Professor Learned-Miller’s interests can be broadly categorized as applying ideas and methods from machine learning to problems in machine vision. Problems he has worked on include learning from a small number of examples, independent component analysis, learned color constancy, developing probability models of shape deformation, and mathematical expression recognition. His Ph.D. thesis focuses on using learned statistical knowledge from one visual task to speed learning of a new, related task. The two major types of statistical knowledge used are distributions over shape variability and distributions over joint color variability. He describes a handwritten digit classifier that gets about 90% accuracy using only a single example of each handwritten digit.
No subject area
Bounding the probability of error for high precision optical character recognition (with Gary B. Huang, Andrew Kae, and Carl Doersch), Journal of Machine Learning Research (2012)
We consider a model for which it is important, early in proces sing, to estimate...
Background modeling using adaptive pixelwise kernel variances in a hybrid feature space (with Manjunath Narayana and Allen Hanson), IEEE Conference on Computer Vision and Pattern Recognition (2012)
Recent work on background subtraction has shown de- velopments on two major fronts. In one,...
Distribution fields for tracking (with Laura Sevilla Lara), IEEE Conference on Computer Vision and Pattern Recognition (2012)
Visual tracking of general objects often relies on the assumption that gradient descent of the...
Half-wits: Software techniques for low-voltage probabilistic storage on microcontrollers with NOR flash memory. (with Mastooreh Salajegheh, Yue Wang, Anxiao Jiang, and Kevin Fu), ACM Transactions on Embedded Computing Systems (2012)
Improvements in joint domain-range modeling for background subtraction (with Manjunath Narayana and Allen Hanson), Proceedings of the British Machine Vision Conference (2012)
In many algorithms for background modeling, a distribution over feature values is modeled at each...