BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The Coaching - Machine Learning interface - Indoor rowing - Simon 
 Fothergill
DTSTART:20130606T101500Z
DTEND:20130606T111500Z
UID:TALK45603@talks.cam.ac.uk
CONTACT:Henrik Lieng
DESCRIPTION:Human rowing coaches\, who are experienced in describing how a
 thletes move\,\ncould deputise machines to provide similar feedback if the
 y could teach them\nwhat to look for.  Ubiquitous and precise kinetic sens
 ors\, along with\nprobabilistic inference algorithms\, have strengthened t
 he foundations of\nmotor-skill based interactive systems.  However\, these
  systems still leave us\nuncertain over how to feel about a particular per
 formance.\n\nI will discuss a framework for developing machines that rate 
 athletes along\nnatural and emotive scales such as "the importance of impr
 oving how much\n`suspension' they achieve".  After presenting a model of t
 he criteria rowing\ncoaches use to describe their judgements of indoor row
 ing techniques\, I will\npropose an iterative scheme for single criteria t
 hat resolves multiple\njudges’ perspectives on a set of performances into
  a consensus of ratings for each\nperformance.  I will share guidelines fo
 r collecting data on indoor rowing\ntechniques for machine learning\, befo
 re presenting initial evidence in support\nof training Bayesian models to 
 predict human ratings.  I will evaluate\ngenerative filters and linear reg
 ression for four criteria\, showing they give\nmore useful predictions tha
 n 1) 10% to 50% of the human coaches\, 2) randomly\nrating performances (1
 x10-6 < p < 0.38) and 3) performance indicators based on\nNewtonian mechan
 ics (1x10-3 < p < 0.6).
LOCATION:Rainbow Room (SS03)\, Computer Laboratory
END:VEVENT
END:VCALENDAR