Machine Learning and Knowledge Representation


Ernest Davis   Yann LeCun   Mehryar Mohri   Foster Provost   David Sontag

Machine learning is concerned with developing of mathematical foundations and algorithm design needed for computers to learn, that is, to adapt their responses based on information extracted from data. For example, learning algorithms may allow a robot to navigate an unknown environment, improving its performance as it acquires more and more data, or a voice-controlled system to improve its recognition of a person's speech after analysis of a sufficient number of samples. Machine learning techniques draw on many fundamental areas from statistics to theoretical computer science, and are used in a broad variety applications: robotics, speech analysis, finance, computer games, handwriting recognition to name just a few.

Ernest Davis studies the problem of representing commonsense knowledge: that is, the problem of taking the basic knowledge about the real world that is common to all humans; expressing it in a form that is systematic enough to be used by a computer program; and providing the program with techniques for effectively using that knowledge. The primary focus of his work is on spatial and physical reasoning (for example, a formalization of commonsense physical reasoning must capture our intuition about qualitative object behavior under gravity). Other directions of work include reasoning about knowledge, belief, plans, and goals, and their interaction with physical reasoning. Development of formal representations of this type has many potential applications in robotics, product and process design and automated training.

Yann LeCun's research interests include the fundamental and practical aspects of machine learning. The main goal of his research is to devise methods through which computers can extract knowledge and automatically acquire "skills" from massive datasets or from experience. Application of learning to perception is a central theme of his work: how can we teach machines to detect and recognize everyday objects in images, how to teach robots to navigate and avoid obstacles solely from visual input. LeCun's group works on a number of fundamental techniques (energy-based models, "deep learning", relational graphical models, and others) which are applied solve problems in computer vision, robotics, image and signal processing, bioinformatics, medical informatics, and economics. LeCun's group also works with the Center for Neural Science on computational models of biological learning.

Mehryar Mohri's primary research areas are machine learning, theory and algorithms, text and speech processing, and computational biology. This includes in particular the study of the theoretical aspects of machine learning, the design of general and accurate learning algorithms, and their applications to large-scale learning problems such as those found in bioinformatics and language processing.

Foster Provost studies data mining, knowledge systems, and machine learning and their alignment with business problems. He has applied advanced technologies to a variety of business problems, including fraud detection and customer contact management. His research directions include robust modeling in the face of imprecision in the business environment, and profiling/monitoring on-line activity.

David Sontag's research interests include theoretical and practical aspects of machine learning and probabilistic inference, with applications to medicine, natural language processing, and information retrieval. In particular, he studies structured prediction, approximate inference in graphical models, unsupervised learning of latent variable models, and computational trade-offs between learning and inference.

Related Web Pages

Computational and Biological Learning Lab   ML Research Group   Machine Learning Seminar  

top | contact webmaster