Research

Over-the-air Update for Energy Harvesting Resource Constraint System

My research focuses on enabling robust and efficient over-the-air (OTA) update mechanisms for energy-harvesting IoT devices, which are increasingly deployed in remote and environmentally constrained settings. While OTA update techniques are well-established for cable-powered systems, they often fail under the intermittent and unstable power conditions characteristic of energy-harvesting platforms.
To address these challenges, I explored update mechanisms tailored for intermittently powered devices, including incremental learning updates and flash memory-specific strategies. Building on this foundation, I proposed a general OTA framework capable of supporting a wide range of applications and hardware, with resilience to varying energy conditions.

Intermittent TinyML Inference

My research focuses on enabling efficient and reliable deep learning inference on intermittently powered, energy-harvesting IoT devices. These devices operate under severe resource constraints and unpredictable power availability, which pose significant challenges to conventional machine learning techniques.
To overcome these challenges, I investigate lightweight and energy-adaptive inference strategies that ensure robust model execution despite frequent power interruptions. My work aims to make TinyML feasible for real-world deployment in ultra-low-power environments by improving inference efficiency, maintaining model accuracy, and eliminating the need for complex power management or checkpointing systems.