ML@GT Blog

Jacob Eisentein (Georgia Tech) "Two new machine learning approaches for text classification"

Speaker: Jacob Eisenstein

12n-1pm, Wednesday, Jan 25, 2017 (Lunch will be served at 11:45am)

Marcus Nano Rm 1117-1118

Title: Two new machine learning approaches for text classification

Abstract: 

Text document classification is one of the most well studied applications of machine learning. Yet this technology is still limited by practical difficulties and invalid underlying assumptions.

First, many people who want text classifiers do not have the time or resources to annotate a dataset. They often employ a heuristic alternative: they create word lists for each label class, and then perform prediction by selecting the class whose list matches the largest number of words in the text. This heuristic is theoretically unjustified, and mistakenly assigns the same importance to every word in the list. I show that list-based classification can be viewed as a (very!) special case of Naive Bayes. Based on this analysis, it is possible to estimate weights for each word without supervision, using the method-of-moments.

Second, machine learning approaches to text classification nearly always begin with an IID assumption. Yet words can mean different things to different people, raising the possibility for misunderstandings even in human-human conversation. One potential solution is to relax the IID assumption by personalizing text classifiers to the author. An apparent roadblock is the challenge of obtaining labeled data for each author. I will present a method that sidesteps this requirement by relying on the sociological theory of homophily, which states that people who are socially connected tend to share personal traits. This idea can be formalized by estimating node embeddings for each individual in a social network, and then using these embeddings to drive a social attentional mechanism in a neural ensemble classifier. The resulting system obtains significant improvements on sentiment analysis in Twitter. This project is joint work with Yi Yang.

BIO: 

Jacob Eisenstein is an Assistant Professor in the School of Interactive Computing at Georgia Tech. He works on statistical natural language processing, focusing on computational sociolinguistics, social media analysis, discourse, and machine learning. He is a recipient of the NSF CAREER Award, a member of the Air Force Office of Scientific Research (AFOSR) Young Investigator Program, and was a SICSA Distinguished Visiting Fellow at the University of Edinburgh. His work has also been supported by the National Institutes for Health, the National Endowment for the Humanities, and Google. Jacob was a Postdoctoral researcher at Carnegie Mellon and the University of Illinois. He completed his Ph.D. at MIT in 2008, winning the George M. Sprowls dissertation award. Jacob's research has been featured in the New York Times, National Public Radio, and the BBC. Thanks to his brief appearance in If These Knishes Could Talk, Jacob has a Bacon number of 2.

Tuesday, January 24, 2017 - 15:26



IDEaS and ARC DL: Jon Kleinberg "Human Decisions and Machine Predictions"

IDEaS and ARC Distinguished Lecture

As part of ARC10: Celebrating 10 years of the Algorithms and Randomness Center

Monday, October 24 at 10 AM

Jon Kleinberg (Cornell University)

Human Decisions and Machine Predictions

Klaus Advanced Computing Building Room 1116

Abstract

An increasing number of domains are providing us with detailed trace data on human decisions, often made by experts with deep experience in the subject matter. This provides an opportunity to use machine-learning prediction algorithms to ask several families of questions -- not only about the extent to which algorithms can outperform expert-level human decision-making in specific domains, but also whether we can use algorithms to analyze the nature of the errors made by human experts, to predict which instances will be hardest for these experts, and to explore some of the ways in which prediction algorithms can serve as supplements to human decision-making in different applications.  In this talk, I'll explore this theme by drawing on a line of recent projects; all are joint with Sendhil Mullainathan, and include collaborations with Ashton Anderson, Himabindu Lakkaraju, Jure Leskovec, Annie Liang, and Jens Ludwig. 

Bio

Jon Kleinberg is the Tisch University Professor in the Departments of Computer Science and Information Science at Cornell University. His research focuses on issues at the interface of networks and information, with an emphasis on the social and information networks that underpin the Web and other on-line media. He is a member of the National Academy of Sciences, the National Academy of Engineering, and the American Academy of Arts and Science; and he is the recipient of research fellowships from the MacArthur, Packard, Simons, and Sloan Foundations, as well as awards including the Nevanlinna Prize, the Harvey Prize, the Newell Award, and the ACM-Infosys Foundation Award in the Computing Sciences.

Friday, October 21, 2016 - 15:42



Welcome to ML@GT by the Director

Welcome to ML@GT, a new interdisciplinary research center at Georgia Tech dedicated to the study and application of machine learning—one of the fastest developing research areas impacting computing, engineering, sciences and many other disciplines.

What makes machine learning so important? If you’re visiting this site, you probably already have an answer to that question. But for those new to the field, think of it this way: For all their power and ever-increasing capabilities, computers historically have actually been pretty dumb. They tended to do only what you explicitly told them to do. Of course, ever more sophisticated software (running on ever more sophisticated machines) could carry a lot of instructions, but at the end of day, humans still needed to tell the machine exactly how to respond to each individual circumstance and how to make inferences.

