Artificial Intelligence Seminar

Spring 2013
Friday 12:00-1:15
Upson 5130

The AI seminar will meet weekly for lectures by graduate students, faculty, and researchers emphasizing work-in-progress and recent results in AI research. Lunch will be served starting at noon, with the talks running between 12:15 and 1:15. The new format is designed to allow AI chit-chat before the talks begin. Also, we're trying to make some of the presentations less formal so that students and faculty will feel comfortable using the seminar to give presentations about work in progress or practice talks for conferences.

February 8th

Speakers: Ishanu Chattopadhyay

Host: Ashutosh Saxena

Bio: Dr. Ishanu Chattopadhyay is currently pursuing post-doctoral research at the Creative Machines Laboratory (Mechanical & Aerospace Engineering, Cornell University) under the supervision of Professor Hod Lipson. He procured his doctoral degree in autonomous decision-making from The Pennsylvania State University, and his current research interests include machine intelligence, robotics and formal languages, with emphasis on unsupervised learning under noise, uncertainty and stochastic system dynamics.

Title: Information Annihilation for Feature-free Classification

Abstract: Information can annihilate “anti-information” to produce flat white noise. We shall discuss the process by which a group-theoretic inverse of an arbitrary stream of information (the anti-stream) may be computed, and how two streams can algorithmically interact with the objective of annihilating the statistical information contained in them. Besides the theoretical implications, this principle offers a fundamentally new way of comparing information streams, in absence of apriori knowledge about their sources. Since a stream of data perfectly annihilates its anti-stream, the degree to which the anti-stream will annihilate a second stream reflects the causal similarity between the hidden sources. We will briefly discuss the general formalism and demonstrate its applicability in the context of unsupervised classification of a variety of real data, including the disambiguation of electro-encephalograph patterns pertaining to epileptic seizures, detection of anomalous cardiac activity from heart sound recordings, and classification of astronomical objects from raw photometry. Classification via information annihilation requires no features or models, and we would see that the resulting feature-free unsupervised classification technique achieves performance that meets or exceeds that produced by state-of-the-art algorithms based on domain specific features tuned by experts.

“The AI-Seminar is sponsored by Yahoo!”

February 15th

Speaker: Ping Li, Cornell University

Host: Ashutosh Saxena

Bio: Ping Li is an assistant professor in CIS/DSS. His research interests include big data algorithms and statistical learning. He is a reciepient of the 2013 AFOSR YIP Award.

Title: :  Data Stream Compressed Sensing with L0 Projections

Abstract: :   Many applications concern sparse signals, for example, detecting anomalies from the differences
between consecutive images taken by surveillance cameras.  In general, anomaly events are sparse. This talk focuses on the problem of recovering a K-sparse signal in N dimensions (coordinates). Classical theories in compressed sensing say the required number of measurement is M = O(K log N). In our most recent work on L0 projections, we show that an idealized algorithm needs about M = 5K measurements, regardless of N.  In particular, 3 measurements suffice when K = 2 nonzeros.  Practically, our method is very fast, accurate, and very robust against measurement noises. Even when there are no sufficient measurements, the algorithm can still accurately reconstruct a significant portion of the nonzero coordinates, without catastraphic failures (unlike popular methods such as linear programming).  This is joint work with Cun-Hui Zhang at Rutgers University. Paper URL:

“The AI-Seminar is sponsored by Yahoo!”

February 22nd

Speaker: Ellen Voorhees, NIST

Host: Claire Cardie

Title: The Text REtrieval Conference

Abstract: TREC is a series of workshops designed to advance the state-of-the-art in information retrieval by providing the infrastructure necessary for large-scale evaluation of retrieval methodologies.  Now beginning its 22nd cycle, TREC has had a significant impact on the field: retrieval effectiveness approximately doubled in the first eight years of TREC; TREC has established research communities for a variety of specialized retrieval tasks such as question answering and cross-language retrieval; and a 2010 economic impact study calculated that for every $1 invested in TREC at least $3-$5 in benefits accrued to researchers.

