Machine Learning and Friends Lunch





home
past talks
resources
conferences

Probability Models for Joint Row Column Clustering and High Dynamic Range Image Processing


Chris Pal
UMass

Abstract


Probability models and their graphical descriptions provide a
principled way of constructing models for data. As well, graphical
models aid with the visualization and development of algorithms for
optimizing probability models. In this talk I present two information
processing problems inspired by cDNA microarray experiments, their
processing and general image processing. The models I present are
described by graphical models with similar structures and this
facilitates the development of similar algorithms for optimizing the
models. First, I present a model for jointly clustering the rows and
columns of a cDNA microarray data matrix, allowing such a matrix to be
permuted into an approximately block constant form. I present a new
algorithm that proves to be superior to many other alternatives based
on synthetic data experiments and real data experiments for the task
of determining subsets of co-expressed genes. Second, I present a new
method for combining images taken with different camera settings into
a higher dynamic range image using only pixel values. I derive priors
for imaging functions using a generative model of the derivative
structure of functions and tie together generative models and
smoothness regularization. I derive a new optimization technique to
estimate both functions and high dynamic range irradiance estimates.
Finally, I present some compelling image processing results expanding
the dynamic range of digital imagery.

Back to ML Lunch home