06/26/2025, 08:05 AM UTC
用于微型设备的小型化人工智能Miniature Artificial Intelligence for Smallest Devices
➀ 格拉茨技术大学、Pro2Future和圣加仑大学的研究人员开发了可在内存仅4KB的物联网设备上本地运行AI模型的方法，无需依赖外部计算资源；
➁ 通过模型分块、子空间可配置网络（SCNs）、量化与剪枝等技术，在保证精度的前提下压缩模型规模，使图像处理速度提升达7.8倍；
➂ 成果适用于工业自动化（如无人机定位）、智能家居遥控器（延长电池寿命）和无钥匙汽车防信号克隆等场景，展现了嵌入式AI的广泛潜力。
➀ Researchers from TU Graz, Pro2Future, and the University of St. Gallen developed methods to deploy AI models on resource-constrained IoT devices with as little as 4KB memory, enabling localized processing without external compute resources;
➁ The team used techniques like model partitioning, Subspace-Configurable Networks (SCNs), quantization, and pruning to balance model size and accuracy, achieving up to 7.8x faster image processing on IoT devices;
➂ Applications include industrial automation (e.g., drone/robot localization), smart home remote controls, and anti-spoofing for keyless car entry systems, demonstrating broad scalability across embedded systems.
---
本文由大语言模型（LLM）生成，旨在为读者提供半导体新闻内容的知识扩展（Beta）。