Each TREC consists of a collection of "tracks", subtasks that focus research attention on specific challenges.  A diverse set of tracks invigorates TREC and makes TREC attractive to a broad set of participants. This talk will give a brief summary of the history of TREC and focus on two of the newest TREC tracks, the Medical Records track and
the so-called Knowledge Base Acceleration track that focuses on cross-document co-reference resolution at web scale.

“The AI-Seminar is sponsored by Yahoo!”

March 1st

Speaker: Ray Ptucha, Eastman Kodak Company

Host: Andy Gallagher

Bio: Ray is a NSF fellow and senior scientist from Eastman Kodak Company specializing in computer vision, machine learning, robotics, and imaging science.  Ray will graduate with a Ph.D in Computer Science from Rochester Institute of Technology in Spring 2013 specializing in dimensionality reduction and sparse representations for human facial understanding.

Title: Joint Optimization of Manifold Learning and Sparse Representations for Face and Gesture Analysis

Link to Presentation:

Abstract: The parsimonious nature of sparse representations (SR) has successfully been exploited for the development of highly accurate classifiers for various applications.  Despite the successes of SR techniques, large dictionaries and high dimensional data can make these classifiers computationally demanding. Further, sparse classifiers are subject to the adverse effects of a phenomenon known as coefficient contamination. This research analyzes the interaction between dimensionality reduction and sparse representations to present a unified sparse representation classification framework that addresses both issues of computational intensity and coefficient contamination.  This framework, termed LGE-KSVD, utilizes variants of Linear extension of Graph Embedding (LGE) along with  flexible K-SVD dictionary learning to jointly learn the dimensionality reduction matrix, sparse representation dictionary, sparse coefficients, and sparsity-based classifier.  Results are shown for facial recognition, facial expression recognition, human activity analysis, and with the addition of a concept called active difference signatures, delivers robust gesture recognition from Kinect or similar depth cameras.

“The AI-Seminar is sponsored by Yahoo!”

March 8th

Speaker: Noah Smith, Associate Professor, CMU

Host: Lillian Lee

Title:Text and Social Context: Analysis and Prediction

Abstract:The rise of the social web presents new opportunities and challenges for computational and statistical analysis of text data. We can now explore corpora of messages written by huge segments of the population and observe many dimensions of the social context of these messages. In this talk, I'll present some of the analyses we've done to explore how message content varies with geographic location in the US and to shed light on message deletion in China. I'll then turn to the use of text for forecasting social outcomes in scientific and political domains. I'll show how simple models can be used to predict how scientific communities will respond to a newly published article and which congressional bills will survive committee. This is joint work with David Bamman, Chris Dyer, Jacob Eisenstein, Michael Heilman, Brendan O'Connor, Bryan Routledge, John Wilkerson, Eric Xing, Tae Yano, and Dani Yogatama.

“The AI-Seminar is sponsored by Yahoo!”

March 15th

Speaker: Phil Long, NEC-labs

Host: Ping Li

Title: On the necessity of irrelevant variables

Abstract: Abstract: An irrelevant variable typically decreases the accuracy of a classifier; after all, it makes the predictions of the classifier depend to a greater extent on random chance.
We show, however, that the harm from irrelevant variables can be much less than the benefit from relevant variables, so that it is possible to learn very accurate classifiers, almost all of whose variables are irrelevant.  It can be advantageous to continue adding variables, even as they become less and less likely to be relevant.  We showed this with theoretical analysis and experiments using artificially generated data (so that we would know which variables were relevant and irrelevant). Both of these use an assumption, conditional independence given the correct class,
formalizing the intuitive idea that variables are not redundant. In the situation that we studied relatively few of the many variables are relevant, and the relevant variables are only weakly predictive.
In this case, algorithms that cast a wide net outperform algorithms whose hypotheses, on average, contain more relevant than irrelevant variables.

(This is joint work with Dave Helmbold of UC Santa Cruz.)

“The AI-Seminar is sponsored by Yahoo!”

March 22nd

Speaker: NO SEMINAR- Spring Break


“The AI-Seminar is sponsored by Yahoo!”

March 29th

Speaker: NO SEMINAR- ACSU Lunch


“The AI-Seminar is sponsored by Yahoo!”

