On-Device Tiny Machine Learning - From Algorithm to Technology

Explore the transformative potential of Tiny Machine Learning (TinyML) in this comprehensive conference talk that examines how everyday devices can make intelligent decisions while learning and evolving autonomously. Delve into the "computing everywhere" paradigm and its role in reshaping how intelligence is embedded into ubiquitous devices, moving beyond traditional inference-only approaches to address the complex challenges of on-device learning. Discover the critical constraints of memory, computation, and energy efficiency that define TinyML implementations, and learn about adaptive mechanisms designed to optimize processing pipelines dynamically during inference. Examine lightweight and efficient on-device training strategies that enable continuous adaptation, personalization, and responsiveness to evolving environmental conditions, providing insights into how these technologies can drive widespread adoption of intelligent systems in daily life.