Matches in SemOpenAlex for { <https://semopenalex.org/work/W3138787737> ?p ?o ?g. }
Showing items 1 to 83 of
83
with 100 items per page.
- W3138787737 abstract "The memory capacity of embedding tables in deep learning recommendation models (DLRMs) is increasing dramatically from tens of GBs to TBs across the industry. Given the fast growth in DLRMs, novel solutions are urgently needed, in order to enable fast and efficient DLRM innovations. At the same time, this must be done without having to exponentially increase infrastructure capacity demands. In this paper, we demonstrate the promising potential of Tensor Train decomposition for DLRMs (TT-Rec), an important yet under-investigated context. We design and implement optimized kernels (TT-EmbeddingBag) to evaluate the proposed TT-Rec design. TT-EmbeddingBag is 3 times faster than the SOTA TT implementation. The performance of TT-Rec is further optimized with the batched matrix multiplication and caching strategies for embedding vector lookup operations. In addition, we present mathematically and empirically the effect of weight initialization distribution on DLRM accuracy and propose to initialize the tensor cores of TT-Rec following the sampled Gaussian distribution. We evaluate TT-Rec across three important design space dimensions -- memory capacity, accuracy, and timing performance -- by training MLPerf-DLRM with Criteo's Kaggle and Terabyte data sets. TT-Rec achieves 117 times and 112 times model size compression, for Kaggle and Terabyte, respectively. This impressive model size reduction can come with no accuracy nor training time overhead as compared to the uncompressed baseline." @default.
- W3138787737 created "2021-03-29" @default.
- W3138787737 creator A5005070359 @default.
- W3138787737 creator A5005712766 @default.
- W3138787737 creator A5028220093 @default.
- W3138787737 creator A5039412958 @default.
- W3138787737 date "2021-01-25" @default.
- W3138787737 modified "2023-10-14" @default.
- W3138787737 title "TT-Rec: Tensor Train Compression for Deep Learning Recommendation Models" @default.
- W3138787737 doi "https://doi.org/10.48550/arxiv.2101.11714" @default.
- W3138787737 hasPublicationYear "2021" @default.
- W3138787737 type Work @default.
- W3138787737 sameAs 3138787737 @default.
- W3138787737 citedByCount "7" @default.
- W3138787737 countsByYear W31387877372021 @default.
- W3138787737 countsByYear W31387877372023 @default.
- W3138787737 crossrefType "posted-content" @default.
- W3138787737 hasAuthorship W3138787737A5005070359 @default.
- W3138787737 hasAuthorship W3138787737A5005712766 @default.
- W3138787737 hasAuthorship W3138787737A5028220093 @default.
- W3138787737 hasAuthorship W3138787737A5039412958 @default.
- W3138787737 hasBestOaLocation W31387877371 @default.
- W3138787737 hasConcept C108583219 @default.
- W3138787737 hasConcept C111919701 @default.
- W3138787737 hasConcept C11413529 @default.
- W3138787737 hasConcept C114466953 @default.
- W3138787737 hasConcept C121332964 @default.
- W3138787737 hasConcept C149635348 @default.
- W3138787737 hasConcept C151730666 @default.
- W3138787737 hasConcept C154945302 @default.
- W3138787737 hasConcept C155281189 @default.
- W3138787737 hasConcept C17349429 @default.
- W3138787737 hasConcept C173608175 @default.
- W3138787737 hasConcept C199360897 @default.
- W3138787737 hasConcept C199683683 @default.
- W3138787737 hasConcept C202444582 @default.
- W3138787737 hasConcept C2779343474 @default.
- W3138787737 hasConcept C2780513914 @default.
- W3138787737 hasConcept C33923547 @default.
- W3138787737 hasConcept C41008148 @default.
- W3138787737 hasConcept C41608201 @default.
- W3138787737 hasConcept C62520636 @default.
- W3138787737 hasConcept C84114770 @default.
- W3138787737 hasConcept C86803240 @default.
- W3138787737 hasConceptScore W3138787737C108583219 @default.
- W3138787737 hasConceptScore W3138787737C111919701 @default.
- W3138787737 hasConceptScore W3138787737C11413529 @default.
- W3138787737 hasConceptScore W3138787737C114466953 @default.
- W3138787737 hasConceptScore W3138787737C121332964 @default.
- W3138787737 hasConceptScore W3138787737C149635348 @default.
- W3138787737 hasConceptScore W3138787737C151730666 @default.
- W3138787737 hasConceptScore W3138787737C154945302 @default.
- W3138787737 hasConceptScore W3138787737C155281189 @default.
- W3138787737 hasConceptScore W3138787737C17349429 @default.
- W3138787737 hasConceptScore W3138787737C173608175 @default.
- W3138787737 hasConceptScore W3138787737C199360897 @default.
- W3138787737 hasConceptScore W3138787737C199683683 @default.
- W3138787737 hasConceptScore W3138787737C202444582 @default.
- W3138787737 hasConceptScore W3138787737C2779343474 @default.
- W3138787737 hasConceptScore W3138787737C2780513914 @default.
- W3138787737 hasConceptScore W3138787737C33923547 @default.
- W3138787737 hasConceptScore W3138787737C41008148 @default.
- W3138787737 hasConceptScore W3138787737C41608201 @default.
- W3138787737 hasConceptScore W3138787737C62520636 @default.
- W3138787737 hasConceptScore W3138787737C84114770 @default.
- W3138787737 hasConceptScore W3138787737C86803240 @default.
- W3138787737 hasLocation W31387877371 @default.
- W3138787737 hasOpenAccess W3138787737 @default.
- W3138787737 hasPrimaryLocation W31387877371 @default.
- W3138787737 hasRelatedWork W1472213334 @default.
- W3138787737 hasRelatedWork W1543798151 @default.
- W3138787737 hasRelatedWork W1832263773 @default.
- W3138787737 hasRelatedWork W2366466109 @default.
- W3138787737 hasRelatedWork W2747563384 @default.
- W3138787737 hasRelatedWork W2962779982 @default.
- W3138787737 hasRelatedWork W2964667014 @default.
- W3138787737 hasRelatedWork W3023225477 @default.
- W3138787737 hasRelatedWork W4315777907 @default.
- W3138787737 hasRelatedWork W2622518229 @default.
- W3138787737 isParatext "false" @default.
- W3138787737 isRetracted "false" @default.
- W3138787737 magId "3138787737" @default.
- W3138787737 workType "article" @default.