Max-Margin Learning of Gaussian Mixtures with Sequential Minimal
Trinh Minh Tri Do and Thierry Artieres
In: ICFHR 2008, 19-21 August 2008, Canada.
This works deals with discriminant training of Gaussian
Mixture Models through margin maximization. We go
one step further previous work, we propose a new formulation
of the learning problem that allows the use of efficient
optimization algorithm popularized for Support Vector
Machines, yielding improved convergence properties
and recognition accuracy on handwritten digits recognition.