Omid Madani

Research Overview: 1. Recall Systems: Efficient Learning and Use of Indices to Categories, and 2. Prediction Games in Infinitely Rich Worlds"

Abstract

Categorization is fundamental to intelligence, especially when a system can effectively handle many categories. Not surprisingly, numerous applications such as personalization, text prediction, and image tagging, can benefit significantly from efficient learning and categorization in the presence of myriad categories (tens of thousands and beyond). I will give an overview of my research in this theme. In the first part of the talk, I will describe the idea of learning indices to categories. The online learning methods that we have developed are simple and scale well, and I will discuss our positive experience, in terms of accuracy as well as efficiency, with the approach. In the second part of the talk, I will briefly describe some ideas on how to effectively discover and learn to recognize many categories in an unsupervised manner. I have done this work in the text domain, in particular text categorization and prediction in text. I am eager to learn about problems and opportunities in vision.