Research map / Learning theory and algorithms / Kernels and regression theory
Exploring Generalization in Deep Learning
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 |
|---|---|---|---|
| Increase hidden-unit count beyond 32 | Two-layer ReLU perceptron · MNIST | test error | Helped |
| Use sharpness alone as a generalization measure | VGG network · CIFAR10, true versus random labels | generalization behavior prediction | Mixed |
| Combine expected sharpness with KL divergence or norm | PAC-Bayes measure · CIFAR10, true versus random labels | generalization behavior prediction | Helped |
| Use norm divided by margin instead of raw norm | Norm-based complexity measures · Deep networks trained to zero training error | scale-invariant complexity assessment | Helped |
| Use Lipschitz continuity alone for capacity control | Feedforward neural networks · Generalization bounds | capacity control | Hurt |
| Increase the size of the random-label confusion set | VGG network · CIFAR10 | test error | Hurt |
| Increase hidden-unit count beyond 128 | Two-layer ReLU perceptron · MNIST | ability of joint PAC-Bayes measure to explain generalization | Hurt |
| Compare raw parameter norms without accounting for output scale | Networks trained with cross-entropy to zero training error · Generalization assessment | meaningfulness of norm comparison | Hurt |
| Use norm-based measures on true labels rather than random labels | Feedforward networks · True versus random labels | measured complexity | Helped |
Closest papers by meaning
- Fantastic Generalization Measures and Where to Find Them (ICLR 2020 · similarity 0.90)
- Deep learning generalizes because the parameter-function map is biased towards simple functions (ICLR 2019 · similarity 0.90)
- Position: Deep Learning is Not So Mysterious or Different (ICML 2025 · similarity 0.89)
- A Modern Look at the Relationship between Sharpness and Generalization (ICML 2023 · similarity 0.89)
- In search of robust measures of generalization (NeurIPS 2020 · similarity 0.89)
- Generalization Measures for Deep Learning Should Be Audited for Fragility (NeurIPS 2026 · similarity 0.89)
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.