Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Understanding the Data Dependency of Mixup-style Training
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 the Mixup Beta parameter from α = 1 to α = 32 | Mixup on X3^2 · X3^2 | classification error on original points | Hurt |
| Increase α from 32 to α = 128 | Mixup on X3^2 · X3^2 | class 1 probability at point 1: 0.603 -> 0.650 | Mixed |
| Use Mixup instead of standard ERM | ResNet-18 · MNIST, CIFAR-10, CIFAR-100 training data | training accuracy: within 1% | No clear effect |
| Train with Mixup rather than standard training | 2-layer network on two moons · two moons, class separation 0.5 | decision-boundary nonlinearity | Helped |
| Use Mixup with α = 1 on the 10-point alternating-line datasets | 2-layer network · X10^2 and X10^10 | original empirical risk | Hurt |
Closest papers by meaning
- Provable Benefit of Mixup for Finding Optimal Decision Boundaries (ICML 2023 · similarity 0.92)
- The Benefits of Mixup for Feature Learning (ICML 2023 · similarity 0.91)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.90)
- Pushing Boundaries: Mixup's Influence on Neural Collapse (ICLR 2024 · similarity 0.87)
- MixMin: Finding Data Mixtures via Convex Minimization (ICML 2025 · similarity 0.85)
- Selective Mixup Fine-Tuning for Optimizing Non-Decomposable Objectives (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.