Research map / Learning theory and algorithms / Kernels and regression theory
Split-kl and PAC-Bayes-split-kl Inequalities for Ternary Random Variables
Yi-Shan Wu, Yevgeny Seldin
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 split-kl instead of kl inequality for ternary variables as p0 grows toward 1 (symmetric case p1=p-1) | split-kl inequality · Simulated ternary variables, n samples, delta=0.05 | bound minus empirical mean (lower is tighter) | Improved |
| use split-kl instead of Unexpected Bernstein for skewed ternary variable (p1=0.99(1-p0), p-1=0.01(1-p0)) | split-kl inequality · Simulated skewed ternary variables | bound minus empirical mean | Improved |
| use Empirical Bernstein instead of other bounds on ternary variables | Empirical Bernstein inequality · Simulated ternary variables | bound tightness | Hurt |
| increase sample size to n=1000 in ternary simulations | Empirical Bernstein vs Unexpected Bernstein · Simulated ternary variables | bound tightness | Mixed |
| use split-kl on Beta-distributed bounded variables with large variance | split-kl inequality · Simulated bounded variables, alpha=beta, p=0.5 | bound tightness | No clear effect |
| use informed priors with excess losses and PAC-Bayes-split-kl vs PAC-Bayes-kl and PAC-Bayes-Unexpected-Bernstein | PAC-Bayes-split-kl · Linear classification, UCI/LibSVM datasets | generalization bound | No clear effect |
| use CCPBSkl and CCPBUB instead of CCPBB for tandem loss with offset | weighted majority vote bound · Heterogeneous classifier ensembles, UCI datasets | bound value | Improved |
| use CCPBSkl/CCPBUB instead of TND/CCTND (tandem loss without offset) | weighted majority vote bound · Heterogeneous classifier ensembles, UCI datasets | bound value | No clear effect |
| apply bounds to reweighting trees in a random forest instead of heterogeneous classifiers | CCPBSkl and CCPBUB · Random forest, UCI datasets | bound value | No clear effect |
Closest papers by meaning
- Concentration Inequalities for General Functions of Heavy-Tailed Random Variables (ICML 2024 · similarity 0.86)
- PAC-Bayes-Chernoff bounds for unbounded losses (NeurIPS 2024 · similarity 0.85)
- Sharp Matrix Empirical Bernstein Inequalities (NeurIPS 2025 · similarity 0.85)
- Intrinsic-dimension empirical Bernstein inequalities for bounded self-adjoint operators (NeurIPS 2026 · similarity 0.84)
- Statistical and Geometrical properties of the Kernel Kullback-Leibler divergence (NeurIPS 2024 · similarity 0.84)
- Instance-Optimality for Private KL Distribution Estimation (NeurIPS 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.
