Modeling neural activity as time-varying signals on graphs across Kuramoto, Ca2+ imaging and fMRI
Pseudomanifold · x · 2026-11-09
In a thread, the Krishnaswamy Lab outlines its approach to neural modeling: represent the neural substrate as a graph whose nodes are neurons, neuron collections, or brain regions, with edges encoding proximity, anatomical connectivity, or functional coupling. Neural activity then becomes a time-varying signal on this graph.
As the accompanying figure shows, three data sources feed into the same representation: Kuramoto model simulations, Ca²⁺ imaging (cell adjacency graphs from neuronal activity), and fMRI (parcellated regions with functional connectivity), all converted into time-lapse signals on a graph.
This sets up a later critique (4/n) that existing methods reduce one axis: PCA, ICA and correlation-based connectivity collapse the spatial pattern, while HMMs and dynamic functional connectivity collapse the temporal pattern — both incur large information loss, whereas the graph framework preserves both dimensions.
Related event: Modeling Neural Activity as Time-Varying Signals on Graphs(3 posts)→
More from Research
- Amazon paper finds rule-following AI benchmarks mark correct answers wrong; one scorer fix lifts LegalBench scores by 80+ points — rohanpaul_ai · 2026-11-09
- DeepMind ships MuJoCo 4.0 with speculative contacts to fix collision tunneling — lukas_m_ziegler · 2026-11-09
- First/last author incentives are the biggest blocker to interdisciplinary team science — RexDouglass · 2026-11-09
- Separatrix 1.0 launches: free open-source desktop app for nonlinear dynamics — stevenstrogatz · 2026-11-09
- Essay challenges AI's computational functionalism: artificial consciousness may be impossible — sebkrier · 2026-11-09
- Report: Unreleased Anthropic model helped solve a major quantum-computing question — zephyr_z9 · 2026-11-09