Medical Image Classification on MedMNIST v2 (test)
84.8Average AccuracyOracle
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OracleBackbone=RN18, Scheduler=FIFO, Trials=1002024.03 | 84.8 | 95.9 | 97.8 | 80.8 | 91.9 | 73.2 | 79.8 | 92.8 | 52.5 | 89.7 | 94.5 | 84.4 | 11 | |
| SelVSBackbone=RN18, Scheduler=FIFO, Trials=100, Held-out validation set=10%2024.03 | 83.2 | 94.9 | 96.9 | 80.3 | 90.5 | 72.8 | 78 | 92.5 | 46 | 85.3 | 94.4 | 83.5 | 5 | |
| TwinBackbone=RN18, Scheduler=FIFO, Trials=100, Held-out validation set=None2024.03 | 83.1 | 95.3 | 97.1 | 80.8 | 90 | 72.8 | 77.3 | 89.3 | 47.8 | 85.9 | 93.7 | 83.8 | 7 | |
| Domain-Aware Reinforcement Fine-TuningShots=82026.01 | 26.8 | 13.3 | 18.9 | 33.1 | 35.4 | 33.1 | — | — | — | — | — | — | — | |
| Domain-Aware Reinforcement Fine-TuningShots=42026.01 | 25.3 | 13.2 | 17.2 | 32.8 | 31.4 | 31.7 | — | — | — | — | — | — | — | |
| Visual-RFTShots=82026.01 | 24.4 | 10.8 | 17.4 | 29.9 | 32.6 | 31.5 | — | — | — | — | — | — | — | |
| SFTShots=82026.01 | 24.1 | 10.7 | 17.1 | 28.9 | 32.3 | 31.4 | — | — | — | — | — | — | — | |
| Domain-Aware Reinforcement Fine-TuningShots=22026.01 | 23.2 | 12.4 | 16.1 | 23.5 | 33.2 | 30.9 | — | — | — | — | — | — | — | |
| Visual-RFTShots=42026.01 | 23 | 10.1 | 16 | 27.6 | 30.9 | 30.6 | — | — | — | — | — | — | — | |
| SFTShots=42026.01 | 22.9 | 10 | 16.1 | 27.3 | 30.6 | 30.3 | — | — | — | — | — | — | — | |
| Visual-RFTShots=22026.01 | 21.6 | 9.81 | 15.6 | 22.3 | 30.3 | 29.8 | — | — | — | — | — | — | — | |
| Domain-Aware Reinforcement Fine-TuningShots=12026.01 | 21.5 | 9.85 | 15.6 | 21.8 | 30.6 | 29.7 | — | — | — | — | — | — | — | |
| SFTShots=22026.01 | 21.4 | 9.76 | 15.5 | 21.8 | 30.1 | 29.8 | — | — | — | — | — | — | — | |
| Visual-RFTShots=12026.01 | 21 | 9.39 | 15.2 | 21.1 | 29.9 | 29.5 | — | — | — | — | — | — | — | |
| SFTShots=12026.01 | 20.9 | 9.1 | 15.2 | 21.3 | 29.8 | 29.5 | — | — | — | — | — | — | — | |
| Qwen2.5-VL2026.01 | 20.7 | 8.7 | 14.8 | 21.7 | 29 | 29.1 | — | — | — | — | — | — | — |