Research map / Learning theory and algorithms / Kernels and regression theory
Diffeomorphism-based feature learning using Poincaré inequalities on augmented input space
Romain Verdière, Clémentine Prieur, Olivier Zahm
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 dimension augmentation (extra random inputs) (ablation) | BACF_IS · u2 on [0,1]^20 (ntrain=100) | NRMSE (%): 9.11 -> 4.29 | Improved |
| add dimension augmentation (ablation) | BACF_IS · u2 on [-1,1]^8 (ntrain=500) | NRMSE (%): 8.05 -> 2.47 | Improved |
| add dimension augmentation (ablation) | BACF_IS · u3 on [-pi/2,pi/2]^8 (ntrain=100) | NRMSE (%): 4.75 -> 1.40 | Improved |
| add dimension augmentation (ablation) | BACF_IS · u3 on [-pi/2,pi/2]^12 (ntrain=100) | NRMSE (%): 6.48 -> 3.26 | Improved |
| add dimension augmentation (ablation) | BACF_IS · Thermal block d=16, m=1 (ntrain=100) | NRMSE (%): 5.18 -> 3.73 | Improved |
| add dimension augmentation (ablation) | BACF_IS · u1 on [-1,1]^8 (ntrain=50) | NRMSE (%): 1.15 -> 1.41 | Hurt |
| add dimension augmentation (ablation) | BACF_IS · Thermal block d=36, m=1 (ntrain=200) | NRMSE (%): 3.44 -> 3.51 | No clear effect |
| add dimension augmentation (ablation) | BACF_FS · Thermal block d=16, m=1 (ntrain=100) | NRMSE (%): 5.27 -> 7.54 | Hurt |
| add dimension augmentation (ablation) | BACF_FS · u3 on [-pi/2,pi/2]^12 (ntrain=500) | NRMSE (%): 1.55 -> 1.28 | Improved |
| use input-space Poincare bound (IS) instead of feature-space bound (FS) (ablation) | BACF · u3 on [-pi/2,pi/2]^8 (ntrain=100) | NRMSE (%): 12.40 -> 4.75 | Improved |
| use BACF_ISd+2 instead of PFM | feature map · u3 on [-pi/2,pi/2]^8 (ntrain=500) | NRMSE (%): 12.08 -> 0.56 | Improved |
| use PFM instead of BACF on a polynomial-friendly target | feature map · u2 on [0,1]^20 (ntrain=500) | NRMSE (%): 1.81 -> 0.39 | Improved |
Closest papers by meaning
- Diffeomorphic Learning (JMLR 2020 · similarity 0.88)
- High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representation (NeurIPS 2022 · similarity 0.86)
- FineMorphs: Affine-Diffeomorphic Sequences for Regression (JMLR 2024 · similarity 0.86)
- Three Mechanisms of Feature Learning in a Linear Network (ICLR 2025 · similarity 0.84)
- Alternating Gradient Flows: A Theory of Feature Learning in Two-layer Neural Networks (NeurIPS 2025 · similarity 0.84)
- Sculpting Features from Noise: Reward-Guided Hierarchical Diffusion for Task-Optimal Feature Transformation (NeurIPS 2025 · 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.
