Research map / Label-efficient and robust learning / Recognition and long-tail
Tailoring: encoding inductive biases by optimizing unsupervised objectives at prediction time
Ferran Alet, Maria Bauza Villalonga, Kenji Kawaguchi, Nurullah Giray Kuru, Tomas Perez, Leslie Pack Kaelbling
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 |
|---|---|---|---|
| meta-tailor randomized smoothing with cosine-feature smoothness loss | Randomized Smoothing · ImageNet sigma=0.25 | ACR: 0.470 -> 0.494 | Improved |
| meta-tailor randomized smoothing with cosine-feature smoothness loss | Randomized Smoothing · ImageNet sigma=0.5 | ACR: 0.720 -> 0.819 | Improved |
| meta-tailor randomized smoothing with cosine-feature smoothness loss | Randomized Smoothing · ImageNet sigma=1.0 | ACR: 0.863 -> 1.032 | Improved |
| meta-tailor randomized smoothing | Randomized Smoothing · CIFAR-10 sigma=0.25/0.5/1.0 | ACR: 8.6%,10.4%,19.2% | Improved |
| meta-tailor the linear layer of SimCLR with a contrastive loss | SimCLR · CIFAR-10 | accuracy: 0.5% and 0.8% | Improved |
| meta-tailoring versus inductive learning with physics tailoring loss | 3-layer MLP · 5-body planetary system | test MSE: 35% | Improved |
| tailor model parameters instead of directly optimizing the output y (ablation) | Physics tailoring loss · 5-body planetary system | test MSE | Improved |
| use first-order instead of second-order CNGrad (ablation) | Meta-tailoring · 5-body planetary system | test MSE | Mixed |
| meta-tailoring with no query-time tailoring steps (ablation) | Meta-tailoring · 5-body planetary system | test MSE | Improved |
| add test-time training (TTT) to SimCLR | SimCLR · CIFAR-10 | accuracy | Hurt |
| meta-tailor MLP with soft energy conservation instead of built-in conservation (HNN) | Vanilla MLP · real pendulum | long-term prediction loss | Improved |
Closest papers by meaning
- How Fine-Tuning Allows for Effective Meta-Learning (NeurIPS 2021 · similarity 0.86)
- Function Contrastive Learning of Transferable Meta-Representations (ICML 2021 · similarity 0.85)
- Inductive biases of multi-task learning and finetuning: multiple regimes of feature reuse (NeurIPS 2024 · similarity 0.85)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.85)
- Training the Untrainable: Introducing Inductive Bias via Representational Alignment (NeurIPS 2025 · similarity 0.85)
- Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization (NeurIPS 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.
