NSE Report: Topological Signatures of Reversal Learning EEG Dataset: ds004295 — Reward Gain and Punishment Avoidance Reversal Learning
Authors/Creators
Description
Abstract
We report a topological analysis of neural dynamics during probabilistic reversal learning using persistent homology (H₁) applied to EEG phase-space trajectories. Using the ds004295 dataset (N = 22, reward gain and punishment avoidance reversal learning), we applied two complementary analyses: (1) a trial-window comparison of pre-reversal, post-reversal, and stable periods, and (2) a sliding-window analysis with 500 ms windows and 100 ms step, tracking H₁ persistence continuously around the reversal point.
The trial-window analysis confirmed a theta-band inverted-U trajectory: persistence increased at post-reversal (p = 0.014, d = +0.162) and decreased at stable (p = 0.004, d = +0.268). The sliding-window analysis revealed a dramatically stronger and temporally precise signal: H₁ persistence rose sharply following the reversal marker, peaking at a mean latency of 390 ms (median 600 ms) post-reversal, with baseline-vs-post significance of p < 0.0001 and Cohen’s d = +1.277. Twenty-five of 56 sliding-window timepoints were significant after FDR correction.
The sliding-window result directly addresses the key theoretical prediction: a new long-lived topological cycle (C-scar) is born approximately 400–600 ms after the reversal point, consistent with the time course of cognitive rule restructuring.
These findings provide the first time-resolved evidence for C-scar formation in neural phase space, offering a precise and experimentally controlled test of the NSE framework.