DTMM: Deploying TinyML Models on Extremely Weak IoT Devices with Pruning

• DOI: 10.48550/arxiv.2401.09068
• Publication Date: 2024-01-01
• Authors: Lixiang Han; Zhen Xiao; Zhenjiang Li
• Keywords: Pruning

DTMM is a library designed for efficient deployment and execution of machine learning models on weak IoT devices such as microcontroller units (MCUs). The motivation designing comes from the emerging field tiny (TinyML), which explores extending reach to many low-end achieve ubiquitous intelligence. Due capability embedded devices, it necessary compress by pruning enough weights before deploying. Although has been studied extensively computing platforms, two key issues with methods are exacerbated MCUs: need be deeply compressed without significantly compromising accuracy, they should perform efficiently after pruning. Current solutions only one these objectives, but not both. In this paper, we find that pruned have great potential MCUs. Therefore, propose unit selection, pre-execution optimizations, runtime acceleration, post-execution low-cost storage fill gap models. It can integrated into commercial ML frameworks practical deployment, prototype system developed. Extensive experiments various show promising gains compared state-of-the-art methods.