EE Seminar: Dr. Raviv Raich

~~(The talk will be given in English)

Speaker: Dr. Raviv Raich,
Associate Professor, School of Electrical Engineering and Computer Science,
Oregon State University, Corvallis, OR 97331-5501

Thursday, January 1st, 2015
13:45 - 14:45
Room 011, Kitot Bldg., Faculty of Engineering

Efficient instance annotation in multiple-instance multiple-label learning
Abstract
Recent work in machine learning focuses on the framework of multiple instance multiple label learning. In this setting, objects consist of multiple components termed instances. For example, an image may consist of multiple segments.
Moreover, objects are labeled by a set of labels. For examples, images can be labels with a list of objects present in the image (e.g., 'sky' and 'building'). The multiple instance multiple label framework is considered for a variety of
application areas including computer vision, acoustic scene analysis, and activity recognition. Bag level prediction, i.e., the prediction of the label set of a multi-instance object has been a focus of research for over a decade. Our interest
is in the somewhat less studied problem of instance annotation, the problem of associating each instance with a label. A novel model and computationally efficient inference, are introduced. The model can be used for both bag level
prediction and instance annotation. Empirical evaluations show that the approach outperforms state-of-the-art methods for instance annotation on a collection of both synthetic and real datasets.