Summary of On-device Training Of Fully Quantized Deep Neural Networks on Cortex-m Microcontrollers, by Mark Deutel et al.
On-Device Training of Fully Quantized Deep Neural Networks on Cortex-M Microcontrollers
by Mark Deutel, Frank Hannig, Christopher Mutschler, Jürgen Teich
First submitted to arxiv on: 15 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 This paper explores the challenge of training deep neural networks (DNNs) on microcontroller units (MCUs), which are critical for on-device adaptation and fine-tuning. The authors present a method that enables efficient DNN training using fully quantized training (FQT) and dynamic partial gradient updates, allowing for on-device training of DNNs on Cortex-M MCUs. The approach is evaluated on multiple vision and time-series datasets, highlighting the tradeoff between training accuracy, memory overhead, energy, and latency on real hardware. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary On-device training of deep neural networks (DNNs) allows models to adapt and fine-tune to new data or changing domains while deployed on microcontrollers. However, training DNNs is a resource-intensive task that requires powerful processors and lots of memory. This makes it hard to train DNNs directly on microcontrollers, which have limited resources. In this paper, the authors present a method that lets them train DNNs efficiently on microcontrollers using fully quantized training (FQT) and dynamic partial gradient updates. They show how their approach works well on different datasets and highlight the tradeoff between accuracy, memory usage, energy consumption, and latency. |
Keywords
* Artificial intelligence * Fine tuning * Time series
