BEGIN:VCALENDAR VERSION:2.0 PRODID:-//IDSS STAGE - ECPv6.15.11//NONSGML v1.0//EN CALSCALE:GREGORIAN METHOD:PUBLISH X-ORIGINAL-URL:https://idss-stage.mit.edu X-WR-CALDESC:Events for IDSS STAGE REFRESH-INTERVAL;VALUE=DURATION:PT1H X-Robots-Tag:noindex X-PUBLISHED-TTL:PT1H BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20160313T070000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20161106T060000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20170312T070000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20171105T060000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20180311T070000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20181104T060000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20190310T070000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20191103T060000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTART;TZID=America/New_York:20181109T110000 DTEND;TZID=America/New_York:20181109T120000 DTSTAMP:20260517T070710 CREATED:20180621T192716Z LAST-MODIFIED:20181107T151946Z UID:7918-1541761200-1541764800@idss-stage.mit.edu SUMMARY:Optimal hypothesis testing for stochastic block models with growing degrees DESCRIPTION:Abstract: In this talk\, we discuss optimal hypothesis testing for distinguishing a stochastic block model from an Erdos–Renyi random graph when the average degree grows to infinity with the graph size. We show that linear spectral statistics based on Chebyshev polynomials of the adjacency matrix can approximate signed cycles of growing lengths when the graph is sufficiently dense. The signed cycles have been shown by Banerjee (2018) to determine the likelihood ratio statistic asymptotically. In this way one achieves sharp asymptotic optimal power of the testing problem within polynomial time complexity. Time permitting\, we will also discuss how linear spectral statistics of a weighted non-backtracking matrix can be used to approximate the likelihood ratio. The talk is based on joint work with Debapratim Banerjee. \n Biography: Dr.Zongming Ma is an Associate Professor of Statistics of the Wharton School at the University of Pennsylvania. He received his PhD in Statistics from Stanford University in 2010 and has since then been on the faculty of the Wharton Statistics Department. Dr.Ma’s research interests include high-dimensional statistical inference\, non-parametric statistics\, network data analysis\, and their applications in biomedical data analysis. He is a recipient of a Sloan Research Fellowship and an NSF CAREER Award. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-18/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20181102T110000 DTEND;TZID=America/New_York:20181102T120000 DTSTAMP:20260517T070710 CREATED:20180621T192450Z LAST-MODIFIED:20181029T150448Z UID:7916-1541156400-1541160000@idss-stage.mit.edu SUMMARY:Joint estimation of parameters in Ising Model DESCRIPTION:Abstract: Inference in the framework of Ising models has received significant attention in Statistics and Machine Learning in recent years. In this talk we study joint estimation of the inverse temperature parameter β\, and the magnetization parameter B\, given one realization from the Ising model\, under the assumption that the underlying graph of the Ising model is completely specified. We show that if the graph is either irregular or sparse\, then both the parameters can be estimated at rate n−1/2 using Besag’s pseudo-likelihood. Conversely\, if the underlying graph is dense and regular\, we show that no consistent estimates exist for (β\, B).\nThis is joint work with Promit Ghosal from Columbia University. \n Biography: Sumit is currently an Assistant Professor in the Statistics Department at Columbia. Prior to this\, he received his PhD in Statistics at Stanford\, under the guidance of Persi Diaconis.\nHis research interests lie in the intersection of Theoretical Statistics and Applied Probability. In Statistics\, his main focus is developing inferential procedures on probability distributions on combinatorial spaces\, such as permutations\, graphs\, and spin configurations. On the probability side\, his main focus is studying persistence of stochastic processes\, and graph coloring problems. