Perspectives on Neuro-Symbolic AI
Speakers: Christoph Wehner (Sony AI; University of Bamberg), Olga Mashkova (KAUST),Samy Badreddine (Sony AI; Fondazione Bruno Kessler)
Date: September 16, 2024
Where: Ciências ULisboa, 6.3.27
Invited by: Cátia Pesquita
10:00 – Olga Mashkova
Title: Enhancing Geometric Ontology Embeddings for EL++ with Negative Sampling and Deductive Closure Filtering
Abstract: This talk presents methods to improve EL++ ontology embeddings using negative sampling and deductive closure, addressing limitations in existing models that blur the line between unprovable and provably false statements. Novel approaches to negative losses and ontology deductive closure will be discussed, demonstrating improvements in knowledge base completion.
Short Bio: I am a PhD student of computer science in Bio-Ontology Research Group under the supervision of Professor Robert Hoehndorf at KAUST interested in mathematical and programming applications to natural sciences. In 2023 I graduated from Moscow State University, Faculty of Mechanics and Mathematics, Department of Logic and Theory of Algorithms. In 2021 I got an additional degree of professional retraining in the field of algorithmic bioinformatics. In period of 2020-2023 I worked as a research assistant in the Institute for System Programming of the Russian Academy of Sciences; my research there was related to ECG signal classification with deep learning techniques and ECG delineation. My current research interests are related to the development of ontology embedding methods and knowledge-enhanced learning with application to natural sciences.
10:50 – Christoph Wehner
Title: Explaining Knowledge Graph Embedding Models
Abstract: Christoph will introduce KGExplainer, a post-hoc explainable AI method for Knowledge Graph Embedding (KGE) models, designed to reveal the underlying subgraph structures and patterns driving predictions. The talk will also cover criteria such as faithfulness and localization in explainable AI and their importance for the transparency of KGE models.
Short Bio: Christoph is a PhD student at the Cognitive Systems Group, University of Bamberg, and a Data Scientist at Sony AI Barcelona. His research focuses on explainable and interactive link prediction methods in knowledge graphs, with applications in manufacturing and AI for science.
11:40 – Samy Badreddine
Title: What Models for What Queries: A General Look into Link Prediction in Knowledge Graphs
Abstract: Samy will provide an overview of various tasks in link prediction, a key method for expanding and refining Knowledge Graphs. He will compare classifiers, energy-based models, and a more recent approach using Probabilistic Circuits, offering insights into selecting the right model for specific link prediction tasks.
Shot Bio: Samy is a Research Scientist at Sony AI in Barcelona and a PhD candidate at Fondazione Bruno Kessler in Italy. His research interests span Neurosymbolic AI and Probabilistic Machine Learning, with applications in biomedical Knowledge Graphs. He is also the developer of the Logic Tensor Networks library.