Research map / Learning theory and algorithms / Kernels and regression theory
Tighter Information-Theoretic Generalization Bounds from Supersamples
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 |
|---|---|---|---|
| Replace the loss pair with its difference in the information term | Supersample information-theoretic generalization bounds · Supersample setting | generalization-bound tightness | Helped |
| Use the disintegrated loss-difference CMI bound | Supersample generalization bounds · Binary classification with separable synthetic Gaussian data | bound tightness | Helped |
| Use the unconditional loss-difference MI bound | Supersample generalization bounds · Binary classification with non-separable synthetic Gaussian data | bound tightness | Helped |
| Use the variance-based MI bound | Supersample generalization bounds · Binary classification with non-separable synthetic Gaussian data | bound tightness | Helped |
| Use the sharpness-based MI bound | Supersample generalization bounds · Ten-class non-separable synthetic classification | bound tightness | Helped |
| Use the weighted fast-rate bound | Supersample generalization bounds · Separable synthetic classification | bound tightness | Helped |
| Use the weighted fast-rate bound | Supersample generalization bounds · Non-separable binary synthetic classification | bound tightness | Hurt |
| Use the sharpness or variance bound | Supersample generalization bounds · Binary MNIST (4 vs 9) | bound tightness | Helped |
| Use the sharpness or variance bound | Supersample generalization bounds · CIFAR10 with fine-tuned ResNet-50 | bound tightness | Helped |
| Use the weighted fast-rate bound | Supersample generalization bounds · Binary MNIST (4 vs 9) | bound tightness | Hurt |
| Use the sharpness bound during early SGLD training | Supersample generalization bounds · SGLD on binary MNIST | bound tightness | Mixed |
| Use the square-root bounds rather than the binary KL bound | Supersample generalization bounds · Non-separable binary synthetic classification | bound tightness | Helped |
Closest papers by meaning
- Exactly Tight Information-theoretic Generalization Bounds via Binary Jensen-Shannon Divergence (ICML 2025 · similarity 0.91)
- Towards Generalization beyond Pointwise Learning: A Unified Information-theoretic Perspective (ICML 2024 · similarity 0.90)
- Generalization Bounds via Conditional $f$-Information (NeurIPS 2024 · similarity 0.90)
- Tighter CMI-Based Generalization Bounds via Stochastic Projection and Quantization (NeurIPS 2025 · similarity 0.89)
- Slicing Mutual Information Generalization Bounds for Neural Networks (ICML 2024 · similarity 0.89)
- Algorithmic Stability Unleashed: Generalization Bounds with Unbounded Losses (ICML 2024 · 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.