Research map / Learning theory and algorithms / Matrix and tensor methods
Beyond Tikhonov: faster learning with self-concordant losses, via iterative regularization
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 |
|---|---|---|---|
| increase iterations t from 1 to 3 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=3.25 | learning rate gamma: 0.73 -> 0.83 | Helped |
| increase iterations t from 1 to 8 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=3.25 | learning rate gamma: 0.73 -> 0.95 | Helped |
| increase iterations t from 1 to 3 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=10.25 | learning rate gamma: 0.72 -> 0.87 | Helped |
| increase iterations t from 1 to 8 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=10.25 | learning rate gamma: 0.72 -> 0.98 | Helped |
| increase iterations t from 1 to 3 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=0.25 (hard task) | learning rate gamma: 0.71 -> 0.75 | No clear effect |
| increase iterations t from 1 to 8 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=0.25 (hard task) | learning rate gamma: 0.71 -> 0.79 | No clear effect |
| increase t from 3 to 8 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=3.25 | learning rate gamma: 0.83 -> 0.95 | Helped |
| increase t from 3 to 8 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=10.25 | learning rate gamma: 0.87 -> 0.98 | Helped |
| use Tikhonov (t=1) on easier task, r from 0.25 to 10.25 (ablation) | Iterated Tikhonov (logistic loss) · Synthetic | learning rate gamma: 0.71 -> 0.72 | No clear effect |
| compare iterated Tikhonov with Tikhonov in absolute excess risk on hard task (ablation) | Iterated Tikhonov (logistic loss) · Synthetic, r=1/4 | learning rate gamma | Mixed |
Closest papers by meaning
- Globally Convergent Newton Methods for Ill-conditioned Generalized Self-concordant Losses (NeurIPS 2019 · similarity 0.88)
- Optimal Rates for Vector-Valued Spectral Regularization Learning Algorithms (NeurIPS 2024 · similarity 0.88)
- Iterative Regularization with k-support Norm: An Important Complement to Sparse Recovery (AAAI 2024 · similarity 0.86)
- Controlling the Flow: Stability and Convergence for Stochastic Gradient Descent with Decaying Regularization (NeurIPS 2025 · similarity 0.85)
- Learning with Fitzpatrick Losses (NeurIPS 2024 · similarity 0.85)
- Generalization error of spectral algorithms (ICLR 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.