Research · 2025
State-Space Neural Decoding
From neural activity to movement.
The project
Built and benchmarked state-space decoders — Kalman filter, RTS smoother, and Preferential Subspace Identification (PSID) — to decode motor kinematics from 96-neuron primate motor cortex recordings (O’Doherty dataset). Validated Kalman optimality on the innovation sequence with Jarque-Bera and Ljung-Box tests; PSID improved velocity decoding from R2 = 0.41 (KF) to R2 = 0.48 (+16.6%).
At a glance
R² 0.48 — velocity decoding — PSID, +16.6% over Kalman.
Built with
Neural Decoding · Kalman Filter · PSID · Python