Research map / Architectures and efficiency / RNNs and normalization
Temperature Balancing, Layer-wise Weight Analysis, and Neural Network Training
Yefan Zhou, Tianyu Pang, Keqin Liu, charles h martin, Michael W. Mahoney, Yaoqing Yang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add TempBalance layer-wise LR scheduling | SGD with cosine annealing (CAL) · CIFAR100, CIFAR10, SVHN, Tiny ImageNet | test accuracy | Improved |
| TempBalance vs spectral norm regularization | SNR · CIFAR100 VGG19 and ResNet34 | test accuracy | Hurt |
| combine TempBalance with SNR | TempBalance · all main settings | test accuracy | Improved |
| TempBalance vs SGDR, SGDP, Lookahead, LARS | SGD · ResNet18/34 CIFAR100 | test accuracy | Improved |
| Adam variants AdamW, AdamP, LAMB instead of SGD CAL | SGD CAL · CIFAR100 | test accuracy | No clear effect |
| tune initial learning rate for both methods (ablation) | TempBalance vs CAL · CIFAR100 ResNet and VGG | test accuracy | Improved |
| vary channel width 50%, 100%, 150% (ablation) | TempBalance vs CAL · CIFAR100 | test accuracy | Improved |
| replace PL_Alpha_Hill with SpectralNorm or AlphaWeighted (ablation) | TempBalance · CIFAR100 | test accuracy | Hurt |
| use Median lambda_min instead of Goodness-of-fit or Fix-finger (ablation) | PL fitting · CIFAR100 | test accuracy | Improved |
| replace linear interpolation with Sqrt, Log2, or Step LR assignment (ablation) | TempBalance · CIFAR100 VGG and ResNet | test accuracy | Hurt |
| narrow LR scaling range (s1,s2) from (0.5,1.5) toward (0.9,1.1) (ablation) | TempBalance · CIFAR100 | test accuracy | Hurt |
| apply TempBalance to object detection and language modeling | CAL · object detection, language modeling | generalization | Improved |
Closest papers by meaning
- To Cool or not to Cool? Temperature Network Meets Large Foundation Models via DRO (ICML 2024 · similarity 0.87)
- Models of Heavy-Tailed Mechanistic Universality (ICML 2025 · similarity 0.85)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.84)
- Analyzing & Reducing the Need for Learning Rate Warmup in GPT Training (NeurIPS 2024 · similarity 0.84)
- On The Concurrence of Layer-wise Preconditioning Methods and Provable Feature Learning (ICML 2025 · similarity 0.84)
- On the Surprising Effectiveness of Large Learning Rates under Standard Width Scaling (NeurIPS 2025 · similarity 0.84)
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.
