On-Device AI Inference Articles

Android On-device AI System Health: Dynamic Inference Degradation by Thermal, Battery, and Memory Pressure

A practical three-dimensional degradation strategy for Android on-device AI inference, coordinating thermal status, battery state, and memory pressure with normalized scoring, model preloading, and state migration.

Android NNAPI End to End: From HAL Abstraction to Qualcomm and MTK NPUs

A practical deep dive into Android NNAPI, covering the HAL layer, vendor drivers, graph partitioning, operator fallback, and on-device AI inference tuning.