Prof. Zahra Moussavi
Speech Title: Application of engineering into Alzheimer’s disease management: Diagnosis and non-medication treatments
Abstract: Memory and cognitive declines are associated with normal brain aging but are also precursors to dementia, in particular the so called the pandemic of the century, Alzheimer’s disease. While currently there is no cure or “vaccine” against dementia, based on brain’s plasticity, there are hopes to delay the onset or to slow the progression of disease.
Alzheimer’s disease is multi-facet condition; thus, the key to its management is in a multi-disciplinary approach. In this talk, I will review diagnostic objective methods and non-medication treatments. In particular, I will elaborate on the application and research outcomes of virtual reality for its diagnosis, and three different treatment modalities (rTMS, tACS and cognitive exercises) on Alzheimer’s management.
Bibliography: Zahra Moussavi is a professor, a Canada Research Chair, and the founder and director of Biomedical Engineering Graduate Program at University of Manitoba. Her current research focuses are on medical devices instrumentation and signal analysis for sleep apnea management and Alzheimer’s diagnosis and treatment using virtual reality, rTMS and EVestG technologies. She is the recipient of several awards including the “Canada’s Most Powerful Women (Top 100)” and “Manitoba Distinguished Women” in 2014 and IEEE EMBS Distinguished Lecturer, 2014 and 2019. She has published more than 259 peer-reviewed papers in journals and conferences, and has given 94 invited talks/seminars including 2 Tedx Talks and 9 keynote speaker seminars at national and international conferences. Aside from academic work, on her spare time, she writes science articles for public; also has developed and offered memory fitness programs for aging population.
Prof. Pierre Duhamel
Speech Title: Evaluating the reliability of machine learning algorithms
Abstract: Machine Learning (ML) algorithms (and especially deep learning algorithms) heavily rely on the availability of large datasets and on their representativity of the situation of interest. This representativity can be questioned by various problems, such as overfitting or unexpected correlations. Examples will be given. It is therefore very important to be able to evaluate if the outcome of an ML algorithm can be fully trusted.
Such an objective can have many very useful consequences, and would allow us to answer such questions as:
1- Given a network that has been trained and is being used on new inputs, when is it necessary to train again the network (due to a shortage in the statistics of the signals to be processed)?
2- Can we obtain a measure of the accuracy of the result of the ML algorithm, so that we can “weight” the corresponding decisions appropriately?
3- Can we detect “outliers”?
4- Can we detect attacks on the network, attempting to push it “out of tune”?
Even if we cannot (obviously) answer all these questions, the talk will first envision several ways of stating the problem, and explain their advantages and drawbacks. Then, we will review the literature on the subject, and evaluate their usefulness with respect to the above objectives. Finally, we concentrate on a “blackbox” approach, insisting on its large applicability, and provide a first set of methods and corresponding first simulated results. Conclusions will elaborate on various practical applications of such algorithms.
Bibliography: Pierre Duhamel (IEEE Fellow, 1998, Eurasip Fellow, 2008, Grand Prix France Télécom of the French Science Academy, 2000) received the Eng. Degree in Electrical Engineering from the National Institute for Applied Sciences (INSA) Rennes, France in 1975, the Dr. Eng. Degree in 1978, and the Doctorat dès sciences degree in 1986, both from Orsay University, France. From 1975 to 1980, he was with Thomson-CSF (now Thales), Paris, France. In 1980, he joined the National Research Center in Telecommunications (CNET, now Orange Labs), Issy les Moulineaux, France. From 1993 to Sept. 2000, he has been professor at ENST, Paris (National School of Engineering in Telecommunications). He is now with CNRS/L2S (Laboratoire de Signaux et Systemes, Gif sur Yvette, France), where he developed studies in Signal processing for communications and signal/image processing for multimedia applications. He is currently investigating the connections between communication theory and networking. Dr Duhamel published more than 100 papers in international journals, more than 300 papers in international conferences, and holds 29 patents. He is a co-author of the book “Joint Source and Channel Decoding: A cross layer perspective with applications in video broadcasting” (2009, Academic Press). He successfully advised or co-advised 60 PhD students, two of them being now IEEE Fellows.