TraceGuard: Adaptive Multimodal Poison Filtering through Cross-Feature Rank Agreement
TraceGuard filters poisoned image-text training data via rank agreement among six corpus-level features, removing 98.4% of poison and 5.4% of clean examples.
1 stories, newest first · one card per story
TraceGuard filters poisoned image-text training data via rank agreement among six corpus-level features, removing 98.4% of poison and 5.4% of clean examples.
Today 1.6K readers · 2.4K views · 771 reads · all time 189K views