2026-09-01 00:00 China Papers Movement & Control Translated from EN

U2Multi-UDA Merges Alignment, Fusion and Fine-Tuning to Lift Motor Imagery Accuracy

Summary Researchers have proposed U2Multi-UDA, a unified multilevel multisource unsupervised domain adaptation framework for motor imagery brain-computer interfaces that raised mean accuracy by 2.69, 1.89 and 3.83 percentage points over the best-performing baselines on two public datasets and one self-constructed dataset, with consistent gains in Kappa. It tackles two persistent obstacles, wide variation between subjects and a shortage of labeled target-domain data, which existing methods address at only one level of adaptation, whether domain alignment, feature interaction or model fine-tuning. The single pipeline aligns source and target distributions with optimal transport while mutual information estimates how relevant each source domain is to the target, fuses spatio-temporal EEG features through multisource cross-attention guided by those weights and reinforced with pseudolabels, and closes with segmented weight-decomposed low-rank adaptation for parameter-efficient fine-tuning that curbs overfitting.

Why it matters Cross-subject transfer, not raw classifier accuracy, is what keeps motor imagery systems tied to long per-user calibration sessions. Combining optimal transport alignment, weighted multisource fusion and low-rank fine-tuning in one pipeline is an engineering answer to that, and the reported gains are stated against best-performing baselines rather than weak ones. The evidence package carries no author affiliations, and the results remain benchmark-level rather than clinical.

BCIwiki (bciwiki.com) – The study was published in IEEE Transactions on Neural Networks and Learning Systems on September 1, 2026. It proposes U2Multi-UDA, a unified multilevel multisource unsupervised domain adaptation framework for motor imagery decoding, a paradigm still constrained by intersubject variability and the scarcity of labeled target-domain data. Existing methods usually focus on a single adaptation level, whether domain alignment, feature interaction or model fine-tuning, which the authors say limits comprehensive cross-domain adaptation.

The framework folds three adaptation levels into one pipeline, according to the paper. Optimal transport first aligns source and target distributions, while mutual information estimates source-domain relevance weights that characterize each source domain’s contribution to the target. Spatio-temporal EEG features are then extracted and fused through multisource cross-attention, with the relevance weights guiding cross-domain feature fusion and pseudolabels strengthening target-domain feature learning. Segmented weight-decomposed low-rank adaptation then enables parameter-efficient target-domain fine-tuning while reducing overfitting.

On BCI Competition IV 2a, BCI Competition IV 2b and the self-constructed MI-GS dataset, mean accuracy improved by 2.69, 1.89 and 3.83 percentage points respectively over the best-performing baselines, with consistent gains in Kappa values. Ablation experiments support the contribution of each component, the authors say. The evidence package carries no author affiliation data.

Compiled by BCIwiki from public sources

Sources · 1

pubmed.ncbi.nlm.nih.gov 2026-09-01

Decoding Algorithms EEG Motor Decoding China

Read original ↗

Suggest a correction Revisions · none 中 / EN