism is “computation through neural population dynamics” (computation through dynamics, CTD), which treats neural circuits as dynamical systems whose population state evolves over time to perform sensory, cognitive, and motor computations (Vyas et al., Annual Review of Neuroscience, 2020).4
  • Low-dimensional latent structure is typically extracted with dimensionality-reduction methods such as principal component analysis, factor analysis, and Gaussian Process Factor Analysis, and evaluated by decoding/reconstruction performance and intrinsic dimensionality.4 5
  • BCIs are emphasized as a uniquely powerful paradigm because the experimenter has complete knowledge of and control over the mapping between recorded neural activity and behavior, enabling causal tests of learning.4
  • Newer analytical tools cited for uncovering manifold representations include CEBRA, MARBLE, and RATS, reflecting a rapidly evolving methodological toolkit.2
  • Recent modeling work (eLife, 2025) argues learning speed is limited not only by manifold geometry but by the network’s controllability and feedback-driven dynamics, and predicts that rapid adaptation to new BCI decoders depends on upstream remapping of sensory feedback beyond local motor-cortex plasticity.6

Footnotes

  1. https://news.google.com/rss/articles/CBMi3AFBVV95cUxPN29fVVFCdURmeU5NbUNXdnA3X2p3cklvUXZFRi11cklRSlZ3VlVHUXdDdFRwSmNZQ1dqeDZsX09xUlN2bEV0NVdlVUdDN2J5dnNIZl9KNjJmSnE5dHpTbVhyZl9vZ2x2SkZSMFZONExoa1M0dHFoYzFqbU1ITmkwNm56dmRBWEpxTk1vQ1NwWjJ2SUVXQUNsOVJUVklHbWRST184aE0yNjd0SEZGeDZPSU5LaF9jdFlOYXUtaXdqU1owWUlDV0kyLU9zR0JuRDU1eEZpVVBISmV2aEJB?oc=5 ↩ ↩2 ↩3 ↩4

  2. https://www.thetransmitter.org/neural-dynamics/neural-population-based-approaches-have-opened-new-windows-into-neural-computations-and-behavior/ ↩ ↩2 ↩3

  3. https://www.thetransmitter.org/neural-dynamics/neural-manifolds-latest-buzzword-or-pathway-to-understand-the-brain/ ↩ ↩2

  4. https://pmc.ncbi.nlm.nih.gov/articles/PMC7402639/ ↩ ↩2 ↩3 ↩4

  5. https://link.springer.com/article/10.1007/s10827-022-00839-3 ↩

  6. https://elifesciences.org/reviewed-preprints/111322 ↩

Graph View