Research map / Learning theory and algorithms / Sample complexity
On Tail Decay Rate Estimation of Loss Function Distributions
Etrit Haxholli, Marco Lorenzi
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 |
|---|---|---|---|
| estimate tail shape per conditional distribution and take the maximum (cross tail estimation) instead of direct POT on pooled samples | Pickands/DEdH tail estimators · Synthetic mixtures with highly varying locations | shape parameter estimation | Improved |
| use CTE instead of direct POT for loss tails of overfitting Gaussian process models | Gaussian process regression · UCR anomaly time series | estimated tail shape vs overfitting | Improved |
| use CTE instead of direct POT for polynomial kernels of degree 1 to 9 | Polynomial kernel regression · UCR anomaly time series | estimated tail shape vs overfitting | Improved |
| increase pooled sample size M from 1e5 to 1e8 for marginal tail estimation | Pickands/DEdH estimators · Synthetic mixtures, zero location | shape parameter estimation | Improved |
| vary the POT threshold percentile (99.7 vs 99.997) (ablation) | CTE and POT on GP experiment · UCR anomaly time series | estimated tail shape | No clear effect |
Closest papers by meaning
- Heavy-Tailed Diffusion Models (ICLR 2025 · similarity 0.84)
- Kernel conditional tests from learning-theoretic bounds (NeurIPS 2025 · similarity 0.84)
- Heavy-Tailed Diffusion with Denoising Levy Probabilistic Models (ICLR 2025 · similarity 0.83)
- Rethinking Benign Overfitting in Two-Layer Neural Networks (ICML 2025 · similarity 0.83)
- Flexible Tails for Normalizing Flows (ICML 2025 · similarity 0.83)
- Emergence of heavy tails in homogenized stochastic gradient descent (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.
