The abstract is read at the publisher; the summary is JClub's.
Learned Poincaré-ball curvature c from extracellular spike recordings is significantly positive for ring-manifold motor circuits (MC_Maze, MC_RTT) and near-zero for unstructured frontal circuits (DMFC_RSG).
- Why it matters: Neural population activity depends on geometry, but linking learned latent structure to circuit topology has been difficult using real spike data.
- What they did: NeuroMASS v4 learns hyperbolic latent geometry end-to-end on the Neural Latents Benchmark (NLB'21), using masked self-prediction of spike counts plus an auxiliary behavior decoding objective, with a diagonal S4D state space model spanning 2–200 Hz.
- The result: Across four NLB datasets in rhesus macaque, learned curvature statistically predicts circuit topology, enabling a biologically interpretable geometric signature tied to specific circuit classes.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.