Conférenciers invités

Cinq exposés invités ponctueront la conférence :

Xin Luna Dong, Meta

Next-Generation Intelligent Assistants for AR/VR Devices

An intelligent assistant shall be an agent that knows you and the world, can receive your requests or predict your needs, and provide you the right services at the right time with your permission. As smart devices such as Amazon Alexa, Google Home, Meta Ray-ban Stories get popular, Intelligent Assistants are gradually playing an important role in people's lives. The Emergence of AR/VR devices brings more opportunities and calls for the next generation of Intelligent Assistants.
In this talk, we discuss the many challenges and opportunities we face to grow intelligent assistants from server-side to on-device, from voice-only to multi-modal, from context-agnostic to context-aware, and from listening to the users' requests to predicting the user's needs. We also describe the roles public and personal knowledge graphs play to empower such an assistant. We expect these new challenges to open doors to new research areas and start a new chapter for providing personal assistance services.

Chiara Ghidini, Fondazione Bruno Kessler

Data-aware Processes and their Executions: what’s in for Knowledge Representation and Graphs

The worlds of Business Process Management (BPM) and Process Mining (ProM) has had only few connections with those of Information Extraction (IE), Knowledge Management (KM), and Semantic Web (SW). Indeed their intersections amounted in few attempts to model semantic business processes or exploit ontologies, such as the BPMN ontology, to reason on semantically enriched process models. One of the reasons of this distance might lie in the fact that the business process oriented communities were mainly focused on handling temporally oriented entities such as activities and their temporal (work)flows relations, while the knowledge oriented ones were mainly focused on the modelling and handling of static entities and relations. In the last few years nonetheless the two groups have started expanding their interests and this may end up in better connecting with each others. Indeed, the business process communities have started looking more and more towards multi-dimensional processes, characterised by a complex network of entities that go beyond the typical event-based ones and include data objects, resources, actors, goals, among others. At the same time the knowledge oriented one has shown a growing interest in temporally denoted entities such as events, stories and narratives. In this talk I will use some of our works on Semantic Modelling and Analysis of Complex Data-aware Processes and their Executions to try to highlight possible connections between these two worlds and challenges where an interaction may provide mutual benefit.

Bio: Chiara Ghidini is a senior Research Scientist at Fondazione Bruno Kessler (FBK), Trento, Italy, where she heads the Process & Data Intelligence (PDI) research unit and is responsible of the scientific ordination of the new centre of digital Health & Well Being. Her scientific work in the areas of Semantic Web, Knowledge Engineering and Representation, Multi-Agent Systems and Process Mining is internationally well known and recognised, and she has made significant scientific contributions in the areas of multi-context logics; deliberative resource bounded agents; ontology mappings and integration; collaborative modeling platforms, business process modelling, and predictive business process monitoring. She has been involved in a number of international research projects, among which the FP7 Organic.Lingua and SO-PC-Pro European projects and the current network of Excellence Humane-AINet, as well as industrial projects in collaboration with companies in the Trentino area.

Axel Ngonga, Paderborn University

Scaling Machine Learning on Knowledge Graphs

Automated knowledge extraction have engendered a plethora of knowledge graphs, which are used in a large number of applications. Symbolic machine learning on these knowledge graphs has a plethora of advantages. First, this family of approaches is often less data-hungry than sub-symbolic models. Moreover, the models computed in this manner are ante-hoc globally explainable. In this talk, we present some recent results on accelerating symbolic machine learning based on inductive logic programming on knowledge graphs with rich semantics. In particular, we focus on algorithms which improve the runtime of ML approaches while maintaining completeness guarantees. We also discuss some of most pertinent challenges faced by this family of approaches.

Camille Roth, Centre Marc Bloch

Graphes sémantiques et réseaux sociaux

La distribution sociale des informations et la structure des interactions sociales sont de plus en plus fréquemment étudiées de manière conjointe, notamment dans les travaux se réclamant des sciences sociales computationnelles. D'une part, l'analyse des contenus, diversement appelée "text mining", "automated text analysis" ou encore "text-as-data methods", s'y appuie sur un vaste éventail de techniques allant de simples statistiques numériques (similarité textuelle, termes saillants) à des approches d'apprentissage automatique s'appliquant au niveau d'ensembles de mots ou de phrases, en particulier en vue d'extraire divers types de graphes sémantiques – qu'il s'agisse simplement de liens de co-occurrence entre termes, de triplets "sujet-prédicat-objet", ou de structures plus élaborées au niveau d'une phrase entière. Ces données et, parfois, ces graphes sémantiques, sont d'autre part associés à des acteurs dont les diverses relations (interaction, collaboration, affiliation) sont également rassemblés fréquemment au sein de graphes sociaux.  Cette présentation vise à proposer un tour d'horizon des approches mêlant contenus et interactions, où les espaces publics numériques et les communautés scientifiques représentent des terrains privilégiés en tant que systèmes sociaux où informations et savoirs sont produits et se propagent de manière décentralisée.

Bio: Chercheur au CNRS en informatique depuis 2008, Camille Roth a également été enseignant-chercheur en sociologie (professeur à Sciences Po Paris et maître de conférences à Toulouse). Docteur de l’École Polytechnique (2005) et ingénieur des Ponts (2002), ainsi que titulaire d’un DEA de sciences cognitives (EHESS, 2002), il a un profil à la croisée entre sciences dures et sciences sociales. En 2012, il a fondé et dirige depuis lors l'équipe de sciences sociales computationnelles du Centre Marc Bloch à Berlin, où il encadre un groupe interdisciplinaire d'une dizaine de personnes mêlant sciences sociales et modélisation mathématique et informatique. Il y dirige notamment un ERC Consolidator (2018-23) sur le thème du confinement et des bulles dans les espaces publics numériques et y mène également des travaux sur le rôle des algorithmes dans l'accès aux contenus en ligne.

Harald Sack, Karlsruher Institut für Technologie

Symbolic vs Subsymbolic Knowledge Representation, an Epic Dilemma?

Over the last decade, deep learning methods made tremendous progress. Massive parallelization via GPUs, huge training data harvested from the Web, and efficient neural network architectures enable humanlike or even superhuman performance in specific areas. Huge pre-trained language models seem to capture complex semantics of natural languages and obtain outstanding results in classification, prediction, or generation tasks. The same holds for the image generation domain with models like Stable Diffusion or Dall-E. As a result, do we still need symbolic knowledge representations and logics? Will Deep Learning models take over and will symbolic logic, ontologies, or knowledge graphs become an obsolete niche product? In this talk, we will look at various examples from both worlds and show that each by itself alone might fail. Both sides will have to join forces to succeed and move forward.

Bio: Harald Sack is Professor of Information Service Engineering at FIZ Karlsruhe - Leibniz Institute for Information Infrastructure and Karlsruhe Institute of Technology (KIT). After graduating in computer science at the University of the Federal Forces Munich, he worked as a network engineer and project manager in the signal intelligence corps of the German Air Force. In 1997 he became an associated member of the graduate program ‘mathematical optimization’ at the University of Trier and obtained a PhD in computer science in 2002. After working as a postdoctoral researcher at the Friedrich-Schiller-University in Jena, he headed the research group Semantic Technologies and Multimedia Retrieval at Hasso Plattner-Institute for IT-Systems Engineering at the University of Potsdam from 2009 to 2016. His current areas of research include semantic technologies, knowledge discovery as well as applications of hybrid symbolic and subsymbolic AI. He has served as General Chair, PC Chair, and (Senior) PC member of numerous international conferences and workshops. Harald Sack has published more than 200 scientific papers in peer reviewed international journals and conferences including several standard textbooks.