April 5th

Speaker: Brian Mcfee, postdoctoral research scholar at Columbia University

Host: Thorston Joachims

Bio: Brian McFee is a postdoctoral research scholar in the Center for Jazz Studies at Columbia University, under the supervision of Tad Shull, Dan Ellis, and Douglas Repetto. He received his PhD in Computer Science and Engineering from the University of California, San Diego in 2012. In 2010, he was a recipient of the Qualcomm Innovation Fellowship.

Title: Modeling playlist dialects

Abstract: A core component of streaming online radio services is the playlist generation algorithm. Playlist generation algorithms can be viewed as probabilistic models over ordered song sequences, and algorithms can be optimized and evaluated according to the likelihood of observed, real playlists. While previous studies have treated playlist collections as an undifferentiated whole, we propose to build models which are tuned to specific categories or contexts. Toward this end, we develop a general class of flexible and scalable models based upon hyper-graph random walks. To evaluate the proposed models, we present a large corpus of categorically annotated, user-generated playlists over the Million Song Dataset. Experimental results indicate that category-specific models can provide substantial improvements in accuracy over global playlist models.

(This is joint work with Gert Lanckriet at UCSD.)

“The AI-Seminar is sponsored by Yahoo!”

April 12th

Speaker: Qiaozhu Mei,  U Michigan

Host: Ping Li

Bio: Qiaozhu Mei is an assistant professor at the School of Information, the University of Michigan. He is widely interested in information retrieval, text mining, natural language processing and their applications in web search, social computing, and health informatics. He has served in the program committee of almost all major conferences in these areas. He is also a recipient of the NSF CAREER Award, two runner-up best student paper awards at KDD, and a SIGKDD dissertation award.

Title: The Foreseer: Integrative Retrieval and Mining of Information in Online Communities

Abstract: With the growth of online communities, the Web has evolved from networks of shared documents into networks of knowledge-sharing groups and individuals. A vast amount of heterogeneous yet interrelated information is being generated, making existing information analysis techniques inadequate. Current data mining tools often neglect the actual context, creators, and consumers of information. Foreseer is a user-centric framework for the next generation of information retrieval and mining for online communities. It represents a new paradigm of information analysis through the integration of the four “C’s”: content, context, crowd, and cloud. 
In this talk, we will introduce our recent effort of integrative analysis and mining of information in online communities. We will highlight the real world problems in online communities to which the Foreseer techniques have been successfully applied. These topics include the prediction of the adoption of hashtags in microblogging communities, and the prediction of social lending behaviors in microfinance communities.

“The AI-Seminar is sponsored by Yahoo!”

April 19th

Speaker: Richard Socher, graduate student, Stanford

Host: Lillian Lee

Bio: Richard Socher is a PhD student at Stanford working with Chris Manning and Andrew Ng. His research interests are machine learning for NLP and vision. He is interested in techniques that learn useful and accurate features, capture recursive and hierarchical structure in multiple modalities and perform well across multiple tasks. He was awarded the 2011 Yahoo! Key Scientific Challenges Award, the Distinguished Application Paper Award at ICML 2011 and a Microsoft Research PhD Fellowship in 2012.

Title: Recursive Deep Learning in Natural Language Processing and Computer Vision

Abstract: Hierarchical and recursive structure is commonly found in different modalities, including natural language sentences and scene images. I will introduce several recursive deep learning models that, unlike standard deep learning methods can learn compositional meaning vector representations for phrases or images.

These recursive neural network based models obtain state-of-the-art performance on a variety of syntactic and semantic language tasks such as parsing, sentiment analysis, paraphrase detection and relation classification for extracting knowledge from the web. Because often no language specific assumptions are made the same architectures can be used for visual scene understanding and object classification from 3d images.

Besides the good performance, the models capture interesting phenomena in language such as compositionality. For instance the models learn that “not good” has worse sentiment than “good” or that high level negation can change the meaning of longer phrases with many positive words.
Furthermore, unlike most machine learning approaches that rely on human designed feature sets, features are learned as part of the model.

“The AI-Seminar is sponsored by Yahoo!”

