Image Classification on miniImageNet 20-way (Full)
93.2AccuracyLinear projection
Evaluation results
| Method | Links | |
|---|---|---|
| Linear projectionAdaptation=Backprop, Inference Compute=O(d_x^2), Inference Memory=4 MB, Learn Cost GPU Secs=780 s2026.05 | 93.2 | |
| FAASTAdaptation=Fast weights, Inference Compute=O(d_x^2), Inference Memory=4 MB, Learn Cost GPU Secs=2.1 s (-94.7%)2026.05 | 93 | |
| k-NN MemoryAdaptation=Memory, Inference Compute=O(N d_x), Inference Memory=80 MB, Learn Cost GPU Secs=41 s2026.05 | 88.6 | |
| Softmax MemoryAdaptation=Memory, Inference Compute=O(N d_x), Inference Memory=80 MB2026.05 | 87.9 | |
| Full FinetuningAdaptation=Backprop, Inference Compute=O(d_x^2), Inference Memory=4 MB, Learn Cost GPU Secs=1,603 s2026.05 | 87.5 | |
| CLIPAdaptation=No adapt, Inference Compute=O(d_x), Inference Memory=0 MB2026.05 | 86.7 |