Research map / Architectures and efficiency / RNNs and normalization
PoF: Post-Training of Feature Extractor for Improving Generalization
Ikuro Sato, Yamada Ryota, Masayuki Tanaka, Nakamasa Inoue, Rei Kawakami
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Post-train SAM with PoF for 10 epochs | SAM-trained WideResNet-28-10 · CIFAR-10 | test error rate (%): 2.50$\pm$0.07 -> 2.41$\pm$0.02 | Improved |
| Post-train SAM with PoF for 10 epochs | SAM-trained WideResNet-28-10 · CIFAR-100 | test error rate (%): 16.27$\pm$0.09 -> 16.07$\pm$0.15 | Improved |
| Post-train SAM with PoF for 50 epochs | SAM-trained WideResNet-28-10 · CIFAR-100 | test error rate (%): 16.27$\pm$0.09 -> 16.60$\pm$0.05 | Hurt |
| Post-train SAM with PoF for 50 epochs | SAM-trained WideResNet-28-10 · SVHN | test error rate (%): 1.63$\pm$0.03 -> 1.55$\pm$0.02 | Improved |
| Post-train SAM with PoF | SAM-trained WideResNet-28-10 · Fashion-MNIST | test error rate (%): 4.12$\pm$0.05 -> 4.25$\pm$0.05 | Hurt |
| Use PoF instead of SGD for post-training | WideResNet-28-10 on CIFAR-10 · CIFAR-10 | time per epoch: 21.8 s -> 25.6 s | Hurt |
| Use PoF instead of SAM for post-training | WideResNet-28-10 on CIFAR-10 · CIFAR-10 | time per epoch: 32.9 s -> 25.6 s | Improved |
| Fine-tune the classifier for 10 epochs after PoF (ablation) | SAM→PoF model · CIFAR-10 | test error rate (%): 2.40% -> 2.39\pm0.03\% | No clear effect |
Closest papers by meaning
- Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models (ICML 2025 · similarity 0.84)
- Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning (NeurIPS 2026 · similarity 0.83)
- OATS: Outlier-Aware Pruning Through Sparse and Low Rank Decomposition (ICLR 2025 · similarity 0.83)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.82)
- Exploring Landscapes for Better Minima along Valleys (NeurIPS 2025 · similarity 0.82)
- Investigating the Overlooked Hessian Structure: From CNNs to LLMs (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
