BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hyperparameter Optimisation - Ross Clarke\, Erik Daxberger\, Austi
 n Tripp
DTSTART:20191106T140000Z
DTEND:20191106T153000Z
UID:TALK134407@talks.cam.ac.uk
CONTACT:Robert Pinsler
DESCRIPTION:As machine learning models become more complex\, so do the spa
 ces of their hyperparameters – those parameters not directly concerned w
 ith the model's adaptability\, but rather the design of its training metho
 d or architecture. With the exposure of pathological comparability issues 
 in the literature\, and the suggestion that many respected contributions s
 ee marked improvement under a more systematic configuration selection stra
 tegy\, hyperparameter optimisation is fast becoming an essential component
  of the research workflow. In this talk\, we will survey the evolution of 
 modern hyperparameter optimisation\, from the key elements of fundamental 
 algorithms\, such as Bayesian optimisation\, to state-of-the-art methods f
 or intelligent and efficient optimisation\, such as BOHB. We will also dis
 cuss the practicalities of implementing hyperparmeter optimsiation in rese
 arch projects\, including an overview of suitable libraries and off-the-sh
 elf implementations.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR