Deb Roy's Teaching and Seminars

MAS.720 Meaning Machines (Spring 2007)

MAS 921: Proseminar (Fall 2006)

MAS 921: Proseminar (Fall 2005)

MAS 962: Meaning Machines (Spring 2005)

We will examine aspects of knowledge representation and language use by machines and humans. Emphasis will be placed on interactions between physically grounded information structures and functional behavior that provide the foundation for meaningful language use. Ideas from semiotics, philosophy of mind, and cognitive psychology will be brought together with methods from computer science and systems engineering. A final project requiring implementation of a complete cognitive system will provide hands-on experience with ideas discussed in class.

Philosophy of Mind (Spring 2005, at Tufts University, with Professor Daniel Dennett)
(I am teaching a series of guest lectures at Tufts, a sort of companion to my MIT seminar that explores philosophical aspects of building "meaning machines".)

MAS 921: Proseminar (Fall 2004)

MAS 966: Meaning Machines (Spring 2004)

Mondays 1-3 PM, E15-468H (first meeting on Feb 9)
Credits: 3-0-9 (H)

Computers don’t grasp meaning in any deep, human sense. From a computer’s point of view, words are meaningless bits of information processed with speed and precision, but without any genuine interpretation of content. The interpretation of words is left to humans. The absence of meaning in computers is not limited to the domain of words -- it is a widespread limitation of current computing systems.

We will study new approaches to creating language processing systems that grasp meanings in human-like ways. In particular, we will explore interactions between physically grounded knowledge representations and goal-directed behavior that provide the foundation for meaningful language use by machines. Readings from semiotics, philosophy of mind, artificial intelligence, and cognitive science.

MAS 963: Curious Machines (Spring 2003) (Co-taught with Profs. Bruce Blumberg and Cynthia Breazeal)
It is ironic that we are witness today to an abundance of machines that learn, yet we would not consider any of them to be curious. Curiosity is a trait of those natural learning systems, such as people and animals, that learn what they ought to learn when they ought to learn it. It implies a pro-active system that is motivated to learn. It also implies a reflective aspect to the learning process: when to learn, what to learn, from whom, how, and why? How can we build machines that are as curious learners as natural systems? How can we build systems that have a deeper understanding of the learning process beyond turning the statistical crank of a learning algorithm? How can synthetic systems interact with and leverage rich environments that include other agents?

This course examines the issues, principles, and challenges toward building curious machines through lecture, lively discussion, critique of course readings, and student projects. Both natural and synthetic systems are explored to investigate how to build machines that are: (1) Curious, self-motivated and pro-active learners, (2) Reflect upon their learning process in order to initiate learning what they ought to learn when they ought to learn it, and (3) Have learning behaviors that are transparent and understandable to a human instructor. They must be easy to train and teach based on their observable behavior.

MAS 962: Computational Semantics (Fall 2002)
How do words get their meanings? How can word meanings be represented and used by machines? We will explore various approaches to these questions from a computational perspective. Relational / structural methods such as semantic networks represent the meaning of words in terms of their relations to other words. Knowledge of the world through perception and action leads to the notion of external grounding, a process by which word meanings are 'attached' to the world. How an agent theorizes about, and conceptualizes its world provides yet another foundation for word meanings. We will examine each of these perspectives, and consider ways to integrate them.

Topics in Natural Language Processing (Fall 2001)
This course introduces principles and algorithms of naturallanguage processing relevant for both text and spoken language processing.Topics include statistical language modeling, part of speech tagging, robustparsing, frame-based language understanding, language acquisition, andinformation retrieval. Emphasis will be placed on semantic analysis andrepresentations. Selected topics from current research in spoken languageprocessing will also be covered.

MAS 964: Concepts, Language, Embodiment and Learning (Spring 2001)
This seminar explores the role of embodiment and perception in the acquisition of concepts and language by studying and building computational models. Recent trends in cognitive science point towards human physiology as a key to understanding how humans develop conceptual knowledge. These trends have been complemented in the artificial intelligence community with a growing focus on the role of robotics and perceptual computing in developing knowledge representations. This seminar will be structured around (1) a set of case studies of computational/robotic efforts to build embodied communication machines, and (2) a group project aimed at building an embodied language learning system.

MAS 622: Pattern Recognition (Fall 2000)
Introduction to pattern recognition, feature detection, classification; Review of probability theory, conditional probability and Bayes rule; Random vectors, expectation, correlation, covariance; Review of linear algebra, linear transformations; Decision theory, ROC curves, Likelihood ratio test; Linear and quadratic discriminants, Fisher discriminant; Sufficient statistics, coping with missing or noisy features; Template-based recognition, eigenvector analysis, feature extraction; Training methods, Maximum likelihood and Bayesian parameter estimation; Linear discriminant/Perceptron learning, optimization by gradient descent; k-nearest-neighbor classification; Non-parametric classification, density estimation, Parzen estimation; Unsupervised learning, clustering, vector quantization, K-means; Mixture modeling, optimization by Expectation-Maximization; Hidden Markov models, Viterbi algorithm, Baum-Welch algorithm; Linear dynamical systems, Kalman filtering and smoothing, system identification; Bayesian networks, independence diagrams; Decision trees, Multi-layer Perceptrons;

MAS 967: Multilingual Computing (Spring 2000)
How might we build computers which can be used by everyone around the planet? The problems are complex. Thousands of languages are in active use worldwide. Almost a billion people are unable to read or write. Many languages don?t even have a written form. Yet most interface technologies are tied to specific written languages. This course explores aspects of natural language and language technologies related to the theme of multilingual computing. Topics to be covered include: text, speech, and visual input and output; semantic representations; language translation and trans-lingual communication; language learning by humans and machines; open system development of language technologies; and applications of multilingual computing in developing nations. Readings will draw from a broad number of areas including artificial intelligence, speech processing, linguistics, and media studies.




Home Page