Researchers have analyzed the performance and power consumption of TinyML systems, which enable machine learning at the edge. They studied various applications on microcontrollers, including neural network models, software libraries, operating systems, and hardware architectures. The study found that abstraction layers can significantly impact performance and energy efficiency, and proposed a model to estimate these costs. This work can help designers optimize Neural Architecture Search and CNN inference on edge devices. --- Why it matters: This research matters because it provides a systematic understanding of the trade-offs between performance and power consumption in TinyML systems. It will help engineers design more efficient edge devices for machine learning applications. Source: https://arxiv.org/abs/2608.21646