BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learned Compression - Gergely Flamich and Stratis Markou (Universi
 ty of Cambridge)
DTSTART:20220202T110000Z
DTEND:20220202T123000Z
UID:TALK169034@talks.cam.ac.uk
CONTACT:Elre Oldewage
DESCRIPTION:In recent years there have been\nsignificant advances in using
  machine learning to improve compression algorithms\, a field known as lea
 rned compression. Learned compression improves on traditional compression 
 methods\, by using ML methods to adapt the compression algorithm to the da
 ta at hand\,\noften outperforming the best traditional methods. In this ta
 lk\, we first provide an accessible introduction to traditional compressio
 n methods. We then give an overview of three approaches for performing lea
 rned compression: quantisation-based approaches\, bits-back\ncoding and re
 lative entropy coding. We discuss the strengths of each of these approache
 s\, their limitations and their practicality\, and give example applicatio
 ns for them.\n\nReading list:\n\nBallé\, Johannes\, Valero Laparra\,\nand
  Eero P. Simoncelli. "End-to-end optimized image compression." arXiv\nprep
 rint arXiv:1611.01704 (2016).\n\nTownsend\, James\, Tom Bird\, and\nDavid 
 Barber. "Practical lossless compression with latent variables using bits b
 ack coding." arXiv\npreprint arXiv:1901.04866 (2019).\n\nMaddison\, Chris 
 J.\, Daniel Tarlow\,\nand Tom Minka. "A* sampling." arXiv preprint arXiv:1
 411.0030\n(2014).\n\nOur reading groups are livestreamed via Zoom and reco
 rded for our Youtube channel. The Zoom details are distributed via our wee
 kly mailing list.
LOCATION: Cambridge University Engineering Department \,LR3A
END:VEVENT
END:VCALENDAR