Research

Research Interests

Broadly speaking I'm interested in harnessing machine learning algorithms and related probabilistic techniques to aid in solving real-world problems.

Most of my work thus far has focused on aspects of document image analysis. Earlier research involved the application of various sequential learning algorithms to carry out page segmentation and region labelling tasks. My MSc thesis work has focused on font independent approaches to optical character recognition that take a top-down, code-breaking style attack to discern symbol labels (completely ignoring glyph shape and font during recognition).

Publications

  • Shape-Free Statistical Information in Optical Character Recognition
    Scott Leishman (2007).
    Master's Thesis, University of Toronto
    Available here: PDF (6.4mb)
  • A Statistical Learning Approach To Document Image Analysis
    Kevin Laven, Scott Leishman, Sam Roweis (2005).
    8th International Conference on Document Analysis and Recognition (ICDAR 2005) pp 357-361
    Available here: PDF (376kb)

Talks

  • Keyboard Acoustic Emanations
    Machine Learning Tea Talk, University of Toronto
    May 3rd, 2006
    Available here: PDF (1.1mb)
  • Cryptogram Decoding for Optical Character Recognition
    2006 CIAR NCAP Summer School
    August 6th, 2006
    Available here: PDF (1.7mb)
    Other talk slides: here
  • The Integral Image Trick for Efficient Rectangular Sum and Area Estimation
    Machine Learning Tea Talk, University of Toronto
    March 28th, 2007
  • Shape-Free Statistical Information in Optical Character Recognition
    Machine Learning Seminar, University of Toronto
    April 2nd, 2007
    Abstract
    Talk slides here: PDF (2.5mb)

Collaborators

  • Sam Roweis: He's my current supervisor, and collaborator on JTAG
  • Kevin Adam Laven: We both did work on JTAG, which he later used in his MSc thesis