Matches in SemOpenAlex for { <https://semopenalex.org/work/W4293100404> ?p ?o ?g. }
Showing items 1 to 74 of
74
with 100 items per page.
- W4293100404 abstract "We propose a new concept called Weight Separation of deep neural networks (DNNs), which enables memory-efficient and accurate deep multitask learning on a memory-constrained embedded system. The goal of weight separation is to achieve extreme packing of multiple heterogeneous DNNs into the limited memory of the system while ensuring the prediction accuracy of the constituent DNNs at the same time. The proposed approach separates the DNN weights into two types of weight-pages consisting of a subset of weight parameters, i.e., shared and exclusive weight-pages. It optimally distributes the weight-pages into two levels of the system memory hierarchy and stores them separately, i.e., the shared weight-pages in primary (level-1) memory (e.g., RAM) and the exclusive weight-pages in secondary (level-2) memory (e.g., flask disk or SSD). First, to reduce the memory usage of multiple DNNs, less critical weight parameters are identified and overlapped onto the shared weight-pages that are deployed in the limited space of the primary (main) memory. Next, to retain the prediction accuracy of multiple DNNs, the essential weight parameters that play a critical role in preserving prediction accuracy are stored intact in the plentiful space of secondary memory storage in the form of exclusive weight-pages without overlapping. We implement two real systems applying the proposed weight separation: 1) a microcontroller-based multitask IoT system that performs multitask learning of 10 scaled-down DNNs by separating the weight parameters into FRAM and flash disk, and 2) an embedded GPU system that performs multitask learning of 10 state-of-the-art DNNs, separating the weight parameters into GPU RAM and eMMC. Our evaluation shows that memory efficiency, prediction accuracy, and execution time of deep multitask learning improve up to 5.9x, 2.0%, and 13.1x, respectively, without any modification of DNN models." @default.
- W4293100404 created "2022-08-26" @default.
- W4293100404 creator A5028983150 @default.
- W4293100404 creator A5044492162 @default.
- W4293100404 date "2022-03-21" @default.
- W4293100404 modified "2023-10-16" @default.
- W4293100404 title "Weight Separation for Memory-Efficient and Accurate Deep Multitask Learning" @default.
- W4293100404 cites W1499332833 @default.
- W4293100404 cites W1861492603 @default.
- W4293100404 cites W1895577753 @default.
- W4293100404 cites W1958932515 @default.
- W4293100404 cites W2034940213 @default.
- W4293100404 cites W2038484192 @default.
- W4293100404 cites W2052666245 @default.
- W4293100404 cites W2057376540 @default.
- W4293100404 cites W2057907879 @default.
- W4293100404 cites W2078224158 @default.
- W4293100404 cites W2112796928 @default.
- W4293100404 cites W2117539524 @default.
- W4293100404 cites W2117876524 @default.
- W4293100404 cites W2194775991 @default.
- W4293100404 cites W2553915786 @default.
- W4293100404 cites W2560647685 @default.
- W4293100404 cites W2585510649 @default.
- W4293100404 cites W2605370493 @default.
- W4293100404 cites W2886851211 @default.
- W4293100404 cites W2888469011 @default.
- W4293100404 cites W2903557836 @default.
- W4293100404 cites W2905515056 @default.
- W4293100404 cites W2937047963 @default.
- W4293100404 cites W2962793481 @default.
- W4293100404 cites W2963163009 @default.
- W4293100404 cites W2963314114 @default.
- W4293100404 cites W2963728985 @default.
- W4293100404 cites W2963839617 @default.
- W4293100404 cites W2988851551 @default.
- W4293100404 cites W3033567383 @default.
- W4293100404 cites W3099206234 @default.
- W4293100404 cites W3103203826 @default.
- W4293100404 doi "https://doi.org/10.1109/percom53586.2022.9762400" @default.
- W4293100404 hasPublicationYear "2022" @default.
- W4293100404 type Work @default.
- W4293100404 citedByCount "0" @default.
- W4293100404 crossrefType "proceedings-article" @default.
- W4293100404 hasAuthorship W4293100404A5028983150 @default.
- W4293100404 hasAuthorship W4293100404A5044492162 @default.
- W4293100404 hasConcept C154945302 @default.
- W4293100404 hasConcept C176649486 @default.
- W4293100404 hasConcept C41008148 @default.
- W4293100404 hasConcept C50644808 @default.
- W4293100404 hasConcept C9390403 @default.
- W4293100404 hasConcept C98986596 @default.
- W4293100404 hasConceptScore W4293100404C154945302 @default.
- W4293100404 hasConceptScore W4293100404C176649486 @default.
- W4293100404 hasConceptScore W4293100404C41008148 @default.
- W4293100404 hasConceptScore W4293100404C50644808 @default.
- W4293100404 hasConceptScore W4293100404C9390403 @default.
- W4293100404 hasConceptScore W4293100404C98986596 @default.
- W4293100404 hasLocation W42931004041 @default.
- W4293100404 hasOpenAccess W4293100404 @default.
- W4293100404 hasPrimaryLocation W42931004041 @default.
- W4293100404 hasRelatedWork W1573925694 @default.
- W4293100404 hasRelatedWork W2363471132 @default.
- W4293100404 hasRelatedWork W2386387936 @default.
- W4293100404 hasRelatedWork W2757277640 @default.
- W4293100404 hasRelatedWork W3001020386 @default.
- W4293100404 hasRelatedWork W3107474891 @default.
- W4293100404 hasRelatedWork W4362499384 @default.
- W4293100404 hasRelatedWork W55822027 @default.
- W4293100404 hasRelatedWork W644753246 @default.
- W4293100404 hasRelatedWork W1629725936 @default.
- W4293100404 isParatext "false" @default.
- W4293100404 isRetracted "false" @default.
- W4293100404 workType "article" @default.