ML/Google Distinguished Lecture - cs.cmu.edu
1
Abstract: ML/Google Distinguished Lecture October 8, 2014 10:30am Newell-Simon Hall 3305 Joint seminar by the Machine Learning Department at Carnegie Mellon University & Google Pittsburgh Hal Daume III Associate Professor, Computer Science, University of Maryland, College Park Algorithms that Learn to Think on their Feet The classic framework of machine learning is: example in, prediction out. This is great when exampleare fully available at all times, and when all parts of an example are relevant for making a prediction. But it is very different from how humans reason. We get some information. We may make a prediction. Or we may decide we need to get more information. For us, it's worth spending effort to acquire more data when making hard and important decisions (e.g., foreign policy); it is not on easy or low-cost decisions (e.g., afternoon snacks). I'll discuss some of our recent work that imbues in machine learning algorithms the notion of information cost, value, and time. I'll show examples from three settings in natural language processing: classification, natural language parsing, question answer and simultaneous machine translation. The last is the problem of incrementally producing a translation of a foreign sentence before the entire sentence is "heard" and is challenging even for well-trained humans. This is joint work with a number of fantastic collaborators: Jordan Boyd-Graber, Leonardo Claudino, Jason Eisner, Lise Getoor, Alvin Grissom II, He He, Mohit Iyyer, John Morgan, Jay Pujara and Richard Socher. BIO: Hal Daumé III is an associate professor in Computer Science at theUniversity of Maryland, College Park. He holds joint appointments in UMIACS and Linguistics. He was previously an Assistant Professor in the School of Computing at the University of Utah. His primary research interest is in developing new learning algorithms for prototypical problems that arise in the context of language processing and artificial intelligence. This includes topics like structured prediction, domain adaptation and unsupervised learning; as well as multilingual modeling and affect analysis. He associates himself most with conferences like ACL, ICML, NIPS and EMNLP. He earned his PhD at the University of Sounthern California with a thesis on structured prediction for language (his advisor was Daniel Marcu). He spent the summer of 2003 working with Eric Brill in the Machine Learning and applied statistics group at Microsoft Research. Prior to that, he studied math (mostly logic) at Carnegie Mellon University.
Transcript of ML/Google Distinguished Lecture - cs.cmu.edu
Abstract:
ML/GoogleDistinguished Lecture
October 8, 2014 10:30amNewell-Simon Hall 3305Joint seminar by the Machine Learning Department at Carnegie Mellon University & Google Pittsburgh
Hal Daume IIIAssociate Professor, Computer Science, University of Maryland, College Park
Algorithms that Learn to Think on their Feet
The classic framework of machine learning is: example in, prediction out. This is great when exampleare fullyavailable at all times, and when all parts of an example are relevant for making a prediction. But it is very differentfrom how humans reason. We get some information. We may make a prediction. Or we may decide we need toget more information. For us, it's worth spending effort to acquire more data when making hard and importantdecisions (e.g., foreign policy); it is not on easy or low-cost decisions (e.g., afternoon snacks).
I'll discuss some of our recent work that imbues in machine learning algorithms the notion of information cost, value, and time. I'll show examples from three settings in natural language processing: classification, naturallanguage parsing, question answer and simultaneous machine translation. The last is the problem ofincrementally producing a translation of a foreign sentence before the entire sentence is "heard" and ischallenging even for well-trained humans.
This is joint work with a number of fantastic collaborators: Jordan Boyd-Graber, Leonardo Claudino, Jason Eisner, Lise Getoor, Alvin Grissom II, He He, Mohit Iyyer, John Morgan, Jay Pujara and Richard Socher.
BIO: Hal Daumé III is an associate professor in Computer Science at theUniversity of Maryland, College Park. He holds joint appointments in UMIACS and Linguistics. He was previously an Assistant Professor in the School of Computing at the University of Utah. His primary research interest is in developing new learning algorithms for prototypical problems that arise in the context of language processing and artificial intelligence. This includes topics like structured prediction, domain adaptation and unsupervised learning; as well as multilingual modeling and affect analysis. He associates himself most with conferences like ACL, ICML, NIPS and EMNLP. He earned his PhD at the University of Sounthern California with a thesis on structured prediction for language (his advisor was Daniel Marcu). He spent the summer of 2003 working with Eric Brill in the Machine Learning and applied statistics group at Microsoft Research. Prior to that, he studied math (mostly logic) at Carnegie Mellon University.
