Seminar: Mathematical Psychology XXVII

Time: 2026-06-11 11:11:00

Date: June 11, 2026

Participants: Prof. Han Biao, Prof. Lin Yizun, Prof. Ye Qi, and student Li Xinyan, among others

Seminar Topic: Reconstruction Algorithms for Intracranial EEG Signals (LCMV / Tensor Completion / Physics-Constrained Fusion)

Content: With sparse spatial sampling of intracranial electrodes and signal blind zones, conventional LCMV yields a short-window correlation of only 0.5, dropping to 0.26 at 2 seconds, and thus fails. Purely data-driven tensor completion (e.g., SPBCD) is more accurate but lacks physiological interpretability. The LCMV lead-field matrix was therefore converted into a graph-manifold regularizer embedded in SPBCD, with three-dimensional tensor channel decoupling to avoid information leakage; blind tests on 118 channels gave a mean correlation of 0.71, reaching 0.93 on some channels.