DAF: An Efficient End-to-End Dynamic Activation Framework for on-Device DNN Training
Recent advancements in on-device training for deep neural networks have underscored the critical need for efficient activation compression to overcome the memory constraints of mobile and edge devices. As activations dominate memory usage during training and are essential for gradient computation, compressing them without compromising accuracy remains a key research challenge. While existing methods for dynamic activation quantization promise theoretical memory savings, their practical deployment is impeded by system-level challenges such as computational overhead and memory fragmentation.
Authors
- Kaiyan Liu (ORCID: https://orcid.org/0009-0009-8663-0698)
- Shaohan Hu (ORCID: https://orcid.org/0000-0002-2877-2665)
- Richard Chen (ORCID: https://orcid.org/0000-0002-5912-5620)
- Yuyang Leng (ORCID: https://orcid.org/0009-0008-9376-0880)
- Renyuan Liu (ORCID: https://orcid.org/0000-0001-9710-6116)
- Heechul Yun (ORCID: https://orcid.org/0000-0002-8515-2622)
- Peijun Zhao (ORCID: https://orcid.org/0000-0003-2843-6941)
- Shuochao Yao (ORCID: https://orcid.org/0000-0001-7446-1430)
Institutions
- JPMorgan Chase & Co (United States) (US)
- University of Kansas (US)
- George Mason University (US)
Publication Details
- Published
- 2025-06-23
- DOI
- https://doi.org/10.1145/3711875.3729145
- Primary Topic
- EEG and Brain-Computer Interfaces
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Science Foundation