Best-arm identification on On-Line Auto Lending auto-loan replay
40.6Mean Stopping TimeTabPFN structured proxy
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TabPFN structured proxyProxy implementation=TabPFN structured proxy2026.07 | 40.6 | 40.8 | 43.1 | 100 | 0.822 | 0.76 | 61.4 | |
| Qwen2.5-7B LoRA raw CLSBase model=Qwen2.5-7B, Evaluation protocol=LoRA fine-tuned2026.07 | 52 | 52.3 | 54.5 | 100 | 0.725 | 0.707 | 50.5 | |
| GPT-5.5 API ICLBase model=GPT-5.5, Evaluation protocol=API ICL2026.07 | 95.6 | 95.7 | 97.6 | 100 | 0.264 | 0.327 | 9.1 | |
| Qwen2.5-7B local ICLBase model=Qwen2.5-7B, Evaluation protocol=local ICL2026.07 | 101.7 | 101.8 | 103.5 | 100 | 0.275 | 0.011 | 3.3 | |
| Reward-only PROBEProxy implementation=None (Baseline)2026.07 | 105.2 | 105.2 | 106.7 | 100 | — | — | 0 |