Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers
Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall
Adaptive decoding of temporally variable neural activity in single-trial time series
Pablo Oyarzo1, Radoslaw Martin Cichy1, Diego Vidaurre2; 1Freie Universität Berlin, 2Aarhus University
Presenter: Pablo Oyarzo
Cognitive processes, specially those involving higher-order functions, often unfold with temporal variability. This complicates the use of time-locked analysis techniques, including standard machine learning-based decoding methods. Although existing methods perform well in tasks with externally timed events, decoding covert processes -such as imagery or recall- remains difficult due to uncertainty in the timing of the underlying neural dynamics. In these cases, task-relevant neural signals may occur at variable latencies across trials, violating the temporal alignment assumptions of standard decoding models. We introduce the Adaptive Decoding Algorithm (ADA), a nonparametric framework for decoding under temporal uncertainty. ADA performs two coupled tasks: (i) it estimates, for each trial, the temporal window most likely to reflect task-relevant neural activity, and (ii) it uses this information to decode the trial label. Using controlled simulations, we show that ADA outperforms conventional methods that assume fixed temporal structure. These results demonstrate that explicitly modeling trial-specific timing can substantially improve decoding performance in scenarios where the timing of relevant neural activity is unknown.
Topic Area: Methods & Computational Tools
Extended Abstract: Full Text PDF