Machine learning changes all that. Now we can provide computers not just with instructions but with the means to learn from their observations and the data they collect. In effect, we enable computers to response to circumstances “on the ground”—or in outer space, or somewhere along the supply chain, or up in the cloud in the midst of billions of terabytes of constantly changing data.

At Georgia Tech, we recognize machine learning to be a game-changer not just in computer science, but in a broad range of scientific, engineering, and business disciplines and practices. And the faculty membership of ML@GT reflects that broad thinking, with dozens of affiliated researchers from the Institute’s Colleges of Computing, Engineering, Sciences, Business, and other areas.

I invite you to peruse our new website to get a feel for what machine learning can do and how Georgia Tech will study and apply it. Whether you already work in machine learning or are a student interested to learn how you can incorporate ML into your studies, we’d like to work with you.

Feel feel to drop us a note. We can all help make the machines all around us just a little bit smarter.

Irfan Essa

Director, ML@GT
Professor & Associate Dean, College of Computing

Friday, October 7, 2016 - 18:58



Welcome new faculty member Devi Parikh to School of Interactive Computing

Please join us in welcoming a new faculty member Devi Parikh in IC. Devi Parikh is an assistant professor at the School of Interactive Computing at Georgia Institute of Technology starting Fall 2016. From 2013 to 2016 she was an Assistant Professor in the Bradley Department of Electrical and Computer Engineering at Virginia Tech. From 2009 to 2012 she was a Research Assistant Professor at Toyota Technological Institute at Chicago (TTIC), a philanthropically endowed computer science academic institute on the University of Chicago campus. She has held visiting positions at Cornell University, University of Texas at Austin, MIT, Carnegie Mellon University, Microsoft Research, and Facebook AI Research.

She received her Ph.D. and M.S. from Carnegie Mellon University in 2009 and 2007 respectively. She received her B.S. in Electrical and Computer Engineering from Rowan University in 2005. 

Her research interests include computer vision and AI in general and visual recognition problems in particular. Her recent work involves exploring problems at the intersection of vision and language, teaching machines common sense, and leveraging human-machine collaboration for building smarter machines. 

For more details, see her website.

Welcome to GA Tech, Devi!

Friday, October 7, 2016 - 17:42



Welcome new faculty member Dhruv Batra to School of Interactive Computing

Please join us in welcoming a new faculty member Dhruv Batra in . Dhruv Batra is an assistant professor at the School of Interactive Computing at Georgia Institute of Technology starting Fall 2016. From 2013 to 2016 he was an Assistant Professor in the Bradley Department of Electrical and Computer Engineering at Virginia Tech. From 2010 to 2012 he was a Research Assistant Professor at Toyota Technological Institute at Chicago (TTIC), a philanthropically endowed computer science academic institute on the University of Chicago campus. He has held visiting positions at Cornell University, MIT, Carnegie Mellon University, Microsoft Research, and Facebook AI Research. 

He received his Ph.D. and M.S. from Carnegie Mellon University in 2010 and 2007 respectively. He received his B.Tech in Electronics Engineering from Indian Institute of Technology (BHU), Varanasi in 2005. 
   
His research lies at the intersection of machine learning, computer vision, and AI, with a focus on developing intelligent systems that are able to concisely summarize their beliefs about the world with diverse predictions, integrate information and beliefs across different sub-components or `modules' of AI (vision, language, reasoning) to extract a holistic view of the world, and explain why they believe what they believe.

For more details, see his website.

Welcome, Dhruv!

Friday, October 7, 2016 - 17:36



IEEE DL Seminar by Paris Smaragdis on "Machine Learning Approaches for Speech Enhancement"

Paris Smaragdis, UIUC & Adobe Research

Date: 25 October 2016
Time: 11:45 AM to 01:00 PM

Technology Square Research Building (TSRB) 125

Abstract

During the last few years, machine learning has started to permeate the world of speech enhancement and has produced results that drastically improve over the state of the art. In this talk I’ll touch on some of the most recent approaches on both multichannel and single channel enhancement, and I will show how traditional signal processing approaches can be reimagined using machine learning tools such as mixture models, matrix factorizations, deep learning regressions, and more.

Biography:

Biography. Paris Smaragdis is an associate professor at the Computer Science and the Electrical and Computer Engineering departments of the University of Illinois at Urbana-Champaign, as well as a senior research scientist at Adobe Research. He completed his masters, PhD, and postdoctoral studies at MIT, performing research on computational audition. In 2006 he was selected by MIT’s Technology Review as one of the year’s top young technology innovators for his work on machine listening, in 2015 he was elevated to an IEEE Fellow for contributions in audio source separation and audio processing, and during 2016-2017 he is an IEEE Signal Processing Society  Distinguished Lecturer. He has authored more than 100 papers on various aspects of audio signal processing, holds more than 40 patents worldwide, and his research has been productized by multiple companies

Friday, October 7, 2016 - 17:24



IRIM Seminar by Byron Boots on "Closing the Gap Between Machine Learning and Robotics"

Wednesday, Oct. 5, 2016
Marcus Nano Bldg. • Rooms 1116-1118
12:00–1:00 p.m.

