Beyond Low-rank Decomposition: A Shortcut Approach for Efficient On-Device Learning
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913958002163712 |
|---|---|
| author | Nguyen, Le-Trung Quelennec, Ael Nguyen, Van-Tam Tartaglione, Enzo |
| author_facet | Nguyen, Le-Trung Quelennec, Ael Nguyen, Van-Tam Tartaglione, Enzo |
| contents | On-device learning has emerged as a promising direction for AI development, particularly because of its potential to reduce latency issues and mitigate privacy risks associated with device-server communication, while improving energy efficiency. Despite these advantages, significant memory and computational constraints still represent major challenges for its deployment. Drawing on previous studies on low-rank decomposition methods that address activation memory bottlenecks in backpropagation, we propose a novel shortcut approach as an alternative. Our analysis and experiments demonstrate that our method can reduce activation memory usage, even up to $120.09\times$ compared to vanilla training, while also reducing overall training FLOPs up to $1.86\times$ when evaluated on traditional benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_05086 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Beyond Low-rank Decomposition: A Shortcut Approach for Efficient On-Device Learning Nguyen, Le-Trung Quelennec, Ael Nguyen, Van-Tam Tartaglione, Enzo Machine Learning Artificial Intelligence On-device learning has emerged as a promising direction for AI development, particularly because of its potential to reduce latency issues and mitigate privacy risks associated with device-server communication, while improving energy efficiency. Despite these advantages, significant memory and computational constraints still represent major challenges for its deployment. Drawing on previous studies on low-rank decomposition methods that address activation memory bottlenecks in backpropagation, we propose a novel shortcut approach as an alternative. Our analysis and experiments demonstrate that our method can reduce activation memory usage, even up to $120.09\times$ compared to vanilla training, while also reducing overall training FLOPs up to $1.86\times$ when evaluated on traditional benchmarks. |
| title | Beyond Low-rank Decomposition: A Shortcut Approach for Efficient On-Device Learning |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2505.05086 |