Summary of Towards Low-energy Adaptive Personalization For Resource-constrained Devices, by Yushan Huang et al.
Towards Low-Energy Adaptive Personalization for Resource-Constrained Devices
by Yushan Huang, Josh Millar, Yuxuan Long, Yuchen Zhao, Hamed Haddadi
First submitted to arxiv on: 23 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes Target Block Fine-Tuning (TBFT), a framework for personalizing machine learning models to address data drift in IoT applications, while considering energy costs. The authors categorize data drift into input-level, feature-level, and output-level and fine-tune different blocks of the model accordingly. This approach is designed for resource-constrained devices and is evaluated on a ResNet model with three datasets, training sizes, and a Raspberry Pi. Compared to full fine-tuning, TBFT achieves an average accuracy improvement of 15.30% while saving 41.57% energy consumption. The framework, named Block Avg, fine-tunes each block individually and averages their performance improvements. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making machine learning models better at adapting to new data in IoT devices without using too much energy. They came up with a new way to personalize the model called Target Block Fine-Tuning (TBFT). It works by identifying different types of changes in the data and fine-tuning specific parts of the model accordingly. This approach is designed for devices that have limited power, like a Raspberry Pi. The results show that TBFT can improve the accuracy of the model by 15% while using 41% less energy compared to just fine-tuning the whole model. |
Keywords
* Artificial intelligence * Fine tuning * Machine learning * Resnet
