Abstract: Machine learning (ML) has great potential in detecting cognitive, mental and functional health disorders from speech, as acoustic properties of speech and corresponding patterns in language are modified by a wide variety of health-related effects. However, there exist many real-life problems that make developing ML models a challenging task. In this talk, I will discuss some of the common challenges researchers are dealing with when developing ML models, such as the lack of appropriate training data, absence of labels, effects of automatic speech recognition, limitations of English-only models and the lack of interpretability. I will explain how these challenges affect the quality and performance of existing ML models used to detect cognitive diseases from speech and will present novel solutions we are developing at Winterlight Labs to deal with them.
Bio: Jekaterina Novikova is a Director of Machine Learning at Winterlight Labs, Toronto/Canada, where she leads a team of research scientists and engineers. She focuses on strategic leadership in machine learning, helping teams to incorporate state-of-the-art research into business priorities, prioritize opportunities, and develop roadmaps. Jekaterina holds a PhD from the University of Bath/UK and did her PostDoc at the Heriot-Watt University in Edinburgh/UK.
Development and evaluation of natural language interfaces is a key area of Jekaterina's research, with applications ranging from human-robot spoken dialogue systems to machine learning-based diagnostic platforms that detect cognitive and mental diseases from human speech. Jekaterina has authored over 40 peer-reviewed papers in this area, that were published and presented at top-tier conferences, such as EMNLP and ACL.
In recent years, Jekaterina was invited to be the keynote speaker at various conferences and workshops across the globe, such as CogX 2017 in London/UK, Re-Work AI Summit 2018 in Toronto/Canada, MLconf 2019 in San Francisco/US, ODSC East 2020 Virtual. Jekaterina's work and outreach activities have been recognised with a nomination of ""30 Influential Women Advancing AI in Canada"", as well as best research paper nominations at the conferences HAI and SigDIAL.