BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Memento EPFL// BEGIN:VEVENT SUMMARY:IC Colloquium: Learning Models over Relational Databases DTSTART:20191007T161500 DTEND:20191007T173000 DTSTAMP:20260406T235858Z UID:664ec50a7a668033c371acde1435fb3b8c270d3e26284c4a430b6f64 CATEGORIES:Conferences - Seminars DESCRIPTION:By: Dan Olteanu - University of Oxford\nVideo of his talk\n\nA bstract:\nIn this talk\, I will make the case for a first-principles appro ach to machine learning over relational databases that exploits recent dev elopment in database systems and theory. The input to learning classificat ion and regression models is defined by feature extraction queries over re lational databases. The mainstream approach to learning over relational da ta is to materialize the training dataset\, export it out of the database\ , and then learn over it using statistical software packages. These three steps are expensive and unnecessary. Instead\, one can cast the machine le arning problem as a database problem by decomposing the learning task into a batch of aggregates over the feature extraction query and by computing this batch over the input database. Ongoing results show that the performa nce of this approach benefits tremendously from structural properties of t he relational data and of the feature extraction query\; such properties m ay be algebraic (semi-ring)\, combinatorial (hypertree width)\, or statist ical (sampling). It also benefits from factorized query evaluation and que ry compilation. For a variety of models\, including factorization machines \, decision trees\, and support vector machines\, this approach may come w ith lower computational complexity than the materialization of the trainin g dataset used by the mainstream approach. This translates to several orde rs-of-magnitude speed-up over state-of-the-art systems such as TensorFlow\ , R\, Scikit-learn\, and mlpack.\n\nThis work is part of the FDB project (https://fdbresearch.github.io) and based on collaboration with Maximilia n Schleich (Oxford)\, Jakub Zavodny (Oxford)\, Milos Nikolic (Edinburgh)\, Mahmoud Abo-Khamis\, Ryan Curtin\, Hung Q. Ngo (RelationalAI)\, Ben Mosel ey (CMU)\, and XuanLong Nguyen (Michigan).\n\nBio:\nDan Olteanu is Profes sor of Computer Science at the University of Oxford and Computer Scientist at RelationalAI. He received his PhD from the University of Munich in 200 5. He spends his time understanding hard computational challenges around d ata processing and designing simple and scalable solutions towards these c hallenges. He has published over 70 papers in the areas of database system s\, AI\, and theoretical computer science\, contributing to XML query proc essing\, incomplete information and probabilistic databases\, factorised d atabases\, scalable and incremental in-database optimisation\, and the com mercial systems LogicBlox and RelationalAI. He co-authored the book « Pro babilistic Databases » (2011). He has served as associate editor for PVLD B and IEEE TKDE\, as track chair for IEEE ICDE’15\, group leader for ACM SIGMOD’15\, vice chair for ACM SIGMOD’17\, and co-chair for AMW’18\ , and he is currently serving as associate editor for ACM TODS and the SIG MOD Record Database Principles column. He is the recipient of an ERC Conso lidator grant (2016)\, an Oxford Outstanding Teaching award (2009)\, and t he ICDT 2019 best paper award.\n\nMore information LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 STATUS:CONFIRMED END:VEVENT END:VCALENDAR