Automated deep-learning model optimization framework for microcontrollers
This paper introduces a framework for optimizing deep-learning models on microcontrollers (MCUs) that is crucial in today’s expanding embedded device market. We focus on model optimization techniques, particularly pruning and quantization, to enhance the performance of neural networks within the lim...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Electronics and Telecommunications Research Institute (ETRI)
2025-04-01
|
| Series: | ETRI Journal |
| Subjects: | |
| Online Access: | https://doi.org/10.4218/etrij.2023-0522 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This paper introduces a framework for optimizing deep-learning models on microcontrollers (MCUs) that is crucial in today’s expanding embedded device market. We focus on model optimization techniques, particularly pruning and quantization, to enhance the performance of neural networks within the lim-ited resources of MCUs. Our approach combines automatic iterative optimization and code generation, simplifying MCU model deployment without requiring extensive hardware knowledge. Based on experiments with architec-tures, such as ResNet-8 and MobileNet v2, our framework substantially reduces the model size and enhances inference speed that are crucial for MCU efficiency. Compared with TensorFlow Lite for MCUs, our optimizations for MobileNet v2 reduce static random-access memory use by 51%-57% and flash use by 17%-62%, while increasing inference speed by approximately 1.55 times. These advancements highlight the impact of our method on perfor-mance and memory efficiency, demonstrating its value in embedded artificial intelligence and broad applicability in MCU-based neural network optimization.
|
|---|---|
| ISSN: | 1225-6463 2233-7326 |