BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:GIBBON: General-purpose Information-Based Bayesian OptimisatioN - 
 Henry Moss\, Senior ML Researcher at Secondmind
DTSTART:20220208T150000Z
DTEND:20220208T160000Z
UID:TALK169607@talks.cam.ac.uk
CONTACT:Andrei Paleyes
DESCRIPTION:Bayesian optimisation (BO) is a popular routine for optimising
  functions that are plagued by high evaluation costs. In BO\, an acquisiti
 on function is used to sequentially focus evaluations into promising areas
  of the search space\, allowing the identification of competitively good s
 olutions within even heavily restricted evaluation budgets. A recent parti
 cularly intuitive and empirically effective class of acquisition functions
  has arisen based around information theory. During this talk\, I will mot
 ivate these powerful entropy-based methods\,  before describing a general-
 purpose extension built around a simple approximation of information gain 
 -- an information-theoretic quantity central to solving a range of BO prob
 lems\, including noisy\, multi-fidelity and batch optimisations. Previousl
 y\, these tasks have been tackled separately within information-theoretic 
 BO\, each requiring a  different sophisticated approximation scheme\, exce
 pt for batch BO\, for which no computationally-lightweight information-the
 oretic approach has previously been proposed. GIBBON (General-purpose Info
 rmation-Based Bayesian OptimisatioN) provides a single principled framewor
 k suitable for all the above\, out-performing existing approaches whilst i
 ncurring substantially lower computational overheads.
LOCATION:SS03\, Computer Lab
END:VEVENT
END:VCALENDAR