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-17/ LOCATION:E18-304\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20181026T110000 DTEND;TZID=America/New_York:20181026T120000 DTSTAMP:20260517T070710 CREATED:20180621T192221Z LAST-MODIFIED:20180626T142057Z UID:7914-1540551600-1540555200@idss-stage.mit.edu SUMMARY:Alan Frieze DESCRIPTION:MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-16/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20181019T110000 DTEND;TZID=America/New_York:20181019T120000 DTSTAMP:20260517T070710 CREATED:20180621T191820Z LAST-MODIFIED:20181010T202034Z UID:7912-1539946800-1539950400@idss-stage.mit.edu SUMMARY:Algorithmic thresholds for tensor principle component analysis DESCRIPTION:Abstract: Consider the problem of recovering a rank 1 tensor of order k that has been subject to Gaussian noise. The log-likelihood for this problem is highly non-convex. It is information theoretically possible to recover the tensor with a finite number of samples via maximum likelihood estimation\, however\, it is expected that one needs a polynomially diverging number of samples to efficiently recover it. What is the cause of this large statistical–to–algorithmic gap? To study this question\, we investigate the thresholds for efficient recovery for a simple family of algorithms\, Langevin dynamics and gradient descent. We view this problem as a member of a broader class of problems which correspond to recovering a signal from a non-linear observation that has been perturbed by isotropic Gaussian noise. We propose a mechanism for success/failure of recovery of such algorithms in terms of the strength of the signal on the high entropy region of the initialization. Joint work with G. Ben Arous (NYU) and R. Gheissari (NYU). \n Biography: Aukosh Jagannath is a Benjamin Pierce Fellow and NSF Postdoctoral fellow at Harvard University with undergraduate and graduate degree from NYU. He works in probability at the interface of statistical physics\, data science\, combinatorics\, and statistics. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-15/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20181012T110000 DTEND;TZID=America/New_York:20181012T120000 DTSTAMP:20260517T070710 CREATED:20180621T191525Z LAST-MODIFIED:20181004T134355Z UID:7910-1539342000-1539345600@idss-stage.mit.edu SUMMARY:Locally private estimation\, learning\, inference\, and optimality DESCRIPTION:Abstract: In this talk\, we investigate statistical learning and estimation under local privacy constraints\, where data providers do not trust the collector of the data and so privatize their data before it is even collected. We identify fundamental tradeoffs between statistical utility and privacy in such local models of privacy\, providing instance-specific bounds for private estimation and learning problems by developing local minimax risks. In contrast to approaches based on worst-case (minimax) error\, which are conservative\, this allows us to evaluate the difficulty of individual problem instances and delineate the possibilities for adaptation in private estimation and inference. As part of this\, we identify an alternative to the Fisher information for private estimation\, giving a more nuanced understanding of the challenges of adaptivity and optimality. We also provide optimal procedures for private inference\, highlighting the importance of a more careful development of optimal tradeoffs between estimation and privacy. One consequence of our results is to identify settings where standard local privacy restrictions may be too strong for practice; time permitting\, I will then discuss a few new directions that maintain limited amounts of privacy while simultaneously allowing the development of high-performance statistical and learning procedures.\nBased on joint work with Feng Ruan.\n\nBiography: John Duchi is an assistant professor of Statistics and Electrical Engineering and (by courtesy) Computer Science at Stanford University\, with graduate degrees from UC Berkeley and undergraduate degrees from Stanford. His work focuses on large scale optimization problems arising out of statistical and machine learning problems\, robustness and uncertain data problems\, and information theoretic aspects of statistical learning. He has won a number of awards and fellowships\, including best paper awards at the Neural Information Processing Systems conference\, the International Conference on Machine Learning\, an NSF CAREER award\, a Sloan Fellowship in Mathematics\, the Okawa Foundation Award\, and the Association for Computing Machinery (ACM) Doctoral Dissertation Award (honorable mention). URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-14/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20181005T110000 