Few-Shot Classification 
| 1-shot 5-way(omniglot) | 5-shot 5-way(omniglot) | 1-shot 5-way(Mini-ImageNet)sota | 5-shot 5-way(Mini-ImageNet) | 1-shot 5-way(FewRel) | |
| Matching the Blanks | 93.86 ± 0.08 | ||||
| MLMAN | 82.98 ± 0.20 | ||||
| PT+MAP(s+f)/WRN | 84.81 ± 0.22 | 90.62 ± 0.11 | |||
| Sill-Net/WRN | 82.99 ± 0.23 | 89.14 ± 0.12 | |||
| PT+MAP/WRN | 82.92 ± 0.26 | 88.82 ± 0.13 | |||
| MCT(Instance)/ResNet-12 | 78.55±0.86 | 86.03±0.42 | |||
| BD-CSPN/WRN-28-10 | 70.31±0.93 | 81.89±0.60 | |||
| SIB/WRN-28-10 | 70.0±0.6 | 78.9±0.4 | |||
| Acc + Amphibian/WRN | 64.21 ± 0.62 | 87.75 ± 0.73 | |||
| S2M2/WRN | 64.93 ± 0.18 | 83.18 ± 0.11 | |||
| FEAT/WRN | 65.10 ± 0.20 | 81.11 ± 0.14 | |||
| SimpleShot (CL2N)/DenseNet | 64.29 ± 0.20 | 81.50 ± 0.14 | |||
| MetaFun-Attention + data augment | 64.13 ± 0.13 | 80.82 ± 0.17 | |||
| CTM/ResNet-18 +data augment | 64.12 ± 0.82 | 80.51 ± 0.13 | |||
| Learning from Adversarial Features for Few-Shot Classification | 63.28 ± 0.46 % | 78.70 ± 0.42% | |||
| wDAE-GNN | 62.96 ± 0.15% | 78.85 ± 0.10% | |||
| MetAdapt | 62.82 | 79.35 | |||
| MetaOptNet-SVM/ResNet-12 | 62.64 ± 0.61 | 78.63 ± 0.46 | |||
| Self-Critique and Adapt + High-End MAML++ 実装 | 62.86% | 77.64% | |||
| LEO 実装 | 61.76 ± 0.08% | 77.59 ± 0.12% | |||
| DEEP COMPARISON NETWORK | 62.88 ± 0.83% | 75.84 ± 0.65% | |||
| iMAML, Hessian-Free | 99.50 ± 0.26% | 99.74 ± 0.11% | 49.30 ± 1.88 % | ||
| SNAIL | 99.07% ± 0.16% | 99.78% ± 0.09% | 55.71% ± 0.99% | 68.88% ± 0.92% | 67.29 ± 0.26 |
| Prototypical Networks 実装 | 98.8% | 99.7% | 49.42 ± 0.78% | 68.20 ± 0.66% | 69.20±0.20 |
| GNN | 99.2% | 99.7% | 50.33% ±0.36% | 66.41% ±0.63% | 66.23 ± 0.75 |
| ES-MAML | |||||
| MAML + Transduction | 98.7 ± 0.4% | 99.9 ± 0.1% | 48.70 ± 1.84% | 63.11 ± 0.92% | |
| Reptile + Transduction? | 97.68 ± 0.04% | 99.48 ± 0.06% | 49.97 ± 0.32% | 65.99 ± 0.58% | |
| MetaNet | 98.95 | 49.21 ± 0.96 | 64.46 ± 0.54 |
Meta-test accuracy on non-mutually-exclusive (NME) classification
| NME Omniglot 20-way 1-shot | NME Omniglot 20-way 5-shot | NME MiniImagenet 5-way 1-shot | NME MiniImagenet 5-way 5-shot | |
| MR-MAML (W) | 83.3 (0.8)% | 94.1 (0.1)% | 43.6 (0.6)% | 53.8 (0.9)% |
| TAML | 9.6 (2.3)% | 67.9 (2.3)% | 26.1 (0.6)% | 44.2 (1.7)% |
| MAML | 7.8 (0.2)% | 50.7 (22.9)% | 26.3 (0.7)% | 41.6 (2.6)% |
コメントをかく