Handwritten digit classification on MNIST (test)
97.79Accuracy (Regular)CIL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CILData Partitioning=i.i.d., A (hyperparameter)=small (A=10)2026.01 | 97.79 | 97.32 | 97.64 | 97.09 | |
| Standard SGDData Partitioning=i.i.d.2026.01 | 97.42 | 97.29 | 97.36 | 97.41 | |
| DeCaForkData Partitioning=i.i.d.2026.01 | 97.31 | 97.25 | 97.49 | 97.12 | |
| CILData Partitioning=i.i.d., A (hyperparameter)=large (A=350)2026.01 | 97.16 | 97.44 | 97.6 | 97.25 | |
| Structured Recurrent SNNLearning rule=WTA + modulatory plasticity, Depth / Recurrence=Deep recurrent2026.05 | 97 | — | — | — | |
| STDPLearning rule=Unsupervised STDP + readout, Depth / Recurrence=Shallow, feedforward2026.05 | 95 | — | — | — | |
| e-propLearning rule=Approx. gradient descent, Depth / Recurrence=Deep recurrent2026.05 | 92 | — | — | — | |
| Reward-modulated STDPLearning rule=Three-factor plasticity, Depth / Recurrence=Shallow / recurrent2026.05 | 88 | — | — | — | |
| WTA-based SNNLearning rule=Margin / competition, Depth / Recurrence=Shallow2026.05 | 88 | — | — | — | |
| Reservoir SNN + readoutLearning rule=Fixed recurrence, Depth / Recurrence=Recurrent, untrained2026.05 | 85 | — | — | — |