Mark Schmidt

  Mark Schmidt

    Department of Computer Science
    University of British Columbia
    201-2366 Main Mall
    Vancouver, BC.
    V6T 1Z4

    604-822-6421 (Lab)
    778-899-5071 (Cell)
   


I'm currently a PhD Candidate in the Department of Computer Science at UBC, working in the Laboratory for Computational Intelligence under the supervision of Kevin Murphy. I used to be at the University of Alberta working for the Alberta Ingenuity Centre for Machine Learning under Russ Greiner as part of the Brain Tumor Analysis Project group. In the summer of 2006, I worked with the Computer-Aided Diagnosis and Therapy Group at Siemens Medical Solutions.

Research

  • List of Publications - with links to related software, presentations/posters, and appendices.

    Recent work:
  • Reviewing optimization methods for L1-regularized optimization
  • Applying non-DAG models for causal prediction
  • Directed cyclic probabilistic graphical models for interventional data
  • Learning the variable groups in blockwise-sparse Gaussian graphical models
  • Discriminative parameter estimation in level-set segmentation methods
  • Large-scale optimization of costly objective functions with simple constraints
  • Group variable selection and structure learning in undirected graphical models
  • Interior-point stochastic approximation methods
  • Learning directed graphical model structures with L1-regularization paths
  • Stochastic gradient methods for training conditional random fields

    Recent applications:
  • Predicting the effects of new combinations of inhibitory and excitatory interventions
  • Modeling purturbations in multivariate intracellular flow cytometry
  • Finding groups of correlated mutual funds
  • Computed tomography muscle segmentation
  • Network models of gene expression data
  • Heart motion abnormality detection
  • E-mail spam classification
  • Network models of newsgroup topics and webpage visits
  • Biomedical named-entity recognition and man-made structure detection
  • Automatic brain tumor segmentation

    Code

  • Matlab Software - list of Matlab software, with descriptions.

    Some Highlights:
  • minConF - Functions for optimization of differentiable real-valued multivariate functions with simple constraints.
  • PQN - Code for optimization of (costly) differentiable multivariate functions over simple convex sets (Examples).
  • UGMlearn - Code for structure learning in discrete-state undirected graphical models using group L1-regularization.
  • crfChain - Learning/inference/decoding/sampling in chain-structured conditional random fields with categorical features.
  • GeneralL1 - Functions implementing strategies for minimizing unconstrained functions subject to L1 regularization (Examples).
  • DAGLearn Functions for maximum a posteriori (MAP) estimation of Gaussian/sigmoid directed acyclic graphical (DAG) models.
  • blogreg - Functions for MCMC simulation of binary probit/logistic regression posterior distributions over parameters.
  • CRF2D - Code for inference and learning in 2D grid-structured binary conditional random fields (CRFs).
  • lasso - Functions implementing a variety of the methods available to solve 'LASSO' regression (and basis selection) problems.
  • minFunc - Function for unconstrained optimization of differentiable real-valued multivariate functions (Examples).

    Miscellaneous

  • Notes on Polar Cones - Notes from a presentation on polar cones and convex non-smooth constrained optimality conditions.
  • Notes on Linear Algebra - Slides from a review of basic concepts in linear algebra.
  • IBM talk - Slides from a talk at IBM research that goes over PQN and other recent projects.
  • CS Refresher Courses - A series of informal sessions with brief reviews of concepts and/or software tools.
  • Structural SVMs - An introduction to structural support vector machines, and an overview of methods for solving them.
  • Lasso Duals - Derivations of two problems that are dual to minimizing squared error with a penalty on the L1-norm of the coefficients.
  • Value Function Approximation - Presentation on function approximation for the course 'Approximate Dynamic Programming'.
  • CRF Mini-Tutorial - Slides for part 1 of a mini-tutorial on Conditional Random Fields
  • Matlab on Arrow - tutorial for running Matlab jobs on the Arrow cluster.