Research Project: Metric learning for dynamical models
Recent work in action recognition has exposed the limitations of methods which directly classify local features extracted from spatio-temporal video volumes. In opposition, encoding the actions’ dynamics via generative dynamical models has a number of attractive features: however, using all-purpose distances for their classification does not necessarily deliver good results. We propose a general framework for learning distance functions for generative dynamical models, given a training set of labeled videos. The optimal distance function is selected among a family of pullback ones, induced by a parameterized automorphism of the space of models. We focus here on hidden Markov models and their manifold, and design an appropriate automorphism there. Experimental results are presented which show how pullback learning greatly improves action recognition performances with respect to base distances.
Relevant papers:
  •  Fabio Cuzzolin and Michael Sapienza
    Learning pullback HMM distances
    IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 36, Issue 7, Pages 1483-1489, July 2014
    http://doi.ieeecomputersociety.org/10.1109/TPAMI.2013.181
    Supplementary material
    (I.F. 6.077)
      Supplementary material Pullback Metrics Code
  •  Fabio Cuzzolin, Michael Sapienza, Patrick Esser, Suman Saha, Marloes Franssen, Johnny Collett and Helen Dawes
    Metric learning for Parkinsonian identification from IMU gait measurements
    Gait and Posture, Volume 54, pages 127-132, May 2017
    (I.F. 2.985)
  •   Fabio Cuzzolin
    Manifold learning for multi-dimensional auto-regressive dynamical models
    in Machine Learning for Vision-based Motion Analysis
    L. Wang, G. Zhao, L. Cheng, M. Pietikäine (Eds.), Springer-Verlag, 2010
    Manifold learning for multi-dimensional auto-regressive dynamical models
  • Fabio Cuzzolin and Stefano Soatto
    Learning Riemannian Metrics for Classification of Dynamical Models
    UCLA Technical Report CSD-TR050054, December 17, 2005
Lab Member(s): Michael Sapienza, Fabio Cuzzolin