My research studies how to evaluate and adapt models when data changes. Recent work examines which training history an evaluation permits, when uncertainty supports excluding a source session, and how model updating affects future performance and individual retention. EEG and brain-computer interfaces provide the main empirical setting, alongside reproducible Python/R research software.
Published · Frontiers in Human Neuroscience - Brain-Computer Interfaces
Yiming Shen and David Degras
Published in Frontiers in Human Neuroscience, 28 Aug 2026 A matched comparison of global and selective source-session strategies for within-subject, cross-session motor-imagery EEG transfer.
Recommended citation: Shen Y and Degras D (2026) A confidence-gated source selection strategy for cross-session transfer in brain–computer interfaces. Front. Hum. Neurosci. 20:1895016. doi:10.3389/fnhum.2026.1895016. https://doi.org/10.3389/fnhum.2026.1895016
Under review at Scientific Reports (solo author) A cross-participant study of MI-EEG system selection: single decoders, uncertainty-routed decoder pairs, and ensembles compared by final accuracy and execution cost, with calibration-based selection tested on held-out participants.
A multi-dataset audit measures how synthetic covariance augmentation affects decoder-relevant structure and cross-session accuracy.
Recommended citation: Shen, Y. (2026). Covariance-Aware Augmentation for Cross-Session BCI: A Multi-Dataset Evaluation. Preprints.org, version 1. doi:10.20944/preprints202610.0758.v1. https://www.preprints.org/manuscript/202610.0758
Controlled replay comparisons and a record-linkage audit distinguish what session similarity identifies from whether retrieved history improves decoder updates.
Matched experiments separate omitted-channel feature prediction, downstream classification, and each participant’s performance on a permanent old test.
Manuscript in preparation · collaborative doctoral research A tensor-based statistical framework extending MCCA to high-dimensional datasets, preserving tensor structure in multi-view data.
Poster project A ‘Linear-First’ evaluation of feature-representation and fixed-feature classifier choices under nested validation, with separate measurements of predictive accuracy and computational cost.
Ph.D. Dissertation · University of Massachusetts Boston
Ph.D. Dissertation, University of Massachusetts Boston (Aug 2026) Reliable MI-EEG decoding as a sequence of representation, transfer, drift measurement, adaptation, and recalibration choices.
A proposed extension of low-channel EEG adaptation research examining EEG’s added value for movement measurement, its reliability across training days, and its potential contribution to motor-learning feedback.
Method development and simulation studies calibrate drift scores against acceptable reference sessions in frozen EEG model representations. Real-EEG evaluation is pending.