Olga Saukh’s inauguration talk at Graz Technical University, titled “Towards embedded intelligence in IoT systems,” will be held as an online lecture.
If you want to attend, please send an email to office@csh.ac.at
Abstract:
As the IoT era expands, we observe more and more sensors embedded into our everyday objects such as medical devices, wearables, home appliances, mobile and static IoT devices. Data processing coming from these devices has been traditionally performed in the cloud or in the backend infrastructure and is now increasingly shifted to edge devices to save energy, bandwidth and preserve ownership rights on sensitive data to gain trust.
It’s incredibly rewarding to build small, cost-effective, autonomous and adaptive systems to make the world a better place! Embedded intelligence is gaining a momentum, with autonomous driving, autonomous robots, interactive assistants and intelligent early warning systems being just a few examples. The society increasingly depends on smart objects and thus dependability and trust are key for their societal acceptance.
But can we really trust embedded intelligence? Sensor readings coming from low-cost miniaturised sensors suffer from noise, drifts and unexpected failures. Severely resource-optimised machine learning models have to deliver high prediction quality on unseen data, provide performance guarantees and robustness to input disturbances and unknowns in the real-world.
Making embedded intelligence systems trustworthy constitutes an outstanding challenge for the years to come.In this talk, I will summarise our research activities towards building embedded intelligence in IoT systems and formulate open questions in this exciting field. First, how to ensure effectiveness of these systems given limited computation, memory and energy resources. Second, how to make these systems work reliably and adapt to changes, despite unpredictability of the physical world. Third, how to make sure embedded intelligence develops into an energy-efficient and sustainable technology. Finally, how to protect user privacy when deploying these systems in sensitive contexts.