Drawing a box around an intended segmentation target has become both a popular user interface and a common output for learning-driven detection algorithms. Despite the ubiquity of...
We present a general Bayesian framework for hyperparameter tuning in L2-regularized supervised learning models. Paradoxically, our algorithm works by first analytically integratin...
Graphical models are usually learned without regard to the cost of doing inference with them. As a result, even if a good model is learned, it may perform poorly at prediction, be...
Multiple-instance Learning (MIL) is a new paradigm
of supervised learning that deals with the classification of
bags. Each bag is presented as a collection of instances
from whi...
Zhouyu Fu (Australian National University), Antoni...
Abstract. A simplified color image formation model is used to construct an algorithm for image reconstruction from CCD sensors samples. The proposed method involves two successive ...