EUVIMEDEUVIMED entdecken
Crossref

Behavioral Latency as Weak Event-Time Supervision for EEG Reaction-Time Decoding

Anuar Aimoldin, Ayana Mussabayeva, Yedige Mussabayev, Xue Liu, Kun Zhang

Neural Computation · 05.11.2026 · 1-38

DOI: 10.1162/neco.a.1604 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Neural Computation, online erschienen am 05.11.2026. Verfasst von Anuar Aimoldin und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Single-trial EEG analyses are often organized around events and latencies, such as stimulus onset, response execution, ERP component timing, and movement onset. Yet EEG-based reaction-time prediction is commonly posed as scalar regression on a fixed stimulus-locked window, so reaction time is treated as a window-level label rather than timing evidence about response-relevant dynamics. Here we reformulate trial-wise reaction-time decoding as event-time posterior modeling. Instead of predicting a single reaction time (RT) directly, the model estimates a posterior distribution over response-relevant event times, p(t event |X), and uses the posterior mean as the RT estimate. This treats behavioral latency as a weak observation of latent response-relevant timing, making the output representation itself part of the computational hypothesis. We evaluate this formulation on the healthy brain network contrast change detection EEG task under a subject-disjoint, release-separated protocol. Across five seeds, distributional event-time supervision consistently improves held-out RT prediction relative to scalar regression and temporal-readout controls. Controlled objective comparisons isolate supervision of the event-time distribution, rather than expectation-based readout alone, as the source of this gain; architecture controls establish that the effect persists across four temporal backbones and is not explained by model scale. Beyond point prediction, the event-time posterior serves as a diagnostic for separating temporal evidence from scalar shortcuts. Posterior geometry characterizes temporal concentration, target alignment, and interval behavior, while observation-noise calibration separates latent event-time concentration from predictive uncertainty over observed RT. We also use shifted-crop inference to probe shortcut use versus temporal localization. A matched shift-jitter intervention improves shifted-crop robustness, increases mean sensitivity, and moves predictions more often in the expected crop-relative direction. Sensitivity remains below ideal crop-relative localization, leaving a clear equivariance gap. Together, these results establish event-time posterior modeling as a probabilistic and interpretable formulation for linking single-trial EEG dynamics to behavioral timing.

Zitieren

APA

Aimoldin, A., Mussabayeva, A., Mussabayev, Y., Liu, X., & Zhang, K. (2026). Behavioral Latency as Weak Event-Time Supervision for EEG Reaction-Time Decoding. Neural Computation, 1-38. https://doi.org/10.1162/neco.a.1604

Vancouver

Aimoldin A, Mussabayeva A, Mussabayev Y, Liu X, Zhang K. Behavioral Latency as Weak Event-Time Supervision for EEG Reaction-Time Decoding. Neural Computation. 2026:1-38. doi:10.1162/neco.a.1604

BibTeX
@article{Aimoldin2026,
  author = {Aimoldin, Anuar and Mussabayeva, Ayana and Mussabayev, Yedige and Liu, Xue and Zhang, Kun},
  title = {Behavioral Latency as Weak Event-Time Supervision for EEG Reaction-Time Decoding},
  journal = {Neural Computation},
  year = {2026},
  pages = {1--38},
  doi = {10.1162/neco.a.1604}
}
Kostenlos starten →