Research map / Learning theory and algorithms / Kernels and regression theory
Functional Output Regression with Infimal Convolution: Exploring the Huber and $ε$-insensitive Losses
Alex Lambert, Dimitri Bouche, Zoltan Szabo, Florence D’Alché-Buc
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 |
|---|---|---|---|
| Use generalized Huber losses instead of squared loss | Functional output regression · Synthetic data with global outliers | MSE | Improved |
| Use Huber loss with p=1 instead of squared loss | Functional output regression · Synthetic data with local outliers | MSE | Improved |
| Use Huber loss with p=2 instead of squared loss | Functional output regression · Synthetic data with local outliers as contamination approaches τ=1 | MSE | Hurt |
| Use Huber loss with p=1 instead of squared loss | Speech regression · Type 3 local outliers, τ=0.1 and ξ=0.1 | MSE: 7.53±0.58 -> 7.0± 0.59; 5.06±0.51 -> 4.72± 0.54; 5.89±0.43 -> 6.62±0.66 | Mixed |
| Use Huber loss with p=2 instead of squared loss | Speech regression · Type 1 global outliers, τ=0.1 | MSE: 6.71±0.96 -> 6.14±0.97; 4.37±0.56 -> 3.76±0.62; 8.15±0.48 -> 7.96±0.49; 9.4±0.75 -> 9.4±0.66 | Mixed |
| Use ε-insensitive loss to ignore small residuals | Functional output regression · Synthetic data with Gaussian noise | sparsity | Improved |
| Use ε-insensitive loss with p=∞ instead of squared loss | DTI · DTI, λ=10^-5 | sparsity: 85.9±10.7% | Improved |
| Use ε-insensitive loss with p=∞ instead of squared loss | DTI · DTI, λ=10^-3 | MSE (10^-1): 2.18±0.27 -> 2.18±0.28 | No clear effect |
| Use ε-insensitive losses instead of squared loss | Speech regression · Untainted speech data | MSE: 6.58±0.62 -> 6.58±0.62; 4.65±0.55 -> 4.64±0.55; 4.26±0.46 -> 4.26±0.46 | No clear effect |
Closest papers by meaning
- Duality in RKHSs with Infinite Dimensional Outputs: Application to Robust Losses (ICML 2020 · similarity 0.87)
- General Loss Functions Lead to (Approximate) Interpolation in High Dimensions (JMLR 2025 · similarity 0.85)
- Vector-Valued Least-Squares Regression under Output Regularity Assumptions (JMLR 2022 · similarity 0.85)
- Generalized Tensor Decomposition for Understanding Multi-Output Regression under Combinatorial Shifts (NeurIPS 2024 · similarity 0.84)
- Solving Sparse \& High-Dimensional-Output Regression via Compression (NeurIPS 2024 · similarity 0.84)
- Learning with Fitzpatrick Losses (NeurIPS 2024 · 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.
