iCaRL_codex

Research workspace for cross-subject class-incremental learning in motor imagery BCI.

Focus

• improve task1, task2, task3 final test accuracy
• prioritize task2/task3 balance and keep the method easy to explain
• keep changes mostly inside this repository

Useful Files

• main.py: training entry
• auto_experiment.py: experiment launcher with env/CLI bridge
• iCaRL.py: incremental learning logic
• mlm.py: pretrained backbone and lightweight adaptation modules
• metrics/latest.md: latest tracked summary
• metrics/experiments.csv: run history
• research/full_results_progression_20260316_zh.md: full-result progression
• research/method_code_index_20260316_zh.md: method-to-code mapping
• research/version_management_20260317_zh.md: version/tag usage