Image Classification on CIFAR-10 (Accuracy and FP32 Parity Check)
99.4AccuracyHuman Best
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Human BestMethod Type=Human Baseline2026.06 | 99.4 | — | |
| ARTS*Base Model=Qwen 4B, Test-time trained=true2026.06 | 98.2 | — | |
| ARTSBase Model=o32026.06 | 97.1 | — | |
| AIRAMethod Type=Prior Works2026.06 | 96.4 | — | |
| MLEvolveMethod Type=Prior Works2026.06 | 95.9 | — | |
| ARTSBase Model=Qwen 4B2026.06 | 95.7 | — | |
| LinearMethod Type=Prior Works2026.06 | 95.6 | — | |
| TUESource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 94.31 | — | |
| GUESource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 94.31 | — | |
| GUESource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 94.28 | — | |
| TUESource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 93.91 | — | |
| Symmetric 4-bit QAT with Tanh Soft ClippingHardware=CPU, Training=From scratch2026.03 | 92.34 | — | |
| DoReFaHardware=GPU, Training=From scratch2026.03 | 85 | — | |
| LSPSource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 38.5 | — | |
| EMNSource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 27.27 | — | |
| LSPSource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 24.16 | — | |
| EMNSource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 14.31 | — | |
| PUESource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 11.61 | — | |
| PUESource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 10.23 | — | |
| FUSESource=SVHN, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 10.04 | — | |
| FUSESource=CIFAR-100, Surrogate Model=ResNet-18, Filter Condition=Unfiltered2026.06 | 9.98 | — |