Research map / Optimization / Optimizers
Singularity-aware Optimization via Randomized Geometric Probing: Towards Stable Non-smooth Optimization
Ruoran Xu, Borong She, Xiaobo Jin, Qiufeng Wang
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 geometric damping (lambda>0) (ablation) | S-Adam (k=2) · QAT TinyImageNet (4-bit) | accuracy (%): 18.29% -> 22.81% | Improved |
| add geometric damping (lambda>0) (ablation) | S-Adam (k=2) · QAT CIFAR-100 (2-bit) | accuracy (%): 15.82% -> 18.11% | Improved |
| use k=2 probes instead of k=1 (ablation) | S-Adam · QAT CIFAR-100 (2-bit) | accuracy (%): 15.86% -> 18.11% | Improved |
| use k=2 probes instead of k=1 (ablation) | S-Adam · QAT Imagewoof2-160 (4-bit) | accuracy (%): 32.93% -> 35.57% | Improved |
| increase probes from k=2 to k=8 (ablation) | S-Adam · QAT TinyImageNet (4-bit) | accuracy (%): 22.81% -> 23.18% | No clear effect |
| increase probes from k=2 to k=8 (ablation) | S-Adam · QAT Imagewoof2-160 (4-bit) | accuracy (%): 35.57% -> 35.19% | No clear effect |
| use perturbation scale delta larger than 0.0001 (ablation) | S-Adam · QAT CIFAR-100 (2-bit) | accuracy (%): 14.78% -> 18.11% | Improved |
| replace AdamW with S-Adam (k=2) | AdamW · QAT ImageNet (8-bit) | accuracy (%): 8.63% -> 10.97% | Improved |
| replace AdamW with S-Adam (k=8) | AdamW · Imagewoof2-160 (BS=2) | accuracy (%): +24.69 points | Improved |
| replace AdamW with S-Adam | AdamW · CIFAR-10 (BS=64) | accuracy (%): 90.62% -> 94.18% | Improved |
| replace AdamW with S-Adam | AdamW · CIFAR-100 (BS=2, lr=5e-4) | accuracy (%): +8.28 points | Improved |
| replace AdamW with S-Adam | AdamW · Imagewoof (BS=64) | accuracy (%): 79.82% -> 84.60% | Mixed |
Closest papers by meaning
- Adam Reduces a Unique Form of Sharpness: Theoretical Insights Near the Minimizer Manifold (NeurIPS 2025 · similarity 0.90)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.90)
- On Convergence of Adam for Stochastic Optimization under Relaxed Assumptions (NeurIPS 2024 · similarity 0.89)
- Understanding the Generalization of Stochastic Gradient Adam in Learning Neural Networks (NeurIPS 2025 · similarity 0.88)
- Adam Exploits $\ell_\infty$-geometry of Loss Landscape via Coordinate-wise Adaptivity (ICLR 2025 · similarity 0.88)
- Adam under Generalized Smoothness with Second-Moment-Type Stochastic Gradients (NeurIPS 2026 · similarity 0.88)
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.
