New AI Sorts Brain Signals Better — Could Speed Brain Research

Researchers Zishuo Feng and Feng Cao propose VanillaSort, a method for spike sorting that combines multichannel detection with spatially augmented, template-guided clustering. The problem it tackles: training spike detectors on real recordings is hard because algorithmically generated labels can be noisy and incomplete. VanillaSort's detector, VanillaDet, uses visibility-aware masking, truncated Gaussian targets and a temporally tolerant positive-bag loss, followed by conditional event-SNR gating; VanillaCluster then combines HuiduRep embeddings with relative-amplitude features for Gaussian mixture clustering and refines assignments using cross-fitted waveform templates built from selected core events. On the static and drift subsets of Hybrid Janelia, VanillaDet improved detection accuracy over SimSort by two and three percentage points, respectively, and the complete pipeline improved sorting performance over the corresponding HuiduRep baselines. The authors say the results support learning from imperfect real-data labels and incorporating waveform consistency into neuronal assignment.