Keynote talk I
Prof. Dr. rer. nat. habil.Thomas Villmann
University of Applied Sciences Mittweida Technikumplatz 17 09648 Mittweida, Germany
Talk Title: Interpretable Neural Networks for Classification Learning – Beyond Pure Accuracy Optimization
Abstract: During the last years a tremendous progress regarding performance and training time is observed. This development comes with improved hardware as well as with new concepts and network architectures accelerating the training process. Thus, deep network models became most favored models in machine learning. Further, publicly available tools enable users to perform efficient end-to-end learning scenarios for many network types. However, the complexity of the resulting networks hinders their interpretability. Many attempts were made to explain such models, e.g. by evaluation of layers after training. Yet, frequently these networks are just applied as a black-box-approach.
We propose network models which inherently provide interpretability. In particular, we focus on prototype-based models as known from vector quantization – the so-called generalized learning vector quantizers (GLVQ). We show that variants of those networks can achieve efficiently similar performance as deep approaches. Further, we explain how interpretability is achieved inherently in these networks, which finally can provide additional knowledge regarding the decision process beyond the pure network performance. In addition, we show that GLVQ-networks are robust against adversarial attacks by mathematical model analysis.
Exemplary applications will illustrate the proposed network properties and training behavior.
Bio.: Prof. Thomas Villmann is with the University of Applied Sciences Mittweida (UASM), Germany. He holds a diploma degree in Mathematics, received his Ph.D. in Computer Science in 1996 and his habilitation as well as venia legendi in the same subject in 2005, all from the University of Leipzig, Germany. From 1997 to 2009 he led the computational intelligence group of the hospital for psychotherapy at Leipzig University. In 2006 he was visiting professor at the University Paris Panthéon-Sorbonne in the dpeartment for statistical analysis and mathematical modelling (SAMM)
Since 2009 he is a full professor for Technomathematics/ Computational Intelligence at the UASM (Saxony), Germany. He is founding member of the German chapter of European Neural Network Society (GNNS) and its president since 2011. Further he leads the Institute of Computational Intelligence and Intelligent Data Analysis e.V. in Mittweida, Germany and the Computational Intelligence Group at the University of Applied Sciences Mittweida. Since 2017 he is director of the Saxony Institute for Computational Intelligence and Machine Learning (SICIM) at the UASM.
Prof. Villmann published more than 90 articles in leading journals. He authored and co-authored more than 250 conference papers and book chapters. Under his supervision, 12 PhD completitions were achieved, three more anticipated this year.
He is editor in chief of the Machine Learning Reports (MLR) and acts as an associate editor for Neural Processing Letters and for Computational Intelligence and Neuroscience.
His research focus includes the theory of prototype based clustering and classification, non-standard metrics, information theoretic and similarity based learning, quantum-enhanced machine learning, interpretable models in machine learning, statistical data analysis and their application in pattern recognition, data mining and knowledge discovery for use in medicine, bioinformatics, remote sensing, hyperspectral analysis and others.
Personally, he is an alpinist and likes climbing in high altitudes. Futher, he is active in judo since more than 40 years.
Keynote talk II
Prof. Paolo Rosso
Universitat Politècnica de Valencia, Spain
Talk Title: Profiling authors, bots and fake news spreaders
Abstract: Author profiling studies how language is shared by people. This helps in identifying aspects such as gender, age, native language, or even personality. Author profiling is a problem of growing importance in forensics, security, and marketing. E.g., from a marketing viewpoint,companies may be interested in knowing, on the basis of the analysis of blogs and online product reviews, the demographics of people that like or dislike their products.
