Research map / Learning theory and algorithms / Kernels and regression theory
On the Generalization of Models Trained with SGD: Information-Theoretic Bounds and Implications
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply GMP with k=3 instead of AMP | VGG16 · SVHN | Top-1 classification accuracy (%): 97.27±0.015 -> 97.18±0.057 | Hurt |
| Apply GMP with k=10 instead of AMP (ablation) | VGG16 · SVHN | Top-1 classification accuracy (%): 97.27±0.015 -> 97.09±0.068 | Hurt |
| Apply GMP with k=3 instead of MixUp | VGG16 · CIFAR-10 | Top-1 classification accuracy (%): 94.52±0.112 -> 94.33±0.094 | Hurt |
| Apply GMP with k=10 instead of MixUp (ablation) | VGG16 · CIFAR-10 | Top-1 classification accuracy (%): 94.52±0.112 -> 94.45±0.158 | Hurt |
| Apply GMP with k=3 instead of AMP | VGG16 · CIFAR-100 | Top-1 classification accuracy (%): 74.40±0.168 -> 74.45±0.256 | Helped |
| Apply GMP with k=10 instead of AMP (ablation) | VGG16 · CIFAR-100 | Top-1 classification accuracy (%): 74.40±0.168 -> 75.09±0.285 | Helped |
| Use GMP with k=3 instead of ERM | VGG16 · SVHN | Top-1 classification accuracy (%): 96.86±0.060 -> 97.18±0.057 | Helped |
| Use GMP with k=3 instead of ERM | VGG16 · CIFAR-10 | Top-1 classification accuracy (%): 93.68±0.193 -> 94.33±0.094 | Helped |
| Use GMP with k=10 instead of ERM (ablation) | VGG16 · CIFAR-100 | Top-1 classification accuracy (%): 72.16±0.297 -> 75.09±0.285 | Helped |
| Increase the number of sampled Gaussian noises from k=3 to k=10 (ablation) | GMP on VGG16 · CIFAR-100 | Top-1 classification accuracy (%): 74.45±0.256 -> 75.09±0.285 | Helped |
| Increase the number of sampled Gaussian noises from k=3 to k=10 (ablation) | GMP on VGG16 · SVHN | Top-1 classification accuracy (%): 97.18±0.057 -> 97.09±0.068 | Hurt |
Closest papers by meaning
- Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power (ICLR 2025 · similarity 0.91)
- Generalization Bounds of Stochastic Gradient Descent for Wide and Deep Neural Networks (NeurIPS 2019 · similarity 0.91)
- SGD Learns Over-parameterized Networks that Provably Generalize on Linearly Separable Data (ICLR 2018 · similarity 0.91)
- Generalization Error Bounds of Gradient Descent for Learning Over-Parameterized Deep ReLU Networks (AAAI 2020 · similarity 0.91)
- Fantastic Generalization Measures are Nowhere to be Found (ICLR 2024 · similarity 0.90)
- Information-theoretic generalization bounds for black-box learning algorithms (NeurIPS 2021 · similarity 0.90)
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.