2008 Poster Sessions : Inferring Inhomogeneous Lattices from Noisy Observations

Student Name : Farshid Moussavi, Fernando Amat, and Sewoong Oh
Advisor : Andrea Montanari
Research Areas: Information Systems
Understanding the underlying structure of a real image is a challenging task. We are interested in identifying repeated structures from noisy observations of curved surfaces with smooth deformations. We formulate the lattice-finding as a inference problem on probabilistic distribution over the set of all possible structures. The inference algorithm finds a plausible lattice by iteratively proposing assignments that maximize the distribution.

Farshid Moussavi is a PhD candidate in the Department of Electrical Engineering at Stanford. Since 2005, he is in the research group of Professor Mark Horowitz, working on applications of probabilistic inference and machine learning to cryogenic electron tomography and biomedical imaging. He holds a Master's degree in electrical engineering from UC Berkeley, a Bachelor's degree (summa cum laude) in electrical engineering from University of Idaho, and has extensive experience in communications and image processing hardware design in various companies like Cisco Systems, Apple Computer, Hewlett Packard, Micron Imaging, and some startups.

Fernando Amat was born in Barcelona, Spain. He did his undergraduate there at the Technical University of Catalonia (UPC). He graduated with a degree in Mathematics and another degree in Telecomunnication Engineering in 2004. He came to Stanford in September, 2004, where he is a PhD student in Electrical Engineering. He is currently working with professor Mark Horowitz in the are are of bioimaging and statistical learning.

Sewoong Oh is a Ph.D. candidate at the Department of Electrical Engineering at Stanford. He is working with Prof. Andrea Montanari in the area of coding theory. He received his Bachelor's degrees in electrical engineering from Seoul National University in Korea in 2002. His current research topic is optimizing code ensemble for finite block length LDPC codes.