Abstract

Given a stream of multimodal sensory data, an autonomous robot must continuously refine its understanding of itself and its environment as it makes decisions on how to act to achieve a goal. These are difficult problems that roboticists have attacked using classical tools from mechanics and controls and, more recently, machine learning. However, classical methods and machine learning algorithms are often seen to be at odds, and researchers continue to debate the merits of engineering vs. learning. 

A recurring theme in this talk will be that prior knowledge and domain insights can make learning and inference easier. I will discuss several fundamental robotics problems including continuous-time motion planning, localization, and mapping from a unified probabilistic inference perspective. I will show how models from statistical machine learning like Gaussian Processes can be tightly integrated with insights from engineering expressed as differential equations to solve these problems efficiently. Finally, I will demonstrate the effectiveness of these algorithms on several existent robotics platforms.

Bio

Byron Boots is an assistant professor in the School of Interactive Computing and the Institute for Robotics and Intelligent Machines at the Georgia Institute of Technology. Prior to joining Georgia Tech, Boots was a postdoctoral researcher working with Dieter Fox in the Robotics and State Estimation Lab at the University of Washington. He received his Ph.D. in Machine Learning from Carnegie Mellon in 2012, where he was advised by Geoff Gordon. Boot’s work on learning models of dynamical systems received the 2010 Best Paper award at ICML. His current research focuses on developing theory and systems that integrate perception, learning, and decision-making. 

Wednesday, October 5, 2016 - 10:48



Welcome new faculty member Chethan Pandarinath in BME

Welcome new faculty member Chethan Pandarinath in Wallace H. Coulter Department of Biomedical Engineering

Please join us in welcoming Chethan Pandarinath in Wallace H. Coulter Department of Biomedical Engineering as an Assistant Professor starting in December 2016. His work centers on understanding how the brain represents information and intention, and using this knowledge to develop high-performance, robust, and practical assistive devices for people with disabilities and neurological disorders. He takes a dynamical systems approach to characterizing the activity of large populations of neurons, combined with rigorous systems engineering (signal processing, machine learning, and real-time systems) to advance the performance of brain-machine interfaces and neuromodulatory devices. 

Welcome to Georgia Tech Chethan!

Wednesday, October 5, 2016 - 08:32



Welcome to new faculty Mayya Zhilova in Math

We would like to welcome Mayya  Zhilova, who joined the School of Mathematics at the Georgia Institute of Technology in Fall 2016. Mayya's  primary research interests lie in the areas of mathematical statistics and probability theory, particularly in statistical inference for complex high-dimensional data. Her current research is focused on probabilistic and statistical properties of resampling procedures for high-dimensional data in presence of modelling errors in statistical models

Mayya Zhilova received her Ph.D. from Humboldt University of Berlin in 2015. Before joining Georgia Tech, she worked at Weierstrass Institute in Berlin, in the research group Stochastic Algorithms and Nonparametric Statistics from 2011 till 2016. She did her undergraduate studies in Lomonosov Moscow State University. In her leisure time, Mayya enjoys listening to music, dancing, and hiking. For more details see her website.

Welcome, Mayya!

Tuesday, October 4, 2016 - 15:57



Welcome to new faculty member Huan Xu in ISyE

 

Please join us in welcoming a new faculty member Huan Xu in ISyE. Huan Xu is an assistant professor at the Milton School of Industrial and Systems Engineering at Georgia Institute of Technology since 2016. Before that, he has been an assistant professor at National University of Singapore, first at the department of Mechanical Engineering, then at the department of Industrial and Systems Engineering, from 2011.

Huan graduated from Shanghai Jiaotong University with a Bachelor’s degree in Automation in 1997 and got his Master’s degree in ECE from the National University of Singapore in 2005. He was fortunate enough to obtain a Ph. D. degree in ECE from McGill University under the supervision of Shie Mannor, in 2009. He then spent two wonderful years in the WNCG group of UT-Austin, as a postdoctoral research fellow working with Constantine Caramanis and David Morton. Prior to starting his Ph.D. study, he has worked in Panasonic, HP, and Oracle for almost six years.

His current research interest focus is on learning and decision-making in large-scale complex systems. Specifically, He is interested in machine learning, high-dimensional statistics, robust and adaptable optimization, robust sequential decision making, and applications to large-scale systems. He is an associate editor of IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) and Computational Management Science and was an area chair for NIPS. For more details, see his website.

Welcome, Huan!

Tuesday, October 4, 2016 - 15:44



Upcoming Events