Andreas Theissler
Andreas Theissler received the Ph.D. degree from Brunel University London, in 2014. He is currently a Professor at the Aalen University of Applied Sciences, Germany, where he researches and lectures on different aspects of machine learning and human-centered AI. Prior to that, he studied software engineering and has worked in different data science positions in industry. He has published on the interplay of machine learning and users, for example on the questions of how we can evaluate, understand, improve, or enable machine learning by incorporating expert knowledge. In addition, he has worked in the field of machine learning in applications, e.g., anomaly detection in time series from automotive systems.
Talk: Explainable AI for time series classification and anomaly detection: current state and open issues
A large part of the research on explainable AI is done on tabular data or on image data. We believe that time series should receive the same research attention since they are omnipresent, e.g., in technical systems, the medical domain, or business applications. In this invited talk, the fields of time series classification and anomaly detection in time series are discussed.
Based on our recently published literature review on XAI for time series classification, the talk will categorize the research field through a taxonomy subdividing the methods into time points-based, subsequences-based and instance-based, and as a main contribution identify open research directions with the aim to inspire future research. While we believe excellent research has been done in the field, there is a variety of unsolved issues, leaving room for numerous future works. To this end, this workshop can help guiding future research directions for XAI on time series data.
George Tzagkarakis
George Tzagkarakis is a Research Scientist with more than 15 years of R&D experience in Quantitative Research and Scientific Data Analysis, with a main focus on multimodal data learning and analysis, distributed information processing, mathematical modeling, risk analytics, and computational finance. He holds a PhD and MSc in Computer Science (1st in class, highest honors) - with a major in Statistical Signal Processing - from the Computer Science Department, University of Crete (UOC), Greece, a PhD in Finance - with a major in Risk Quantification - from the Economics, Business and Society Doctoral School, Research Institute for the Management of Organizations, University of Bordeaux, France, and a BSc in Mathematics (1st in class, highest honors) - with a major in Applied and Computational Mathematics - from the Department of Mathematics, University of Crete, Greece. In the period 2002-2010, he was a Research Associate at the Foundation for Research and Technology-Hellas (FORTH), Institute of Computer Science (ICS), Crete, Greece, as a member of the Telecommunications and Networks Laboratory and the Signal Processing Laboratory. A number of academic distinctions have been awarded to him as both an undergraduate and a post-graduate student from UOC, FORTH-ICS, as well as from external state foundations. From 2010 to 2012, he was a Marie Curie Postdoctoral Researcher at the Cosmology and Statistics Laboratory, CEA/Saclay, France, working on the design and implementation of compressive sensing algorithms for remote imaging in areal and terrestrial surveillance systems. In the period from 2012 to 2018, he was a Senior Researcher and Scientific Director at EONOS Investment Technologies, Paris, France, being responsible for developing and managing scientific research projects strategically aligned with the state-of-the-art in quantitative analysis, econometrics, and computational finance using advanced signal and data processing methodologies. Since 2018, he has been at FORTH-ICS, where he currently holds a Principal Researcher position, affiliated with the Signal Processing Laboratory. His current research interests primarily focus on statistical signal and image processing, non- Gaussian heavy-tailed modeling, compressive sensing and sparse representations, distributed signal processing for smart sensor networks, computational finance, and machine learning applications in operations research. George is empowered with pure problem-solving skills and analytical thinking power, in conjunction with an extended experience in transferring research and interacting with the industry, which he obtained through his involvement in European and national R&D and R&I projects.
Talk: Feature Engineering for Graph-based Analysis of Recurrent Behavior in Biosignal Ensembles
As the field of brain monitoring is evolving rapidly, there is an increasing demand for new approaches to handle and analyze relevant signals, such as electroencephalogram (EEG) ensembles. EEG signals, which are often corrupted by impulsive (heavy-tailed) noise, admit naturally graph representations that encode the inherent spatio-temporal interdependencies between the electrodes. Furthermore, there is a growing interest in dynamic approaches to functional brain connectivity, and their potential applications in understanding atypical brain function. To address these problems, this talk will elaborate on two distinct directions: (i) the design of an efficient regularized graph filtering method based on fractional lower-order moments, which better adapts to the heavy-tailed statistics of impulsive noise; (ii) recurrence quantification analysis for the engineering of appropriate features characterizing the dynamic evolution of the underlying EEG data generating process, which are further combined with conventional classifiers for the identification of epileptic seizures.
Panagiotis Papapetrou
Panagiotis Papapetrou is a Professor at the Department of Computer and Systems Sciences of Stockholm University. His area of expertise is algorithmic data mining with particular focus on time series classification, explainability, and emphasis on healthcare applications. Panagiotis received his PhD in Computer Science at Boston University in 2009 and his Masters degree at the same university in 2007. He was a postdoctoral researcher at Aalto University during 2009-2012, and a lecturer at Birkbeck University of London, UK, during 2012-2013. He has participated in several national and international research projects, among which a 4-year starting grant funded by the Swedish Research Council. He is serving as Action Editor at the Data Mining and Knowledge Discovery journal and he is a Board Member of the Swedish Artificial Intelligence Society. Panagiotis has been involved in the organization of several Workshops and Tutorials at KDD, ICDM, and ECML/PKDD. Moreover, he has served as the general chair of IDA 2016, PhD consortium co-chair at ICDM 2018, and Workshops co-chair at ICDM 2019.
Talk: Towards explainable time series classification
Time series classification has received a lot of attention over the past decade including recent deep learning-based approaches that show impressive performance against state-of-the-art solutions. At the same time, there is a rising need for explainability either by design or post-hoc, especially when the underlying models are opaque and are also built using data of increased complexity. This talk will focus on recently proposed explainable time series classification methods, and will present an overview of existing approaches towards this direction. Emphasis will be given to both univariate and multivariate explainable time series, as well as on explainable feature representations and time series counterfactual generation. Finally, the talk will conclude with current challenges and future directions in the area.