Multi-task image classification on Multi-Fashion MNIST (test)
83.78Accuracy 1DSelect-k
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DSelect-kGate selection type=Static2021.06 | 83.78 | 83.34 | 1.8 | — | |
| DSelect-kGate selection type=Per-example2021.06 | 83.69 | 83.13 | 1.5 | — | |
| Gumbel SoftmaxGate selection type=Static2021.06 | 83.66 | 83.28 | 1.5 | — | |
| Top-kGate selection type=Per-example2021.06 | 83.66 | 83.15 | 4 | — | |
| Softmax MoEGate selection type=Baseline2021.06 | 83.48 | 82.81 | 8 | — | |
| Top-kGate selection type=Static2021.06 | 83.44 | 82.66 | 4 | — | |
| Shared BottomGate selection type=Baseline2021.06 | 82.05 | 81.37 | — | — |