Research map / Learning theory and algorithms / Kernels and regression theory
A Reparametrization-Invariant Sharpness Measure Based on Information Geometry
Cheongjae Jang, Sungyoon Lee, Frank C. Park, Yung-Kyun Noh
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 IGS sharpness regularizer to SGD | SGD, 3FCN · MNIST | test accuracy (%): 98.13 -> 98.65 | Improved |
| add IGS sharpness regularizer to SGD | SGD, VGG11-BN · CIFAR-10 | test accuracy (%): 92.75 -> 93.81 | Improved |
| add IGS sharpness regularizer to SGD | SGD, ResNet-56 · CIFAR-10 | test accuracy (%): 94.08 -> 94.96 | Improved |
| add IGS sharpness regularizer to SGD | SGD, WRN-16-8 · CIFAR-10 | test accuracy (%): 95.98 -> 96.57 | Improved |
| add IGS sharpness regularizer to SGD | SGD, VGG11-BN · CIFAR-100 | test accuracy (%): 72.25 -> 74.24 | Improved |
| add IGS sharpness regularizer to SGD | SGD, ResNet-56 · CIFAR-100 | test accuracy (%): 73.14 -> 76.07 | Improved |
| add IGS sharpness regularizer to SGD | SGD, WRN-16-8 · CIFAR-100 | test accuracy (%): 80.55 -> 82.44 | Improved |
| IGS regularizer vs gradient regularization (GR) | SGD+IGS · CIFAR-100 ResNet-56 | test accuracy (%): 74.67 -> 76.07 | Improved |
| IGS regularizer vs ASAM | SGD+IGS · CIFAR-10 WRN-16-8 | test accuracy (%): 97.03 -> 96.57 | Hurt |
| IGS regularizer vs SAM/ASAM | SGD+IGS · CIFAR-10/100, MNIST | test accuracy (%) | Mixed |
| add IGS regularizer on poisoned-label toy data | SGD, 3-layer FCN · synthetic 3-class poison | test accuracy (%) | Improved |
| IGS vs other sharpness measures under nonlinear reparametrization | IGS · Figure 2 | test accuracy (%) | Improved |
Closest papers by meaning
- A Universal Class of Sharpness-Aware Minimization Algorithms (ICML 2024 · similarity 0.86)
- On the Limitations of Fractal Dimension as a Measure of Generalization (NeurIPS 2024 · similarity 0.86)
- Super Consistency of Neural Network Landscapes and Learning Rate Transfer (NeurIPS 2024 · similarity 0.85)
- Simplicity Bias via Global Convergence of Sharpness Minimization (ICML 2024 · similarity 0.85)
- Understanding Nonlinear Implicit Bias via Region Counts in Input Space (ICML 2025 · similarity 0.85)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 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.
