Subspectral Algorithms for Sparse Learning, Optimization & Inference
Speaker: Baback Moghaddam , Mitsubishi Electric Research Laboratories
Date: February 28 2007
Time: 3:00PM to 4:00PM
Location: Star Seminar Room (32-D463)
Host: C. Mario Christoudias, Gerald Dalley, MIT CSAIL
Contact: C. Mario Christoudias, Gerald Dalley, 3-4278, 3-6095, cmch@csail.mit.edu, dalleyg@mit.edu
Relevant URL: http://www.merl.com/projects/spca/ Abstract:
In this talk I will present a general class of "subspectral" algorithms (sparse eigenvector techniques) for solving NP-hard combinatorial optimization problems in three basic domains: (Un)Supervised Learning (e.g. PCA & LDA), Quadratic/Entropic Optimization (e.g. Least-Squares & MaxEnt) and 3) Bayesian Inference (e.g. Automatic Relevance Determination & variational approximations). Efficient algorithms for finding both optimal and greedy solutions are derived using analytic eigenvalue bounds. Sample applications presented are in "sparse PCA" for variable selection (in statistics), "sparse LDA" for classification (gene discovery), sparse kernel regression (robotics & control), sparse quadratic programming (portfolio optimization), graph model selection (sensor networks) and sparse Bayesian inference for computer vision (face recognition & OCR).
Joint work with Yair Weiss (The Hebrew University)
See other events that are part of MIT Machine Vision Colloquium 2006/2007
See other events happening in February 2007
