Matches in SemOpenAlex for { <https://semopenalex.org/work/W3049559114> ?p ?o ?g. }
Showing items 1 to 89 of
89
with 100 items per page.
- W3049559114 abstract "We investigate optimal posteriors for recently introduced cite{begin2016pac} chi-squared divergence based PAC-Bayesian bounds in terms of nature of their distribution, scalability of computations, and test set performance. For a finite classifier set, we deduce bounds for three distance functions: KL-divergence, linear and squared distances. Optimal posterior weights are proportional to deviations of empirical risks, usually with subset support. For uniform prior, it is sufficient to search among posteriors on classifier subsets ordered by these risks. We show the bound minimization for linear distance as a convex program and obtain a closed-form expression for its optimal posterior. Whereas that for squared distance is a quasi-convex program under a specific condition, and the one for KL-divergence is non-convex optimization (a difference of convex functions). To compute such optimal posteriors, we derive fast converging fixed point (FP) equations. We apply these approaches to a finite set of SVM regularization parameter values to yield stochastic SVMs with tight bounds. We perform a comprehensive performance comparison between our optimal posteriors and known KL-divergence based posteriors on a variety of UCI datasets with varying ranges and variances in risk values, etc. Chi-squared divergence based posteriors have weaker bounds and worse test errors, hinting at an underlying regularization by KL-divergence based posteriors. Our study highlights the impact of divergence function on the performance of PAC-Bayesian classifiers. We compare our stochastic classifiers with cross-validation based deterministic classifier. The latter has better test errors, but ours is more sample robust, has quantifiable generalization guarantees, and is computationally much faster." @default.
- W3049559114 created "2020-08-21" @default.
- W3049559114 creator A5007214077 @default.
- W3049559114 creator A5084320426 @default.
- W3049559114 date "2020-08-14" @default.
- W3049559114 modified "2023-09-27" @default.
- W3049559114 title "Optimal Posteriors for Chi-squared Divergence based PAC-Bayesian Bounds and Comparison with KL-divergence based Optimal Posteriors and Cross-Validation Procedure" @default.
- W3049559114 cites W1622633983 @default.
- W3049559114 cites W1628829797 @default.
- W3049559114 cites W171835870 @default.
- W3049559114 cites W2014384147 @default.
- W3049559114 cites W2026653933 @default.
- W3049559114 cites W2029029543 @default.
- W3049559114 cites W2112065930 @default.
- W3049559114 cites W2116314036 @default.
- W3049559114 cites W2116942894 @default.
- W3049559114 cites W2123871098 @default.
- W3049559114 cites W2130464645 @default.
- W3049559114 cites W2149991487 @default.
- W3049559114 cites W2170207925 @default.
- W3049559114 cites W2221524898 @default.
- W3049559114 cites W2296319761 @default.
- W3049559114 cites W2329283878 @default.
- W3049559114 cites W2412885812 @default.
- W3049559114 cites W2787523326 @default.
- W3049559114 cites W2794149522 @default.
- W3049559114 cites W2964276654 @default.
- W3049559114 cites W2989823397 @default.
- W3049559114 hasPublicationYear "2020" @default.
- W3049559114 type Work @default.
- W3049559114 sameAs 3049559114 @default.
- W3049559114 citedByCount "1" @default.
- W3049559114 countsByYear W30495591142020 @default.
- W3049559114 crossrefType "posted-content" @default.
- W3049559114 hasAuthorship W3049559114A5007214077 @default.
- W3049559114 hasAuthorship W3049559114A5084320426 @default.
- W3049559114 hasConcept C105795698 @default.
- W3049559114 hasConcept C107673813 @default.
- W3049559114 hasConcept C11413529 @default.
- W3049559114 hasConcept C126255220 @default.
- W3049559114 hasConcept C138885662 @default.
- W3049559114 hasConcept C139945424 @default.
- W3049559114 hasConcept C154945302 @default.
- W3049559114 hasConcept C207390915 @default.
- W3049559114 hasConcept C2776135515 @default.
- W3049559114 hasConcept C28826006 @default.
- W3049559114 hasConcept C33923547 @default.
- W3049559114 hasConcept C41008148 @default.
- W3049559114 hasConcept C41895202 @default.
- W3049559114 hasConceptScore W3049559114C105795698 @default.
- W3049559114 hasConceptScore W3049559114C107673813 @default.
- W3049559114 hasConceptScore W3049559114C11413529 @default.
- W3049559114 hasConceptScore W3049559114C126255220 @default.
- W3049559114 hasConceptScore W3049559114C138885662 @default.
- W3049559114 hasConceptScore W3049559114C139945424 @default.
- W3049559114 hasConceptScore W3049559114C154945302 @default.
- W3049559114 hasConceptScore W3049559114C207390915 @default.
- W3049559114 hasConceptScore W3049559114C2776135515 @default.
- W3049559114 hasConceptScore W3049559114C28826006 @default.
- W3049559114 hasConceptScore W3049559114C33923547 @default.
- W3049559114 hasConceptScore W3049559114C41008148 @default.
- W3049559114 hasConceptScore W3049559114C41895202 @default.
- W3049559114 hasLocation W30495591141 @default.
- W3049559114 hasOpenAccess W3049559114 @default.
- W3049559114 hasPrimaryLocation W30495591141 @default.
- W3049559114 hasRelatedWork W1679170311 @default.
- W3049559114 hasRelatedWork W1830254893 @default.
- W3049559114 hasRelatedWork W2037199905 @default.
- W3049559114 hasRelatedWork W2087109949 @default.
- W3049559114 hasRelatedWork W2121131150 @default.
- W3049559114 hasRelatedWork W2130251486 @default.
- W3049559114 hasRelatedWork W2159280716 @default.
- W3049559114 hasRelatedWork W2166944917 @default.
- W3049559114 hasRelatedWork W2293221180 @default.
- W3049559114 hasRelatedWork W2339457533 @default.
- W3049559114 hasRelatedWork W2752903275 @default.
- W3049559114 hasRelatedWork W2774039628 @default.
- W3049559114 hasRelatedWork W2775216473 @default.
- W3049559114 hasRelatedWork W2817027724 @default.
- W3049559114 hasRelatedWork W2894562391 @default.
- W3049559114 hasRelatedWork W2947390487 @default.
- W3049559114 hasRelatedWork W2963885708 @default.
- W3049559114 hasRelatedWork W2971969744 @default.
- W3049559114 hasRelatedWork W3008049397 @default.
- W3049559114 hasRelatedWork W3103576036 @default.
- W3049559114 isParatext "false" @default.
- W3049559114 isRetracted "false" @default.
- W3049559114 magId "3049559114" @default.
- W3049559114 workType "article" @default.