Olga Saukh, associate professor

Embedded Learning and Sensing Systems Group
Institute of Technical Informatics
Graz University of Technology

Abstract: TinyML has transformed embedded AI by making machine learning practical on resource-constrained devices through efficient models, compression, and hardware-aware optimization. Yet making AI fit was never the whole challenge: embedded systems remain costly and difficult to engineer, optimize, and operate. Foundation models now offer a new way to tackle this long-standing challenge. In this talk, I explore the shift from AI running on embedded systems to AI shaping them, as models begin to assist with code, configuration, optimization, debugging, and adaptation. Drawing on recent work and our own experiments, I examine what capabilities these models actually bring to engineering, how strongly those capabilities depend on reasoning, tools, and adaptation, and where they break down. These developments point toward AI-native embedded engineering, while raising a critical question: can we make these systems reliable enough to trust?

Bio: Olga Saukh is an associate professor and leads the Embedded Learning and Sensing Systems group at TU Graz, affiliated with the Institute of Technical Informatics (ITI). Since 2020, she holds a habilitation in Embedded Systems from TU Graz. She did her postdoctoral training at ETH Zurich in 2010-2016 working in the group headed by Prof. Lothar Thiele in the Computer Engineering and Networks Laboratory. Following her Bachelors in Applied Mathematics from the Taras Shevchenko University of Kyiv, Ukraine in 2002 and her Masters in Applied Computer Science from the University of Freiburg, Germany in 2004, she received her Ph.D. in Computer Science from the University of Bonn, Germany in 2009. She received the 2010 CONET Ph.D. Academic Award for her thesis "Efficient Algorithms for Structuring Wireless Sensor Networks". Her research focuses on efficient machine learning and the design of practical AI-based systems at the intersection of deep learning and embedded systems. She is interested in both the theoretical foundations of deep learning optimization for resource-constrained devices and their practical applications in real-world systems. She serves on program committees of leading international conferences in both the Machine Learning and Embedded Systems communities.