Learning to Defer on CIFAR-10 with redundant synthetic experts
91.9System AccuracyDecoupled
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DecoupledNumber of experts (J)=162026.04 | 91.9 | 90.9 | |
| DecoupledNumber of experts (J)=322026.04 | 91.9 | 90.8 | |
| DecoupledNumber of experts (J)=82026.04 | 91.8 | 90.9 | |
| Classif. onlyNumber of experts (J)=162026.04 | 91.1 | 91.1 | |
| Classif. onlyNumber of experts (J)=82026.04 | 91 | 91 | |
| Classif. onlyNumber of experts (J)=322026.04 | 90.8 | 90.8 | |
| OvANumber of experts (J)=82026.04 | 90.1 | 87.5 | |
| OvANumber of experts (J)=162026.04 | 89 | 85.9 | |
| OvANumber of experts (J)=322026.04 | 88 | 84.4 | |
| PiCCENumber of experts (J)=162026.04 | 86 | 89.5 | |
| A-SMNumber of experts (J)=82026.04 | 86 | 83.1 | |
| PiCCENumber of experts (J)=322026.04 | 85.9 | 88 | |
| PiCCENumber of experts (J)=82026.04 | 85.8 | 88.8 | |
| A-SMNumber of experts (J)=162026.04 | 82.7 | 77.2 | |
| Add. CENumber of experts (J)=82026.04 | 81.5 | 90.2 | |
| Add. CENumber of experts (J)=162026.04 | 79 | 89.5 | |
| A-SMNumber of experts (J)=322026.04 | 78.3 | 68 | |
| Add. CENumber of experts (J)=322026.04 | 71.2 | 87.2 | |
| Mao25Number of experts (J)=82026.04 | 70.3 | 10.5 | |
| Mao25Number of experts (J)=162026.04 | 70.3 | 11.1 | |
| Mao25Number of experts (J)=322026.04 | 70.3 | 10.5 |