Matches in SemOpenAlex for { <https://semopenalex.org/work/W4288804536> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W4288804536 abstract "Exemplar-free Class-incremental Learning (CIL) is a challenging problem because rehearsing data from previous phases is strictly prohibited, causing catastrophic forgetting of Deep Neural Networks (DNNs). In this paper, we present iVoro, a holistic framework for CIL, derived from computational geometry. We found Voronoi Diagram (VD), a classical model for space subdivision, is especially powerful for solving the CIL problem, because VD itself can be constructed favorably in an incremental manner -- the newly added sites (classes) will only affect the proximate classes, making the non-contiguous classes hardly forgettable. Further, in order to find a better set of centers for VD construction, we colligate DNN with VD using Power Diagram and show that the VD structure can be optimized by integrating local DNN models using a divide-and-conquer algorithm. Moreover, our VD construction is not restricted to the deep feature space, but is also applicable to multiple intermediate feature spaces, promoting VD to be multi-centered VD (CIVD) that efficiently captures multi-grained features from DNN. Importantly, iVoro is also capable of handling uncertainty-aware test-time Voronoi cell assignment and has exhibited high correlations between geometric uncertainty and predictive accuracy (up to ~0.9). Putting everything together, iVoro achieves up to 25.26%, 37.09%, and 33.21% improvements on CIFAR-100, TinyImageNet, and ImageNet-Subset, respectively, compared to the state-of-the-art non-exemplar CIL approaches. In conclusion, iVoro enables highly accurate, privacy-preserving, and geometrically interpretable CIL that is particularly useful when cross-phase data sharing is forbidden, e.g. in medical applications. Our code is available at https://machunwei.github.io/ivoro." @default.
- W4288804536 created "2022-07-30" @default.
- W4288804536 creator A5037816092 @default.
- W4288804536 creator A5039291107 @default.
- W4288804536 creator A5041163845 @default.
- W4288804536 creator A5044560598 @default.
- W4288804536 creator A5066792926 @default.
- W4288804536 creator A5076488287 @default.
- W4288804536 date "2022-07-28" @default.
- W4288804536 modified "2023-09-24" @default.
- W4288804536 title "Progressive Voronoi Diagram Subdivision: Towards A Holistic Geometric Framework for Exemplar-free Class-Incremental Learning" @default.
- W4288804536 doi "https://doi.org/10.48550/arxiv.2207.14202" @default.
- W4288804536 hasPublicationYear "2022" @default.
- W4288804536 type Work @default.
- W4288804536 citedByCount "0" @default.
- W4288804536 crossrefType "posted-content" @default.
- W4288804536 hasAuthorship W4288804536A5037816092 @default.
- W4288804536 hasAuthorship W4288804536A5039291107 @default.
- W4288804536 hasAuthorship W4288804536A5041163845 @default.
- W4288804536 hasAuthorship W4288804536A5044560598 @default.
- W4288804536 hasAuthorship W4288804536A5066792926 @default.
- W4288804536 hasAuthorship W4288804536A5076488287 @default.
- W4288804536 hasBestOaLocation W42888045361 @default.
- W4288804536 hasConcept C111919701 @default.
- W4288804536 hasConcept C11413529 @default.
- W4288804536 hasConcept C138885662 @default.
- W4288804536 hasConcept C142362112 @default.
- W4288804536 hasConcept C143392562 @default.
- W4288804536 hasConcept C153349607 @default.
- W4288804536 hasConcept C154945302 @default.
- W4288804536 hasConcept C166957645 @default.
- W4288804536 hasConcept C177264268 @default.
- W4288804536 hasConcept C199360897 @default.
- W4288804536 hasConcept C205649164 @default.
- W4288804536 hasConcept C24881265 @default.
- W4288804536 hasConcept C2524010 @default.
- W4288804536 hasConcept C2776401178 @default.
- W4288804536 hasConcept C2777212361 @default.
- W4288804536 hasConcept C2778572836 @default.
- W4288804536 hasConcept C2910391586 @default.
- W4288804536 hasConcept C33923547 @default.
- W4288804536 hasConcept C41008148 @default.
- W4288804536 hasConcept C41895202 @default.
- W4288804536 hasConcept C7149132 @default.
- W4288804536 hasConcept C71559656 @default.
- W4288804536 hasConcept C80444323 @default.
- W4288804536 hasConceptScore W4288804536C111919701 @default.
- W4288804536 hasConceptScore W4288804536C11413529 @default.
- W4288804536 hasConceptScore W4288804536C138885662 @default.
- W4288804536 hasConceptScore W4288804536C142362112 @default.
- W4288804536 hasConceptScore W4288804536C143392562 @default.
- W4288804536 hasConceptScore W4288804536C153349607 @default.
- W4288804536 hasConceptScore W4288804536C154945302 @default.
- W4288804536 hasConceptScore W4288804536C166957645 @default.
- W4288804536 hasConceptScore W4288804536C177264268 @default.
- W4288804536 hasConceptScore W4288804536C199360897 @default.
- W4288804536 hasConceptScore W4288804536C205649164 @default.
- W4288804536 hasConceptScore W4288804536C24881265 @default.
- W4288804536 hasConceptScore W4288804536C2524010 @default.
- W4288804536 hasConceptScore W4288804536C2776401178 @default.
- W4288804536 hasConceptScore W4288804536C2777212361 @default.
- W4288804536 hasConceptScore W4288804536C2778572836 @default.
- W4288804536 hasConceptScore W4288804536C2910391586 @default.
- W4288804536 hasConceptScore W4288804536C33923547 @default.
- W4288804536 hasConceptScore W4288804536C41008148 @default.
- W4288804536 hasConceptScore W4288804536C41895202 @default.
- W4288804536 hasConceptScore W4288804536C7149132 @default.
- W4288804536 hasConceptScore W4288804536C71559656 @default.
- W4288804536 hasConceptScore W4288804536C80444323 @default.
- W4288804536 hasLocation W42888045361 @default.
- W4288804536 hasOpenAccess W4288804536 @default.
- W4288804536 hasPrimaryLocation W42888045361 @default.
- W4288804536 hasRelatedWork W124579 @default.
- W4288804536 hasRelatedWork W12954281 @default.
- W4288804536 hasRelatedWork W14268380 @default.
- W4288804536 hasRelatedWork W29332 @default.
- W4288804536 hasRelatedWork W5374421 @default.
- W4288804536 hasRelatedWork W5650064 @default.
- W4288804536 hasRelatedWork W7073430 @default.
- W4288804536 hasRelatedWork W7491381 @default.
- W4288804536 hasRelatedWork W8060374 @default.
- W4288804536 hasRelatedWork W8565002 @default.
- W4288804536 isParatext "false" @default.
- W4288804536 isRetracted "false" @default.
- W4288804536 workType "article" @default.