Since 2013 at the PAN Lab at CLEF (https://pan.webis.de/) we have addressed several aspects of author profiling in social media: age and gender, personality, language variety, and gender from a multimodal perspective). In 2019 we investigated the feasibility of distinguishing whether the author of a Twitter feed is a bot or a human. The identification of bots from an author profiling perspective is of high importance. In fact, they may influence users with commercial, political or ideological purposes. For example, bots could artificially inflate the popularity of a product by promoting it and/or writing positive ratings, as well as undermine the reputation of competitive products through negative valuations. The threat is even greater when the purpose is political such as for the Brexit referendum or the US Presidential election (fearing the effect of this influence, the German political parties have rejected the use of bots in their electoral campaign for the general elections). Moreover, bots are commonly related to fake news spreading, although also humans often play a key role in disseminating false claims. Therefore, in 2020 at PAN we will be addressing the problem of profiling those authors that are more likely to spread fake news In Twitter. In this talk I will briefly describe the findings of the these years'PAN author profiling shared tasks
Bio.: Paolo Rosso is full professor at the Universitat Politecnica de Valencia, Spain where he is also member of the PRHLT research center. His research interests focus mainly on author profiling, irony detection, fake reviews detection, plagiarism detection, and recently hate speech and fake news detection. Since 2009 he has been involved in the organisation of PAN benchmark activities at CLEF and at FIRE evaluation forums, mainly on plagiarism / text reuse detection and author profiling. At SemEval he has been co-organiser of shared tasks on sentiment analysis of figurative language in Twitter (2015), and on multilingual detection of hate speech against immigrants and women in Twitter (2019). He is co-ordinator of the activities of FIRE and IberEval evaluation forums. He has been PI of national and international research projects funded by EC and U.S. Army Research Office. At the moment, in collaboration with Carnegie Mellon University, he is involved in a project funded by Qatar National Research Fund on author profiling for cyber-security. He serves as deputy steering committee chair for the CLEF conference and as associate editor for the Information Processing & Management journal. He has been chair of *SEM-2015, and organisation chair of CERI-2012,CLEF-2013 and EACL-2017. He is the author of 400+ papers, published in journals, book chapters, conference and workshop proceedings.
Keynote talk II
Prof. Jan Treur
Artificial Intelligence in the Social AI Group of the Vrije Universiteit Amsterdam, The Netherlands
Talk Title: Network-Oriented Modeling for Simulation and Analysis of a Dynamic, Adaptive and Evolving World
Abstract: Network-Oriented Modeling is a more and more often used approach to model a wide variety of phenomena in the world resulting in, for example, biological networks, neural networks, mental networks, economic networks, and social networks or integrations thereof. As the world often can be dynamic, adaptive and evolving, a challenge is to design networks that can address that in a suitable and transparent manner. Although networks by themselves have a transparent declarative structure based on nodes and connections between them, to model non-static aspects, by tradition procedural algorithmic, programming-like specifications are added, which leads to a less transparent hybrid modeling format. In the current paper a modeling approach is described which still uses declarative mathematical notions to specify the non-static aspects as well in a transparent network oriented manner. Its applicability is illustrated by addressing a number of challenging examples of network models for dynamic, adaptive and evolutionary phenomena such as a social network model for opinion dynamics, a mental network model showing plasticity and metaplasticity, and an evolutionary network model showing disgust during first-trimester pregnancy as a fourth-order evolutionary adaptation.
Bio.: Jan Treur is a full professor of Artificial Intelligence in the Social AI Group of the Vrije Universiteit Amsterdam. He is an internationally well-recognized expert in human-directed AI and cognitive and social modelling. His research during the past 15 years covers methods and techniques for modelling and analysis of human-directed AI systems in a number of application areas, including human-aware AI systems and cognitive and social modelling and simulation. Currently his research focuses on Network-Oriented Modeling for (multi-order) adaptive networks to model adaptive biological, cognitive, affective and social processes, with a book dedicated to this published in 2020. Modeling, simulation and analysis addresses parts of reality in the context of dynamic and adaptive aspects of human functioning as described in scientific disciplines such as Biology, Cognitive Neuroscience, Cognitive Science, Social Neuroscience and Social Sciences. personal website: https://www.researchgate.net/profile/Jan_Treur