FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3117301244> ?p ?o ?g. }

• W3117301244 abstract "Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared to general linear maps. However, choosing which of the myriad structured transformations to use (and its associated parameterization) is a laborious task that requires trading off speed, space, and accuracy. We consider a different approach: we introduce a family of matrices called kaleidoscope matrices (K-matrices) that provably capture any structured matrix with near-optimal space (parameter) and time (arithmetic operation) complexity. We empirically validate that K-matrices can be automatically learned within end-to-end pipelines to replace hand-crafted procedures, in order to improve model quality. For example, replacing channel shuffles in ShuffleNet improves classification accuracy on ImageNet by up to 5%. K-matrices can also simplify hand-engineered pipelines -- we replace filter bank feature computation in speech data preprocessing with a learnable kaleidoscope layer, resulting in only 0.4% loss in accuracy on the TIMIT speech recognition task. In addition, K-matrices can capture latent structure in models: for a challenging permuted image classification task, a K-matrix based representation of permutations is able to learn the right latent structure and improves accuracy of a downstream convolutional model by over 9%. We provide a practically efficient implementation of our approach, and use K-matrices in a Transformer network to attain 36% faster end-to-end inference speed on a language translation task." @default.
• W3117301244 created "2021-01-05" @default.
• W3117301244 creator A5001041485 @default.
• W3117301244 creator A5013698287 @default.
• W3117301244 creator A5025386668 @default.
• W3117301244 creator A5046328705 @default.
• W3117301244 creator A5069499364 @default.
• W3117301244 creator A5070267604 @default.
• W3117301244 creator A5070318397 @default.
• W3117301244 creator A5091734792 @default.
• W3117301244 date "2020-12-29" @default.
• W3117301244 modified "2023-09-27" @default.
• W3117301244 title "Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps" @default.
• W3117301244 cites W1489355995 @default.
• W3117301244 cites W1524333225 @default.
• W3117301244 cites W1561337879 @default.
• W3117301244 cites W175700287 @default.
• W3117301244 cites W1800356822 @default.
• W3117301244 cites W1814624729 @default.
• W3117301244 cites W1965960902 @default.
• W3117301244 cites W2007988810 @default.
• W3117301244 cites W2010961009 @default.
• W3117301244 cites W2037221767 @default.
• W3117301244 cites W2047769394 @default.
• W3117301244 cites W2058641082 @default.
• W3117301244 cites W2092590148 @default.
• W3117301244 cites W2103559027 @default.
• W3117301244 cites W2103869314 @default.
• W3117301244 cites W2111460811 @default.
• W3117301244 cites W2115881141 @default.
• W3117301244 cites W2132105090 @default.
• W3117301244 cites W2134316164 @default.
• W3117301244 cites W2135461721 @default.
• W3117301244 cites W2138106543 @default.
• W3117301244 cites W2149599446 @default.
• W3117301244 cites W2157331557 @default.
• W3117301244 cites W2158426391 @default.
• W3117301244 cites W2163605009 @default.
• W3117301244 cites W2182629226 @default.
• W3117301244 cites W2194775991 @default.
• W3117301244 cites W2228229434 @default.
• W3117301244 cites W2398826216 @default.
• W3117301244 cites W2405098100 @default.
• W3117301244 cites W2508048623 @default.
• W3117301244 cites W2520160253 @default.
• W3117301244 cites W2544176167 @default.
• W3117301244 cites W2547718140 @default.
• W3117301244 cites W2549139847 @default.
• W3117301244 cites W2569750427 @default.
• W3117301244 cites W2581719241 @default.
• W3117301244 cites W2596625124 @default.
• W3117301244 cites W2754084392 @default.
• W3117301244 cites W2764043458 @default.
• W3117301244 cites W2769802733 @default.
• W3117301244 cites W2792764867 @default.
• W3117301244 cites W2794209590 @default.
• W3117301244 cites W2899771611 @default.
• W3117301244 cites W2907886210 @default.
• W3117301244 cites W2908336025 @default.
• W3117301244 cites W2913637113 @default.
• W3117301244 cites W2933138175 @default.
• W3117301244 cites W2945461886 @default.
• W3117301244 cites W2949382160 @default.
• W3117301244 cites W2956434358 @default.
• W3117301244 cites W2962697559 @default.
• W3117301244 cites W2962901777 @default.
• W3117301244 cites W2962949934 @default.
• W3117301244 cites W2963042606 @default.
• W3117301244 cites W2963061092 @default.
• W3117301244 cites W2963125010 @default.
• W3117301244 cites W2963175699 @default.
• W3117301244 cites W2963403868 @default.
• W3117301244 cites W2963471276 @default.
• W3117301244 cites W2963813662 @default.
• W3117301244 cites W2963815651 @default.
• W3117301244 cites W2963992014 @default.
• W3117301244 cites W2964052309 @default.
• W3117301244 cites W2964156591 @default.
• W3117301244 cites W2964196118 @default.
• W3117301244 cites W2964227577 @default.
• W3117301244 cites W2964299589 @default.
• W3117301244 cites W2964316651 @default.
• W3117301244 cites W2964348070 @default.
• W3117301244 cites W2972442919 @default.
• W3117301244 cites W2981014499 @default.
• W3117301244 cites W2990844796 @default.
• W3117301244 cites W3022720367 @default.
• W3117301244 cites W3098543800 @default.
• W3117301244 cites W3100023701 @default.
• W3117301244 cites W3101584733 @default.
• W3117301244 cites W3127686677 @default.
• W3117301244 cites W3198160809 @default.
• W3117301244 doi "https://doi.org/10.48550/arxiv.2012.14966" @default.
• W3117301244 hasPublicationYear "2020" @default.
• W3117301244 type Work @default.
• W3117301244 sameAs 3117301244 @default.
• W3117301244 citedByCount "1" @default.
• W3117301244 countsByYear W31173012442021 @default.
• W3117301244 crossrefType "posted-content" @default.
• W3117301244 hasAuthorship W3117301244A5001041485 @default.

• next