Adaptive Tiny Machine Learning for Smart Eyewear
Smarts eyewear represents a new frontier in AI technology, with AI capabilities embedded directly into glasses. This is made possible by the usage of the Tiny Machine Learning (TinyML) paradigm, which integrates Machine Learning (ML) within embedded devices, constrained by limited memory, low computational power, and low power consumption, like smart eyewear. Existing TinyML approaches lack the capability of on-device personalization, hence the embedded AI capabilities in smart glasses do not adapt to individual users wearing them. This limitation becomes even more critical when considering the concept drift effect, where the data distribution changes over time, resulting in a decreased model accuracy. Addressing this limitation is the purpose of this thesis, in which A new methodology for designing ML/DL solutions capable of operating on smart eyewear and adapting over time within the perspective of adaptive AI is presented.
Back to Current Students