Research map / Learning theory and algorithms / Prediction and decision losses
Learning nonlinear level sets for dimensionality reduction in function approximation
Guannan Zhang, Jiaxin Zhang, Jacob Hinkle
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 |
|---|---|---|---|
| NLL nonlinear transformation instead of no transformation (brute-force NN), 500 samples | 20+20 NN · f4 | validation relative RMSE: 61.22% -> 13.41% | Improved |
| NLL instead of AS, 500 samples | 20+20 NN · f4 | validation relative RMSE: 65.98% -> 13.41% | Improved |
| NLL instead of SIR, 500 samples | 20+20 NN · f5 | validation relative RMSE: 28.17% -> 13.36% | Improved |
| NLL with small network (1 hidden layer, 10 neurons), 10,000 samples | 1x10 NN · f4 | validation relative RMSE: 30.36% -> 3.11% | Improved |
| NLL with small network, 10,000 samples | 1x10 NN · f5 | validation relative RMSE: 16.32% -> 6.74% | Improved |
| NLL with 100 training samples | 20+20 NN · f4 | validation relative RMSE: 96.74% -> 98.23% | No clear effect |
| NLL with 100 training samples | 20+20 NN · f5 | validation relative RMSE: 40.95% -> 77.79% | No clear effect |
| NLL with 10,000 training samples (large data) | 20+20 NN · f5 | validation relative RMSE: 3.56% -> 3.04% | No clear effect |
| NLL for composite shell, 500 samples | 1x20 NN · composite shell | validation relative RMSE: 67.57% -> 11.96% | Improved |
| NLL for composite shell, 100 samples | 1x20 NN · composite shell | validation relative RMSE: 65.74% -> 63.18% | No clear effect |
| NLL reduces sensitivity of inactive dimension Dim 4 vs AS | NLL · composite shell | relative sensitivity: 0.20 -> 0.011 | Improved |
| NLL vs AS/SIR on sensitivity of inactive dimensions (20D functions) | NLL · f4, f5 | relative sensitivity | Improved |
Closest papers by meaning
- Compress to Impress: Efficient LLM Adaptation Using a Single Gradient Step on 100 Samples (NeurIPS 2025 · similarity 0.84)
- Function-space Parameterization of Neural Networks for Sequential Learning (ICLR 2024 · similarity 0.83)
- Pretrained Transformer Efficiently Learns Low-Dimensional Target Functions In-Context (NeurIPS 2024 · similarity 0.83)
- Maintaining Structural Integrity in Parameter Spaces for Parameter Efficient Fine-tuning (ICLR 2025 · similarity 0.83)
- High-Dimensional Analysis for Generalized Nonlinear Regression: From Asymptotics to Algorithm (AAAI 2024 · similarity 0.83)
- A Riemannian Framework for Learning Reduced-order Lagrangian Dynamics (ICLR 2025 · similarity 0.83)
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.
