Research map / Label-efficient and robust learning / Recognition and long-tail
Quantifying and Optimizing Simplicity via Polynomial Representations
Tianren Zhang, Xiangxin Li, Minghao Xiao, Guanyu Chen, Feng Chen
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 effective-degree (ED) regularization with label anchoring | ViT-Tiny · CIFAR-10 (from scratch) | top-1 accuracy (%): 87.80 -> 90.82 | Improved |
| add SAM, ASAM or Jacobian regularization | ViT-Tiny · CIFAR-10 (from scratch) | top-1 accuracy (%): 87.80 -> 87.85 (SAM) | No clear effect |
| remove label anchoring from ED (ablation) | ED regularization · CIFAR-10 (ViT-Tiny) | top-1 accuracy (%): 90.82 -> 90.00 | Hurt |
| add ED regularization during CLIP fine-tuning | CLIP ViT-B/32 · ImageNet ID and average OOD | accuracy (%): ID 76.20 -> 77.14; avg OOD 44.04 -> 45.31 | Improved |
| add ED regularization during CLIP fine-tuning | CLIP ViT-B/16 · ImageNet ID and average OOD | accuracy (%): ID 81.35 -> 82.19; avg OOD 53.69 -> 55.29 | Improved |
| add ED regularization to ViT-S/16 trained from scratch on ImageNet | ViT-S/16 · ImageNet (original recipe) | top-1 accuracy (%): 71.37 -> 72.76 | Improved |
| add ED regularization to ViT-S/16 with strong recipe | ViT-S/16 · ImageNet (strong recipe) | top-1 accuracy (%): 74.42 -> 75.01 | Improved |
| add ED regularization with embedding-space paths | BERT-base · GLUE RTE, MRPC, CoLA | accuracy / Matthews corr.: RTE 70.28 -> 71.12; MRPC 86.74 -> 87.66; CoLA 62.31 -> 62.45 | Improved |
| add embedding mixup | BERT-base · GLUE RTE, MRPC, CoLA | accuracy / Matthews corr.: MRPC 86.74 -> 86.30; CoLA 62.31 -> 61.04; RTE unchanged 70.28 | Hurt |
| add ED regularization to PPO actor | PPO · Procgen Dodgeball, Fruitbot, Jumper, StarPilot (unseen levels) | test score | Improved |
| replace real-image interpolation paths with random noise (ablation) | ED · CIFAR-10 (Appendix C.1) | correlation and regularization performance | Hurt |
| replace Chebyshev basis with Legendre basis (ablation) | ED · CIFAR-10 (Appendix C.2) | performance | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ASAM — ASAM: Adaptive Sharpness-Aware Minimization for Scale-Invariant Learning of Deep Neural Networks (ICML 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Simplicity Bias in Overparameterized Machine Learning (AAAI 2024 · similarity 0.87)
- Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild (CVPR 2025 · similarity 0.86)
- Simplicity Bias via Global Convergence of Sharpness Minimization (ICML 2024 · similarity 0.86)
- Interpretability Illusions in the Generalization of Simplified Models (ICML 2024 · similarity 0.86)
- Simplicity Bias of Two-Layer Networks beyond Linearly Separable Data (ICML 2024 · similarity 0.86)
- On the Minimal Degree Bias in Generalization on the Unseen for non-Boolean Functions (ICML 2024 · similarity 0.85)
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.
