BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Probabilistic Latent Tensor Factorisation - Taylan Cemgil\, Bogazi
 ci University
DTSTART:20121001T100000Z
DTEND:20121001T110000Z
UID:TALK40428@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:Algorithms for decompositions of matrices are of central impor
 tance in machine learning\, signal processing and information retrieval\, 
 with SVD and NMF (Nonnegative Matrix Factorisation) being the most widely 
 used examples. Probabilistic interpretations of matrix factorisation model
 s are also well known and are useful in many applications (Salakhutdinov a
 nd Mnih 2008\; Cemgil 2009\; Fevotte et. al. 2009). In the recent years\, 
 decompositions of multiway arrays\, known as tensor factorisations have ga
 ined significant popularity for the analysis of large data sets with more 
 than two entities (Kolda and Bader\, 2009\; Cichocki et. al. 2008\; Mohame
 d 2011). We will discuss a subset of these models from a statistical model
 ling perspective\, building upon probabilistic generative models and gener
 alised linear models (McCulloch and Nelder). In both views\, the factorisa
 tion is implicit in a well-defined statistical model and factorisations ca
 n be computed via maximum likelihood.\n\nWe express a tensor factorisation
  model using a factor graph and the factor tensors are optimised iterative
 ly. In each iteration\, the update equation can be implemented by a messag
 e passing algorithm\, reminiscent to variable elimination in a discrete gr
 aphical model. This setting provides a structured and efficient approach t
 hat enables very easy development of application specific custom models\, 
 as well as algorithms for the so called coupled (collective) factorisation
 s where an arbitrary set of tensors are factorised simultaneously with sha
 red factors. Extensions to full Bayesian inference for model selection\, v
 ia variational approximations or MCMC are also feasible. Well known models
  of multiway analysis such as Nonnegative Matrix Factorisation (NMF)\, Par
 afac\, Tucker\, and audio processing (Convolutive NMF\, NMF2D\, SF-SSNTF) 
 appear as special cases and new extensions can easily be developed. We wil
 l illustrate the approach with applications in audio and music processing 
 and link prediction for recommendation.
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7 J J Thomson Av
 enue (Off Madingley Road)\, Cambridge
END:VEVENT
END:VCALENDAR