BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Multitask Learning - Prof Massimiliano Pontil - University College
  London
DTSTART:20140507T130000Z
DTEND:20140507T140000Z
UID:TALK51143@talks.cam.ac.uk
CONTACT:David Greaves
DESCRIPTION:Machine learning studies the problem of learning to perform\na
  given task from a dataset of examples. A fundamental limitation of\nstand
 ard machine learning methods is the cost incurred in preparing\nlarge trai
 ning datasets. Often in applications a limited number of\nexamples is avai
 lable and the task cannot be solved in isolation. A\npotential remedy is o
 ffered by multitask learning\, which aims to learn\nseveral related tasks 
 simultaneously. If the tasks share some\nconstraining or generative proper
 ty which is sufficiently simple it\nshould allow for better learning of th
 e individual tasks even when the\nindividual training datasets are small. 
 In the talk\, I will present a\nwide class of multitask learning methods w
 hich encourage different\nforms of task relatedness and involve certain no
 tions of structured\nsparsity and low rank tensor representations. I will 
 also discuss\niterative algorithms to implement these methods\, building u
 pon ideas\nfrom convex optimisation. Finally\, I will illustrate the perfo
 rmance\nof these methods in applications arising in affective computing\,\
 ncomputer vision and user modelling.\n
LOCATION:Lecture Theatre 1\, Computer Laboratory
END:VEVENT
END:VCALENDAR