← 研踪 PaperTrailarXiv:2610.10850 · v2
Similar Predictive Fit but Different Latent Dynamics: Characterizing Learned Dynamical Structure in Personalized Models of Brain Disorders
首次提交 2026-10-08 03:54
版本更新 2026-11-13 05:23
雷达观测 2026-11-17 04:58
摘要
As AI models move toward clinical decision-making and personalized treatment, understanding \emph{what} a model learns is important beyond predictive accuracy alone. We investigate whether personalized latent dynamics reveal clinically associated differences even when predictive fit is similar. A lightweight CNN--Transformer EEG foundation model pretrained on the Temple University EEG Corpus (TUEG) extracts segment-level representations. Using the Temple University Epilepsy Corpus (TUEP), representations are mapped to a shared latent-state space, and sparse multinomial logistic transition distributions (mLTD) are fit independently to each subject to obtain personalized transition-dependency graphs . Analyses include subjects (99 epilepsy / 99 non-epilepsy). At , epilepsy subjects exhibit substantially denser learned dependency structure (), with the same pattern at (19.90 vs. 13.46; ). Graph-derived features provide moderate group discrimination under 5-fold subject-wise cross-validation (AUROC 0.68 at ; 0.65 at ). In contrast, held-out log-likelihood is nearly identical between groups at ( vs. ; ), with similarly matched next-state prediction (AUROC 0.855 vs. 0.861; ). Thus, similar predictive fit does not imply similar learned dynamics: groups can be comparably predictable while differing substantially in the internal dynamical structure learned by personalized models. This distinction motivates evaluating learned structure alongside predictive performance in personalized clinical models.
作者备注:Withdrawn pending administrative review of publication-related requirements
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-13 05:23 · 最新观测版本
Similar Predictive Fit but Different Latent Dynamics: Characterizing Learned Dynamical Structure in Personalized Models of Brain Disorders
As AI models move toward clinical decision-making and personalized treatment, understanding \emph{what} a model learns is important beyond predictive accuracy alone. We investigate whether personalized latent dynamics reveal clinically associated differences even when predictive fit is similar. A lightweight CNN--Transformer EEG foundation model pretrained on the Temple University EEG Corpus (TUEG) extracts segment-level representations. Using the Temple University Epilepsy Corpus (TUEP), representations are mapped to a shared latent-state space, and sparse multinomial logistic transition distributions (mLTD) are fit independently to each subject to obtain personalized transition-dependency graphs . Analyses include subjects (99 epilepsy / 99 non-epilepsy). At , epilepsy subjects exhibit substantially denser learned dependency structure (), with the same pattern at (19.90 vs. 13.46; ). Graph-derived features provide moderate group discrimination under 5-fold subject-wise cross-validation (AUROC 0.68 at ; 0.65 at ). In contrast, held-out log-likelihood is nearly identical between groups at ( vs. ; ), with similarly matched next-state prediction (AUROC 0.855 vs. 0.861; ). Thus, similar predictive fit does not imply similar learned dynamics: groups can be comparably predictable while differing substantially in the internal dynamical structure learned by personalized models. This distinction motivates evaluating learned structure alongside predictive performance in personalized clinical models.
查看此版本解读 →查看此版本原文 ↗