Quantization
Quantization is a promising technique to reduce the computation cost of neural network training, which can replace high-cost floating-point numbers (e.g., float32) with low-cost fixed-point numbers (e.g., int8/int16).
Source: Adaptive Precision Training: Quantify Back Propagation in Neural Networks with Fixed-point Numbers
Papers
Showing 1–10 of 4925 papers
All datasetsImageNetCIFAR-10Wiki-40BAgeDB-30CFP-FPCOCO (Common Objects in Context)IJB-BIJB-CKnowledge-based:LFW
Benchmark Results
| # | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| 1 | FQ-ViT (ViT-L) | Top-1 Accuracy (%) | 85.03 | — | Unverified |
| 2 | FQ-ViT (ViT-B) | Top-1 Accuracy (%) | 83.31 | — | Unverified |
| 3 | FQ-ViT (Swin-B) | Top-1 Accuracy (%) | 82.97 | — | Unverified |
| 4 | FQ-ViT (Swin-S) | Top-1 Accuracy (%) | 82.71 | — | Unverified |
| 5 | FQ-ViT (DeiT-B) | Top-1 Accuracy (%) | 81.2 | — | Unverified |
| 6 | FQ-ViT (Swin-T) | Top-1 Accuracy (%) | 80.51 | — | Unverified |
| 7 | FQ-ViT (DeiT-S) | Top-1 Accuracy (%) | 79.17 | — | Unverified |
| 8 | Xception W8A8 | Top-1 Accuracy (%) | 78.97 | — | Unverified |
| 9 | ADLIK-MO-ResNet50-W4A4 | Top-1 Accuracy (%) | 77.88 | — | Unverified |
| 10 | ADLIK-MO-ResNet50-W3A4 | Top-1 Accuracy (%) | 77.34 | — | Unverified |