Research map / Learning theory and algorithms / Kernels and regression theory
The Teaching Dimension of Regularized Kernel Learners
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Adding L2 regularization to the linear kernel learner | linear kernel learner · exact teaching, theory | teaching dimension: Θ(d) to 1 | Helped |
| Adding L2 regularization to the polynomial kernel learner | polynomial kernel learner · exact teaching, theory | teaching dimension: TD strictly smaller than C(d+p-1,p) for some targets | Helped |
| Adding L2 regularization to the Gaussian kernel learner | Gaussian kernel learner · approximate teaching, epsilon-TD | epsilon-approximate teaching dimension: d^O(log2(1/ϵ)) to O(1/ϵ²) | Helped |
| Regularization for Gaussian kernel learner | Gaussian kernel learner · exact teaching, theory | teaching dimension | No clear effect |
| Adding L2 regularization to the linear learner in exact teaching (ablation) | linear kernel learner · R^3 target, square and hinge loss | teaching set size: from 3 to 1 (square), 5 to 1 (hinge) | Helped |
| Adding L2 regularization to the polynomial learner in exact teaching (ablation) | polynomial kernel learner · given target, square and hinge loss | teaching set size: from 4 to 2 (square loss); unregularized cannot be taught for hinge loss | Helped |
| Adding L2 regularization in approximate teaching (Gaussian kernel) (ablation) | Gaussian kernel learner on regression and classification datasets · epsilon-TD across Sin, MR, MPG, Eunite, Circle, Moon, Adult, Sonar | difference in epsilon-TD vs unregularized: all differences positive where both reach threshold; unregularized fails on Sin, Eunite, Circle, Adult, Sonar at several | Helped |
| Increasing subset size m of Nystrom-approximated H_m (Gaussian kernel) | H_m approximation of RKHS · Sin, MR, MPG, Eunite, Circle, Moon, Adult, Sonar | smallest m to reach excess risk ratio: Sin: m=1 at 100% ratio, m=8 at 0%; Adult: m=9 at 100%, >400 at 40% | Helped |
| Adding L2 regularization on Sonar/Adult classification | Gaussian kernel learner · Adult and Sonar | epsilon-TD difference vs unregularized: Adult unregularized fails at most ratios; Sonar fails at 0% for regularized too | Mixed |
Closest papers by meaning
- The Teaching Dimension of Linear Learners (ICML 2016 · similarity 0.90)
- How a Student becomes a Teacher: learning and forgetting through Spectral methods (NeurIPS 2023 · similarity 0.86)
- Iterative Machine Teaching (ICML 2017 · similarity 0.86)
- Preference-Based Batch and Sequential Teaching: Towards a Unified View of Models (NeurIPS 2019 · similarity 0.86)
- Teaching with Limited Information on the Learner’s Behaviour (ICML 2020 · similarity 0.85)
- Learning from Teaching Regularization: Generalizable Correlations Should be Easy to Imitate (NeurIPS 2024 · similarity 0.85)
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.