Optical Character Recognition on MNIST (test)
0.71Error RateFastfood 1024
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Fastfood 1024Number of features=1024, Dropout=true, Params=38,8212014.12 | 0.71 | — | |
| Fastfood 2048Number of features=2048, Dropout=true, Params=52,1242014.12 | 0.71 | — | |
| Adaptive Fastfood 1024Number of features=1024, Dropout=true, Params=38,8212014.12 | 0.72 | — | |
| Adaptive Fastfood 2048Number of features=2048, Dropout=true, Params=52,1242014.12 | 0.73 | — | |
| Fastfood 1024 (ND)Number of features=1024, Dropout=false, Params=38,8212014.12 | 0.83 | — | |
| Adaptive Fastfood 1024 (ND)Number of features=1024, Dropout=false, Params=38,8212014.12 | 0.86 | — | |
| Reference ModelArchitecture=LeNet, Params=430,5002014.12 | 0.87 | — | |
| Fastfood 2048 (ND)Number of features=2048, Dropout=false, Params=52,1242014.12 | 0.9 | — | |
| Adaptive Fastfood 2048 (ND)Number of features=2048, Dropout=false, Params=52,1242014.12 | 0.92 | — | |
| CLIPEvaluation Protocol=Linear2021.02 | — | 99.2 | |
| CLIPEvaluation Protocol=Zero-Shot2021.02 | — | 88.4 | |
| ES BestEvaluation Protocol=Linear2021.02 | — | 98.9 | |
| Raw PixelsEvaluation Protocol=Linear2021.02 | — | 92.5 | |
| SOTAEvaluation Protocol=Finetune2021.02 | — | 99.8 |