Summary of Ditmos: Delving Into Diverse Tiny-model Selection on Microcontrollers, by Xiao Ma et al.
DiTMoS: Delving into Diverse Tiny-Model Selection on Microcontrollers
by Xiao Ma, Shengfeng He, Hezhe Qiao, Dong Ma
First submitted to arxiv on: 14 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 DiTMoS, a novel deep neural network (DNN) training and inference framework, enables efficient and accurate DNN inference on microcontrollers. By constructing small/weak models directly and improving their accuracy, DiTMoS outperforms current methodologies that compress larger models at the expense of model accuracy. The framework employs a selector-classifiers architecture, where the selector routes each input sample to the appropriate classifier for classification. To improve accuracy, DiTMoS introduces three strategies: diverse training data splitting to increase classifiers’ diversity, adversarial selector-classifiers training to ensure synergistic interactions and maximize complementarity, and heterogeneous feature aggregation to improve classifiers’ capacity. The framework also includes a network slicing technique to alleviate memory overhead incurred by feature aggregation. Evaluation on three time-series datasets for human activity recognition, keywords spotting, and emotion recognition demonstrates that DiTMoS achieves up to 13.4% accuracy improvement compared to the best baseline. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary DiTMoS is a new way to make deep neural networks work well on tiny computers called microcontrollers. Usually, these computers can’t handle big neural networks because they don’t have enough memory or power. But DiTMoS makes small neural networks that are good at specific tasks and then combines them to get even better results. It uses a special kind of architecture where each input gets sent to the right part of the network for processing. To make it even better, DiTMoS has some tricks like using different training data, making sure the parts work well together, and combining features in a smart way. It also helps with memory by breaking things down into smaller pieces. When tested on three types of data, DiTMoS did significantly better than other methods. |
Keywords
* Artificial intelligence * Activity recognition * Classification * Inference * Neural network * Time series
