BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Cambridge ELLIS Seminar Series- Jasper Snoek - Speaker to be confi
 rmed
DTSTART:20230627T130000Z
DTEND:20230627T140000Z
UID:TALK201622@talks.cam.ac.uk
CONTACT:Kimberly Cole
DESCRIPTION:The Cambridge ELLIS Unit Seminar Series holds talks by leading
  researchers in the area of machine learning and AI. Our next  speaker of 
 2023 will be Dr. Jasper Snoek. Details of his talk can be found below. \n
 \nTitle: “Kernel (GP) Regression with Infinite-Width Neural Networks on 
 Millions of Examples "\n\nAbstract: I'll talk about some of our recent wor
 k on scaling up Gaussian process regression with expressive kernels on str
 uctured data.  Neural kernels have drastically increased performance on d
 iverse and nonstandard data modalities but require significantly more comp
 ute\, which previously limited their application to smaller datasets. In t
 his work\, we address this by massively parallelizing their computation ac
 ross many GPUs. We combine this with a distributed\, preconditioned conjug
 ate gradients algorithm to enable kernel regression at a large scale (i.e.
  up to five million examples). Using this approach\, we study scaling laws
  of several neural kernels across many orders of magnitude for the CIFAR-5
 m dataset. Using data augmentation to expand the original CIFAR-10 trainin
 g dataset by a factor of 20\, we obtain a test accuracy of 91.2\\% (SotA f
 or a pure kernel method). Moreover\, we explore neural kernels on other da
 ta modalities\, obtaining results on protein and small molecule prediction
  tasks that are competitive with SotA methods.\n\n
LOCATION:https://eng-cam.zoom.us/j/89163631728?pwd=R0NxRGVDVVptSmtmRGZDeHl
 lWUtmdz09 
END:VEVENT
END:VCALENDAR