Ventilator setting prediction on Retrospective ICU Trajectory (Full)
0.102MSEGPT-5.2 (VDSS)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5.2 (VDSS)Framework=VDSS (multi agent), Backbone=GPT-5.22026.05 | 0.102 | 0.162 | 0.743 | |
| Gemini 3 Pro (VDSS)Framework=VDSS (multi agent), Backbone=Gemini 3 Pro2026.05 | 0.124 | 0.166 | 0.693 | |
| NoPref (Ablation)Framework=VDSS (multi agent), Backbone=GPT-5.2, Ablation=No Preference2026.05 | 0.128 | 0.183 | 0.707 | |
| Qwen3-VL-235B (VDSS)Framework=VDSS (multi agent), Backbone=Qwen3-VL-235B2026.05 | 0.13 | 0.183 | 0.661 | |
| NoImg (Ablation)Framework=VDSS (multi agent), Backbone=GPT-5.2, Ablation=No Image2026.05 | 0.145 | 0.194 | 0.66 | |
| NoImgNoPref (Ablation)Framework=VDSS (multi agent), Backbone=GPT-5.2, Ablation=No Image & No Preference2026.05 | 0.158 | 0.202 | 0.62 | |
| Qwen3(VL)-8B (VDSS)Framework=VDSS (multi agent), Backbone=Qwen3-VL-8B2026.05 | 0.175 | 0.21 | 0.354 | |
| GPT-5.2Framework=Single model end to end, Backbone=GPT-5.22026.05 | 0.343 | 0.381 | 0.153 | |
| Gemini 3 ProFramework=Single model end to end, Backbone=Gemini 3 Pro2026.05 | 0.362 | 0.386 | 0.13 | |
| Qwen3-VL-235BFramework=Single model end to end, Backbone=Qwen3-VL-235B2026.05 | 0.39 | 0.426 | 0.019 | |
| Qwen3-VL-8BFramework=Single model end to end, Backbone=Qwen3-VL-8B2026.05 | 0.486 | 0.595 | -0.132 |