BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Exact Learning - Ben W.J. Irwin
DTSTART:20200622T153000Z
DTEND:20200622T160000Z
UID:TALK142690@talks.cam.ac.uk
CONTACT:Bingqing Cheng
DESCRIPTION:We present a collection of mathematical tools and emphasise a\
 nfundamental representation of analytic functions. Connecting these\nconce
 pts leads to a framework for `exact learning'\, where an unknown\nnumeric 
 distribution or function could in principle be assigned an exact\nmathemat
 ical description. This is a new\nperspective on machine learning with pote
 ntial applications in all\ndomains of the mathematical sciences and the ge
 neralised representations\npresented here have not yet been widely conside
 red in the context of\nmachine learning and data analysis. The moments of 
 a multivariate\nfunction or distribution are extracted using a Mellin tran
 sform and the\ngeneralised form of the coefficients is trained assuming a 
 highly\ngeneralised Mellin-Barnes integral representation. The fit functio
 ns\nuse many fewer parameters contemporary machine learning methods and an
 y\nimplementation that connects these concepts successfully will likely\nc
 arry across to non-exact problems and provide approximate solutions. We\nc
 ompare the equations for the exact learning method with those for a\nneura
 l network which leads to a new perspective on understanding\nwhat a neural
  network may be learning and how to interpret the\nparameters of those net
 works.\n\nA pre-print can be found at:\nhttps://papers.ssrn.com/sol3/paper
 s.cfm?abstract_id=3596606
LOCATION:virtual ZOOM meeting ID: 263 591 6003\, https://zoom.us/j/2635916
 003
END:VEVENT
END:VCALENDAR