Web App: MSDA-Bench

Milestone:

MSDA-Bench cover

Application: MSDA-Bench
Title: Multi-Source Domain Adaptation Benchmark for Cross-Session EEG Classification
Language: Python / Streamlit
Authors: Yiming Shen and David Degras

Overview

MSDA-Bench is an interactive dashboard for comparing how source-session utilization strategies behave under cross-session EEG distribution shift. It is designed to inspect benchmark outputs rather than to run the pipelines directly.

Toolkit Role

MSDA-Bench is the benchmark interpretation layer for multi-source cross-session experiments.

CrossDA results -> MSDA-Bench -> pipeline comparison, stability analysis, and mechanism inspection

Main Capabilities

  • Pipeline comparison across MAP, DWP, MMP, and BDP variants.
  • Configuration effects across feature, classifier, DA method, and distance choices.
  • Selection sensitivity and ranking stability analysis.
  • Subject-level and target-session deep dives.
  • DA gain/harm analysis.
  • Session-role visualizations for source weighting, bridge/far partitioning, and MMP source selection.
  • Runtime and accuracy-time tradeoff views.

Statistical Convention

Summary statistics are computed on subject-level values rather than raw cells, so the dashboard emphasizes cross-subject behavior instead of overweighting subjects with more experiment cells.

Live Demo View on GitHub