April 26th

Speaker: Brian Kulis, Ohio State University

Host: Thorston Joachims

Bio: Brian Kulis is an assistant professor of computer science at Ohio State University.   His research focuses on machine learning, data mining, and large-scale optimization.  Previously, he was a postdoctoral fellow at UC Berkeley EECS and was also affiliated with the International Computer Science Institute.  He obtained his PhD in computer science from the University of Texas in 2008, and his BA degree from Cornell University in computer science and mathematics in 2003.

Title: Small-Variance Asymptotics for Large-Scale Learning

Abstract: It is widely known that the Gaussian mixture model is related to k-means by “small-variance asymptotics”: as the covariances of the clusters shrink, the EM algorithm approaches the k-means algorithm and the negative joint log-likelihood approaches the k-means objective.  Similar asymptotic connections exist for other machine learning models, including dimensionality reduction (probabilistic PCA becomes PCA), multiview learning (probabilistic CCA becomes CCA), and classification (a restricted Bayes optimal classifier becomes the SVM).  The asymptotic non-probabilistic counterparts to the probabilistic models are almost always more scalable, and are typically easier to analyze, making them useful alternatives to the probabilistic models in many situations.  I will explore how we can extend such asymptotics to a richer class of probabilistic models, with a focus on large-scale graphical models, Bayesian nonparametric models, and time-series data.  I will develop the necessary mathematical tools needed for these extensions and will describe a framework for designing scalable optimization problems derived from the rich probabilistic models.  Applications are diverse, and include topic modeling, network evolution, and feature learning.

“The AI-Seminar is sponsored by Yahoo!”

May 3rd

Speaker: Lana Lazebnik

Host: Noah Snavely

Title: Towards Open-Universe Image Parsing with Broad Coverage (Joint work with J. Tighe)

Abstract: I will present our work on image parsing, or labeling each pixel in an image with its semantic category (e.g., sky, ground, tree, person, etc.). Our aim is to achieve broad coverage across hundreds of object categories in large-scale datasets that can continuously evolve. I will first describe our baseline nonparametric region-based parsing system that can easily scale to datasets with tens of thousands of images and hundreds of labels. Next, I will describe our approach to combining this region-based system with per-exemplar sliding window detectors to improve parsing performance on small object classes, which achieves state-of-the-art results on several challenging datasets.

Bio: Svetlana Lazebnik received her Ph.D. at the University of Illinois at Urbana-Champaign in 2006. From 2007 to 2011, she was an assistant professor of computer science at the University of North Carolina in Chapel Hill, and in 2012 she has returned to the University of Illinois as a faculty member. She is the recipient of an NSF CAREER Award, a Microsoft Research Faculty Fellowship, and a Sloan Foundation Fellowship. She is a member of the DARPA Computer Science Study Group and of the editorial board of the International Journal of Computer Vision. Her research interests focus on scene understanding and modeling the content of large-scale photo collections.

“The AI-Seminar is sponsored by Yahoo!”

May 10th

Speakers: Hema S. Koppula, Ian Lenz & Zhaoyin Jia

Host: Ashutosh Saxena

Anticipating Human Activities using Object Affordances for Reactive Robotic Response,
Hema S Koppula, Ashutosh Saxena.
In Robotics: Science and Systems (RSS), 2013. 

Deep Learning for Detecting Robotic Grasps,
Ian Lenz, Honglak Lee, Ashutosh Saxena.
In Robotics: Science and Systems (RSS), 2013.

3D-Based Reasoning with Blocks, Support, and Stability.
Zhaoyin Jia, Andy Gallagher, Ashutosh Saxena, Tsuhan Chen.
In Computer Vision and Pattern Recognition (CVPR), 2013. 

“The AI-Seminar is sponsored by Yahoo!”






See also the AI graduate study brochure.

Please contact any of the faculty below if you'd like to give a talk this semester. We especially encourage graduate students to sign up!

Sponsored by

CS7790, Spring '12
Claire Cardie
Carla Gomes
Joe Halpern
Dan Huttenlocher
Thorsten Joachims
Lillian Lee
Ashutosh Saxena
Bart Selman
Ramin Zabih

Back to CS course websites