CREST framework optimizes AI models for low-power embedded systems

Researchers have developed CREST, a hardware-in-the-loop neural architecture search (NAS) framework specifically designed for low-power embedded sensing systems. This framework addresses the limitations of existing methods by considering realistic deployment factors such as memory, latency, energy constraints, and sensing schedules, rather than relying solely on static proxies like FLOPs or parameters. Evaluations on Arm Cortex-M targets demonstrated that CREST can significantly reduce energy consumption and identify more suitable model architectures compared to traditional NAS approaches. AI

IMPACT Enables more efficient deployment of AI models on resource-constrained embedded devices.

RANK_REASON Research paper detailing a new framework for neural architecture search on embedded systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• ARM Cortex-M
• arXiv
• CREST
• DS-CNN
• STM32 N657

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →