KEYNOTE SPEAKERS
Prof. Arjan Durresi
Algorithms and computers have been used for a long time in supporting decision making in various fields of human endowers. Examples include optimization techniques in engineering, statistics in experiment design, modeling of different natural phenomena, and so on. In all such uses of algorithms and computers, an essential question has been how much we can trust them, what are the potential errors of such models, what is the field range of their applicability? With time the algorithms and computers we use have become more powerful and more complex, and we call them today as Artificial Intelligence that includes various machine learning and other algorithmic techniques. But the increase of power and complexity of algorithms and computers and with extended use of them the question of how much we should trust them becomes more crucial. Their complexity might hide more potential errors and especially the interdependencies; their solution might be difficult to be explained, and so on. To deal with these problems, we have developed an evidence and measurement-based trust management system; our system can be used to measure trust in human to human, human to machine, and machine to machine interactions. In this talk, we will introduce our trust system and its validation on real stock market data. Furthermore, we will discuss the use of our trust system to build more secure computer systems, filter fake news on social networks and develop better collective decision making support systems in managing natural resources, as well as future potential uses.
Arjan Durresi is a Professor of Computer Science at Indiana University Purdue University in Indianapolis, Indiana. In the past, he held positions at LSU and The Ohio State University. His research interests include trustworthy decision making and AI, networking, and security. He has published about 100 articles in journals and over 200 articles in conference proceedings and seven book chapters. He also has authored over thirty contributions to standardization organizations such as IETF, ATM Forum, ITU, ANSI and TIA.
Prof. Shengrui Wang, University of Sherbrooke, Canada
Time series are a type of time-dependent data found in many fields such as finance, medicine, meteorology, ecology and utility industry. In such fields, time series forecasting is an issue of great importance as it can help predict future index/equity values and behavioral changes in a stock market, health trajectories of a patient and probabilities of failure events such death and (re)-hospitalization, varying electricity consumption of individual households, etc. It also poses significant theoretical and methodological challenges. One such challenge is identification and prediction of a regime and regime shifts since most time series models, whether they are linear or non-linear, work well only within one regime. In this talk, I will introduce our recent work on building a novel framework to model time series interaction and evolution as an ecosystem. The focus of the talk is to show how to make use of the graph or network structures and community analysis approaches to account for interactions or interdependencies between time series. For this purpose, we build a time-evolving network graph from moving window segments of time series, where nodes represent time series profiles and links correspond to similarity between the time series. By using a community detection algorithm, we discover communities displaying similar behaviours, or regimes, and trace the discovered behaviours in the form of trajectories of structural changes in these communities. Such changes are considered as behaviour events that may appear, persist, or disappear w.r.t. a community, reflecting the survival of regimes and abrupt transition between regimes. Using such network structures for modelling the interactions allows also discovering "relational" features explaining why certain behaviours may persist longer than others. These relational features, together with behaviour profiles, constitute input to machine learning models for regime analysis and trajectory forecasting. In our work, we tackle the problem of learning regime shifts by modeling a time-dependent probability of transition between regimes, using a full time-dependent Cox-regression model. We evaluate the whole approach by testing it on both synthetic and real data sets and compare its performance with that of state-of-the-art learning algorithms.
Shengrui Wang, PhD from Institut National Polytechnique de Grenoble, France, is a full professor of Computer Science and Vice-Dean Education, Faculty of Sciences at University of Sherbrooke. Prof. Wang's research interests include data mining, pattern recognition, machine learning artificial intelligence, knowledge discovery, and neural networks. He is an internationally recognized expert in high-dimensional and sequence data mining, and is well known for his expertise and achievements in areas that include clustering, outlier detection, social network analysis, sequential and temporal data analysis, and health informatics. He has successfully applied his research in areas such as bioinformatics, healthcare data analytics, finance data analysis, online social networks, image databases, smart environments, radar monitoring and others. Prof. Wang’s research has been supported by the Natural Science and Engineering Research Council of Canada (NSERC) since 1992. He was an NSERC Discovery Accelerator Supplements recipient in 2010 and 2020. He has frequently been invited to serve in national and international grant review committees. He was a member of the NSERC computer science grant selection committee in 2010-14, and co-chair of this committee in 2013-14. He also served as the chair of the NSERC Research Tools and Instruments Selection Committee in the areas of Computer, Mathematical and Statistical Sciences in 2016 and 2018.