Language Models and Symbolic Al скачать в хорошем качестве

Language Models and Symbolic Al

Language models have been around for a long time, forming a foundational component of various Natural Language Processing systems (NLP). In recent years, however, advances in neural networks have given rise to powerful, masked, large-scale neural language models that capture vast amounts of textual semantic information. While these models demonstrate remarkable results in NLP tasks, they still face challenges in reasoning, dynamic knowledge access, and interpretability. Conversely, language models can enhance symbolic systems through tasks like knowledge graph completion. Mehwish Alam is an Associate Professor and is leading a research group in Neurosymbolic Artificial Intelligence at Télécom Paris, Institut Polytechnique de Paris, France. Her research topics include Language Models, Natural Language Processing, Graph Learning, Machine/Deep Learning, and Knowledge Graphs. In her talk at the research colloquium at ZB MED Mehwish Alam explores the intersection of language models and symbolic AI, highlighting various algorithms and frameworks developed based on language models for knowledge graph completion, taxonomy refinement, and other downstream tasks such as author name disambiguation. This talk also gives an overview of the ongoing research in the field. #largelanguagemodels #LLM #KnowledgeGraphs 00:00 Introduction 01:30 Evolution of Language Models 05:32 Large Language Models (LLMs) & Knowledge Graphs (KGs) 09:05 Knowledge Graph Embeddings and Multimodal KGs 13:35 Extracting Entity Context & Predicate Frequency 15:35 MADLINK & Comparison with other Text-based Models 18:58 LiterallyWikidata – A Benchmark for KG Completuin using Literals 21:46 Inducitve Link Prediction 25:32 Literally Author Name Disambiguation (LAND) 30:33 Evaluation of Knowledge Graph Embedding Models 32:30 From Factual Knowledge to Type Information 38:02 Taxonomic Information on Knowledge Graphs? 45:40 Enriching Taxonomies with LLMs 48:50 Missing of semantics 52:30 Structured Data ------ ZB MED – Informationszentrum Lebenswissenschaften wird gefördert durch das Ministerium für Kultur und Wissenschaft des Landes Nordrhein-Westfalen. Dieses Video ist lizenziert unter einer Creative Commons Namensnennung 4.0 International Lizenz - CC BY 4.0.