Multimodal Machine Unlearning on Forget-Set 15% forgetting ratio
61.45Class. VQA AccuracyKL_Min
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| KL_MinBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 61.45 | 59.69 | 68.3 | 73.3 | 19.2 | 28.7 | |
| BeforeBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 61.23 | 60.73 | 70.7 | 74.3 | 19.6 | 28.8 | |
| GABackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 60.13 | 59.91 | 24.8 | 68.9 | 17.2 | 27.9 | |
| NPOBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 52.46 | 55.64 | 22.5 | 58.7 | 14.9 | 23.9 | |
| MANUBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 50.77 | 61.28 | 60.7 | 72.4 | 14.3 | 24.2 | |
| MMunlearnerBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 31.49 | 63.53 | 32.3 | 74.6 | 10.1 | 27.6 | |
| GA_diffBackbone=Qwen3-VL-8B, Forgetting Ratio=15%2026.05 | 7.28 | 62.49 | 8.9 | 73.3 | 3.1 | 26.8 |