Presentations at the CBLL weekly meetings:
# November 3, 2009: "Learning Realistic Human Actions", Ivan Laptev, Marcin Marszalek, Cordelia Schmid, Benjamin Rozenfeld
# May 6, 2009: "Gaussian Process Dynamical Models", Jack Wang, David Fleet, Aaron Hertzmann
# September 24, 2008: "Epileptic Seizure Prediction from EEG": review and our new successful technique
# February 6, 2008: "Temporal Lobe Epilepsy Model", based on research by Fabrice Wendling
# November 22, 2006: "Pyramidal Methods in Image Processing and Recognition and Pyramid Matching Kernels"
# October 5, 2005: "MEG Source Localization Using an MLP with Distributed Output Representation", Sung Chan Jun, Barak Pearlmutter, Guido Nolte
Dynamic Factor Graphs for time series modeling
This is my thesis research project under
Prof. Yann LeCun's supervision, published in
Lecture Notes in Artificial Intelligence and presented at
A DFG includes factors modeling joint probabilities between hidden and observed variables, and factors modeling dynamical constraints on hidden variables. The DFG assigns a scalar energy to each configuration of hidden and observed variables. A gradient-based inference procedure finds the minimum-energy state sequence for a given observation sequence. Because the factors are designed to ensure a constant partition function, they can be trained by minimizing the expected energy over training sequences with respect to the factors' parameters. These alternated inference and parameter updates can be seen as a deterministic gradient-based EM-like procedure.
Using smoothing regularizers, DFGs have been shown to reconstruct chaotic attractors and to separate a mixture of independent oscillatory sources perfectly. DFGs outperformed the best known algorithm on the CATS competition benchmark for time series prediction. DFGs also successfully reconstruct missing motion capture data. For more details, download the article, or follow the video lecture at ECML-PKDD 2009, with corresponding slides.
Belief propagation stereo vision
(Computer Vision class, Spring 2006 with
Prof. Davi Geiger).
The cost function was based on image measurements: intensity accumulation and derivatives
That intensity derivative attribute yielded good results with edgels-based and dynamic programming-based contour detection
Using belief propagation, disparity maps were constructed from cost functions based on intensity derivatives, then the depth maps of the stereo images were used for 3D reconstruction