DC Field | Value | Language |
---|---|---|
dc.contributor.author | Shin, Jaekang | ko |
dc.contributor.author | Choi, Seungkyu | ko |
dc.contributor.author | Choi, Yeongjae | ko |
dc.contributor.author | Kim, Lee-Sup | ko |
dc.date.accessioned | 2020-11-30T09:50:28Z | - |
dc.date.available | 2020-11-30T09:50:28Z | - |
dc.date.created | 2020-11-30 | - |
dc.date.created | 2020-11-30 | - |
dc.date.created | 2020-11-30 | - |
dc.date.issued | 2020-07 | - |
dc.identifier.citation | 57th ACM/IEEE Design Automation Conference, DAC 2020 | - |
dc.identifier.issn | 0738-100X | - |
dc.identifier.uri | http://hdl.handle.net/10203/277771 | - |
dc.description.abstract | Incremental learning is drawing attention to widen capabilities of device-AI. Previous works have researched to reduce numerous computations and memory accesses required for the training process of IL, but they could not show a noticeable improvement in the weight gradient computation (WGC) phase. Therefore, we propose a selective weight update technique that searches for critical weights to be updated by applying the IL algorithm that training per-task binary masks. Also, we introduce a novel dataflow for the implementation of selective WGC on typical NPUs with minimum overheads. On average, our system shows a 2.9× speed up and 2.5× energy efficiency in WGC without degrading training quality. | - |
dc.language | English | - |
dc.publisher | Institute of Electrical and Electronics Engineers Inc. | - |
dc.title | A pragmatic approach to on-device incremental learning system with selective weight updates | - |
dc.type | Conference | - |
dc.identifier.wosid | 000628528400019 | - |
dc.identifier.scopusid | 2-s2.0-85093961539 | - |
dc.type.rims | CONF | - |
dc.citation.publicationname | 57th ACM/IEEE Design Automation Conference, DAC 2020 | - |
dc.identifier.conferencecountry | US | - |
dc.identifier.conferencelocation | Virtual | - |
dc.identifier.doi | 10.1109/DAC18072.2020.9218507 | - |
dc.contributor.localauthor | Kim, Lee-Sup | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.