Research map / Label-efficient and robust learning / Recognition and long-tail
Mixing Configurations for Downstream Prediction
Juntang Wang, Hao Wu, Yihan Wang, Dongmian Zou, Shixin Xu
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 |
|---|---|---|---|
| add energy statistics to the selector (ablation) | MixConfig · CIFAR-100 | Top-1 accuracy: .738 -> .758 | Improved |
| add energy statistics to the selector (ablation) | MixConfig · BBBP | accuracy: .882 -> .903 | Improved |
| add sample context h to the selector (ablation) | MixConfig · CIFAR-100 | Top-1 accuracy: .747 -> .758 | Improved |
| add cluster embeddings to the selector (ablation) | MixConfig · CIFAR-100 | Top-1 accuracy: .743 -> .758 | Improved |
| use all m configurations instead of only 2 (coarsest + finest) (ablation) | MixConfig · CIFAR-100 | Top-1 accuracy: .716 -> .758 | Improved |
| use 6 vs all configurations (ablation) | MixConfig · BBBP | accuracy: .895 -> .903 | Improved |
| replace configurations with random partitions of same dimensionality (ablation) | MixConfig features · CIFAR-100 | Top-1 accuracy: .742 -> .737 | Hurt |
| learn per-sample adaptive weights instead of global learnable weights (ablation) | MixConfig · BBBP | accuracy: .878 -> .903 | Improved |
| use the energy-aware selector instead of a stronger 2-layer MLP fusion head (ablation) | configuration features · BBBP | accuracy: .884 -> .903 | Improved |
| add CSPA consensus clustering (ablation) | Base · CIFAR-100 | Top-1 accuracy: .742 -> .743 | No clear effect |
| add DeepCluster single-resolution k-means features | Base · OpenML-CC18 multi-class, CIFAR-100, MolHIV | accuracy / AUC: .782 -> .779 (multi-class) | Hurt |
| replace linear probe with 2-layer MLP probe (ablation) | CLIP on CIFAR-100 · CIFAR-100 | Top-1 accuracy: .742 -> .753 | Improved |
Closest papers by meaning
- MixMin: Finding Data Mixtures via Convex Minimization (ICML 2025 · similarity 0.85)
- Understanding Fixed Predictions via Confined Regions (ICML 2025 · similarity 0.84)
- Few-shot Adaptation to Distribution Shifts By Mixing Source and Target Embeddings (ICML 2024 · similarity 0.83)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.83)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.82)
- Fantastic Features and Where to Find Them: A Probing Method to combine Features from Multiple Foundation Models (NeurIPS 2025 · similarity 0.82)
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.
