Dan Spielman (Yale University) Title: Discrepancy Theory and Randomized Controlled Trials Abstract: Discrepancy theory tells us that it is possible to partition vectors into sets so that each set looks surprisingly similar to every other. By “surprisingly similar” we mean much more similar than a random partition. I will begin by surveying fundamental results in […]
Speaker: Carl Allen and Ivana Balažević - University of Edinburgh School of Informatics Title: Knowledge graph representation: From recent models towards a theoretical understanding Abstract: Knowledge graphs (KGs), or knowledge bases, are large repositories of facts in the form of triples (subject, relation, object), e.g. (Edinburgh, capital_of, Scotland). Many models have been developed to succinctly represent KGs […]
https://youtu.be/OoimTbnSe7I Speaker: Nikunj Saunshi, Dept. of Computer Science, Princeton University Title: A Mathematical Exploration of Why Language Models Help Solve Downstream Tasks Abstract: Autoregressive language models pretrained on large corpora have been successful at solving downstream tasks, even with zero-shot usage. However, there is little theoretical justification for their success. This paper considers the following […]