FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3175206595> ?p ?o ?g. }

• W3175206595 abstract "Deep ensembles have recently gained popularity in the deep learning community for their conceptual simplicity and efficiency. However, maintaining functional diversity between ensemble members that are independently trained with gradient descent is challenging. This can lead to pathologies when adding more ensemble members, such as a saturation of the ensemble performance, which converges to the performance of a single model. Moreover, this does not only affect the quality of its predictions, but even more so the uncertainty estimates of the ensemble, and thus its performance on out-of-distribution data. We hypothesize that this limitation can be overcome by discouraging different ensemble members from collapsing to the same function. To this end, we introduce a kernelized repulsive term in the update rule of the deep ensembles. We show that this simple modification not only enforces and maintains diversity among the members but, even more importantly, transforms the maximum a posteriori inference into proper Bayesian inference. Namely, we show that the training dynamics of our proposed repulsive ensembles follow a Wasserstein gradient flow of the KL divergence with the true posterior. We study repulsive terms in weight and function space and empirically compare their performance to standard ensembles and Bayesian baselines on synthetic and real-world prediction tasks." @default.
• W3175206595 created "2021-07-05" @default.
• W3175206595 creator A5010706885 @default.
• W3175206595 creator A5034845812 @default.
• W3175206595 date "2021-06-22" @default.
• W3175206595 modified "2023-09-23" @default.
• W3175206595 title "Repulsive Deep Ensembles are Bayesian" @default.
• W3175206595 cites W1061737 @default.
• W3175206595 cites W1522301498 @default.
• W3175206595 cites W1522579744 @default.
• W3175206595 cites W1567512734 @default.
• W3175206595 cites W1677182931 @default.
• W3175206595 cites W2038117130 @default.
• W3175206595 cites W2071048859 @default.
• W3175206595 cites W2072555316 @default.
• W3175206595 cites W2102955267 @default.
• W3175206595 cites W2111051539 @default.
• W3175206595 cites W2118020555 @default.
• W3175206595 cites W2135293965 @default.
• W3175206595 cites W2139801605 @default.
• W3175206595 cites W2145513251 @default.
• W3175206595 cites W2148571195 @default.
• W3175206595 cites W2167433878 @default.
• W3175206595 cites W2194775991 @default.
• W3175206595 cites W2254249950 @default.
• W3175206595 cites W2335728318 @default.
• W3175206595 cites W2441221256 @default.
• W3175206595 cites W2605488176 @default.
• W3175206595 cites W2612983688 @default.
• W3175206595 cites W2620283749 @default.
• W3175206595 cites W2750384547 @default.
• W3175206595 cites W2766678531 @default.
• W3175206595 cites W2788838181 @default.
• W3175206595 cites W2806584432 @default.
• W3175206595 cites W2809090039 @default.
• W3175206595 cites W2919841361 @default.
• W3175206595 cites W2945024121 @default.
• W3175206595 cites W2949496227 @default.
• W3175206595 cites W2950220847 @default.
• W3175206595 cites W2951266961 @default.
• W3175206595 cites W2952625664 @default.
• W3175206595 cites W2962752165 @default.
• W3175206595 cites W2963177640 @default.
• W3175206595 cites W2963238274 @default.
• W3175206595 cites W2963558557 @default.
• W3175206595 cites W2963693742 @default.
• W3175206595 cites W2963956018 @default.
• W3175206595 cites W2970859221 @default.
• W3175206595 cites W2990020588 @default.
• W3175206595 cites W2992525328 @default.
• W3175206595 cites W2996144997 @default.
• W3175206595 cites W2996184071 @default.
• W3175206595 cites W2996246027 @default.
• W3175206595 cites W3005304850 @default.
• W3175206595 cites W3005861412 @default.
• W3175206595 cites W3006861283 @default.
• W3175206595 cites W3012922053 @default.
• W3175206595 cites W3026026123 @default.
• W3175206595 cites W3034669169 @default.
• W3175206595 cites W3040074755 @default.
• W3175206595 cites W3042158842 @default.
• W3175206595 cites W3096736359 @default.
• W3175206595 cites W3098341014 @default.
• W3175206595 cites W3099149583 @default.
• W3175206595 cites W3102879800 @default.
• W3175206595 cites W3105285613 @default.
• W3175206595 cites W3118608800 @default.
• W3175206595 cites W3119420068 @default.
• W3175206595 cites W3122034064 @default.
• W3175206595 cites W3129383691 @default.
• W3175206595 cites W3130833330 @default.
• W3175206595 cites W3133620575 @default.
• W3175206595 cites W3157086450 @default.
• W3175206595 cites W3160580080 @default.
• W3175206595 cites W3161916718 @default.
• W3175206595 cites W3166859839 @default.
• W3175206595 cites W3167104264 @default.
• W3175206595 cites W3128146547 @default.
• W3175206595 hasPublicationYear "2021" @default.
• W3175206595 type Work @default.
• W3175206595 sameAs 3175206595 @default.
• W3175206595 citedByCount "7" @default.
• W3175206595 countsByYear W31752065952021 @default.
• W3175206595 crossrefType "posted-content" @default.
• W3175206595 hasAuthorship W3175206595A5010706885 @default.
• W3175206595 hasAuthorship W3175206595A5034845812 @default.
• W3175206595 hasConcept C107673813 @default.
• W3175206595 hasConcept C119857082 @default.
• W3175206595 hasConcept C138885662 @default.
• W3175206595 hasConcept C14036430 @default.
• W3175206595 hasConcept C153258448 @default.
• W3175206595 hasConcept C154945302 @default.
• W3175206595 hasConcept C160234255 @default.
• W3175206595 hasConcept C207390915 @default.
• W3175206595 hasConcept C2776214188 @default.
• W3175206595 hasConcept C41008148 @default.
• W3175206595 hasConcept C41895202 @default.
• W3175206595 hasConcept C45942800 @default.
• W3175206595 hasConcept C50644808 @default.
• W3175206595 hasConcept C57830394 @default.

• next