On-Device Neural Architecture Search

Short summary

Researchers propose a lightweight Neural Architecture Search algorithm that runs directly on edge devices like Raspberry Pi 4 to automatically optimize tiny neural networks for analyzing real-time sensor data. Validated on Italian Sign Language gesture recognition (using surface electromyography) and industrial fault diagnosis, the approach achieves 5.96% better accuracy with 37% less RAM than state-of-the-art methods. Enables practical on-device ML adaptation for human-machine interfaces without cloud dependency.

• •On-device NAS algorithm automatically optimizes neural networks for edge devices without cloud dependency
• •Achieves 37% RAM reduction and 5.96% accuracy improvement on Raspberry Pi 4
• •Enables user-specific model adaptation for gesture recognition and fault diagnosis applications

Generated with AI, which can make mistakes.