BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Fabian Ruehle (Northeastern University) DTSTART:20240517T070000Z DTEND:20240517T080000Z DTSTAMP:20260423T021020Z UID:UNISTMath/2 DESCRIPTION:Title: Kolmogorov-Arnold Networks\nby Fabian Ruehle (Northeastern Universit y) as part of UNIST Mathematical Sciences Seminar Series\n\n\nAbstract\nWe introduce Kolmogorov-Arnold Networks (KANs) as an alternative to standard feed-forward neural networks. KANs are based on Kolmogorov-Arnold represe ntation theory\, which means for our purposes that we can represent any fu nction we want to learn by a weighted sum over basis functions\, taken to be splines. In contrast to standard MLPs\, the function basis of KANs is f ixed to be piecewise polynomial rather than a combination of weights and n on-linearities\, and we only learn the parameters that control the individ ual splines. While this is more expensive than a standard MLP\, KANs have two properties that can offset this cost. First\, KANs can typically work with much fewer parameters. Second\, they exhibit better neural scaling la ws\, meaning the error decreases faster when increasing the number of para meters as compared to MLPs. Fewer parameters also mean that KANs are much more interpretable\, especially when combined with the sparsification and pruning techniques we introduce. This makes KANs interesting as tools for symbolic regression and for scientific discovery. We discuss an example fr om knot theory\, where we could recover (trivial and non-trivial) relation s among knot invariants.\n\nRegister for zoom link:\nhttps://us06web.zoom. us/meeting/register/tZ MkcOqtrDMuGNAQcMlvp3-MJwcWXVU6fzXl\n LOCATION:https://researchseminars.org/talk/UNISTMath/2/ END:VEVENT END:VCALENDAR