Research map / Label-efficient and robust learning / Recognition and long-tail
Embedding Space Interpolation Beyond Mini-Batch, Beyond Pairs and Beyond Examples
Shashanka Venkataramanan, Ewa Kijak, laurent amsaleg, Yannis Avrithis
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 |
|---|---|---|---|
| Increase generated mixed examples from n=10 to n=1000 (ablation) | MultiMix · CIFAR-100 | accuracy: 92.19 -> 93.43 | Improved |
| Use pairwise mixing (m=2) rather than a full-batch tuple (m=128), keeping n=1000 (ablation) | MultiMix · CIFAR-100 | accuracy: 0.7\% lower accuracy | Hurt |
| Increase the interpolation depth toward the deepest layer (ablation) | MultiMix · CIFAR-100 | accuracy | Improved |
| Increase the number n of generated mixed examples (ablation) | MultiMix and Dense MultiMix · CIFAR-100 | accuracy | Improved |
| Add dense MultiMix | MultiMix · CIFAR-100-C | classification error (CE): 33.8 -> 31.7 | Improved |
| Replace AugMix with MultiMix | image classification · CIFAR100 | accuracy: 79.83 -> 81.81 | Improved |
| Replace AugMix with MultiMix | image classification · ImageNet | accuracy: 77.68 -> 78.81 | Improved |
| Use Input Mixup instead of the baseline | CIFAR-100 classification · CIFAR-100 | ECE: 10.25 -> 18.50 | Hurt |
| Use MultiMix instead of the baseline | CIFAR-100 classification · CIFAR-100 | ECE: 10.25 -> 5.63 | Improved |
| Use Dense MultiMix instead of the baseline | CIFAR-100 classification · CIFAR-100 | OE: 1.11 -> 0.27 | Improved |
| Use MultiMix instead of DAML | Office-Home open-domain classification · Office-Home Clipart | accuracy: 45.13 -> 46.01 | Improved |
| Use Dense MultiMix instead of DAML | Office-Home open-domain classification · Office-Home Product | accuracy: 61.54 -> 62.28 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AugMix — AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty (ICLR 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tailoring Mixup to Data for Calibration (ICLR 2025 · similarity 0.85)
- Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond (AAAI 2024 · similarity 0.84)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.84)
- Inversion Circle Interpolation: Diffusion-based Image Augmentation for Data-scarce Classification (CVPR 2025 · similarity 0.84)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.84)
- Enhance Image Classification via Inter-Class Image Mixup with Diffusion Model (CVPR 2024 · similarity 0.84)
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.