DTEND;TZID=America/New_York:20181005T120000 DTSTAMP:20260517T070710 CREATED:20180621T191105Z LAST-MODIFIED:20181004T152601Z UID:7908-1538737200-1538740800@idss-stage.mit.edu SUMMARY:Efficient Algorithms for the Graph Matching Problem in Correlated Random Graphs DESCRIPTION:Abstract: The Graph Matching problem is a robust version of the Graph Isomorphism problem: given two not-necessarily-isomorphic graphs\, the goal is to find a permutation of the vertices which maximizes the number of common edges. We study a popular average-case variant; we deviate from the common heuristic strategy and give the first quasi-polynomial time algorithm\, where previously only sub-exponential time algorithms were known.\nBased on joint work with Boaz Barak\, Chi-Ning Chou\, Zhixian Lei\, and Yueqi Sheng.\n\n\nBiography: Tselil Schramm is a postdoc in theoretical computer science at Harvard and MIT\, hosted by Boaz Barak\, Jon Kelner\, Ankur Moitra\, and Pablo Parrilo. She obtained her PhD in computer science from U.C. Berkeley under the advisement of Prasad Raghavendra and Satish Rao. Her research interests include inference and average-case problems\, optimization via convex programs (especially the sum-of-squares hierarchy)\, spectral algorithms\, spectral graph theory\, and more. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-13/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180928T110000 DTEND;TZID=America/New_York:20180928T120000 DTSTAMP:20260517T070710 CREATED:20180621T194142Z LAST-MODIFIED:20180919T020411Z UID:7928-1538132400-1538136000@idss-stage.mit.edu SUMMARY:Jingbo Liu DESCRIPTION:Abstract Concentration of measure refers to a collection of tools and results from analysis and probability theory that have been used in many areas of pure and applied mathematics. Arguably\, the first data science application of measure concentration (under the name ‘‘blowing-up lemma’’) is the proof of strong converses in multiuser information theory by Ahlswede\, G’acs and K”orner in 1976. Since then\, measure concentration has found applications in many other information theoretic problems\, most notably the converse (impossibility) results in information theory. Motivated by this\, information theorists (e.g. Marton) have also contributed to the mathematical foundations of measure concentration using their information-theoretic techniques. \nNow\, after all the past 40 years of such progress\, we found that\, amusingly\, measure concentration is not the right hammer for many of these information theoretic applications. We introduce a new machinery based on functional inequalities and reverse hypercontractivity which yields strict improvements in terms of sharpness of the bounds\, generality of the source/channel distributions\, and simplicity of the proofs. Examples covered in the talk include: 1. optimal second-order converses to distributed source-type problems (hypothesis testing\, common randomness generation\, and source coding); 2. sharpening the recent relay channel converse bounds by Wu and Ozgur with much simpler proofs. \nThe work benefited from collaborations with Thomas Courtade\, Paul Cuff\, Ayfer Ozgur\, Ramon van Handel\, and Sergio Verd’u \n Biography: jingbo Liu received the B.E. degree from Tsinghua University\, Beijing\, China in 2012\, and the M.A. and Ph.D. degrees from Princeton University\, Princeton\, NJ\, USA\, in 2014 and 2018\, all in electrical engineering. His research interests include signal processing\, information theory\, coding theory\, high dimensional statistics\, and the related fields. His undergraduate thesis received the best undergraduate thesis award at Tsinghua University (2012). He gave a semi-plenary presentation at the 2015 IEEE Int. Symposium on Information Theory\, Hong-Kong\, China. He was a recipient of the Princeton University Wallace Memorial Honorific Fellowship in 2016. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-12/ LOCATION:E18-304\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180921T110000 DTEND;TZID=America/New_York:20180921T120000 DTSTAMP:20260517T070710 CREATED:20180621T184900Z LAST-MODIFIED:20180626T141436Z UID:7904-1537527600-1537531200@idss-stage.mit.edu SUMMARY:Boaz Nadler DESCRIPTION:MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-11/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180914T110000 DTEND;TZID=America/New_York:20180914T120000 DTSTAMP:20260517T070710 CREATED:20180621T184405Z LAST-MODIFIED:20180914T154952Z UID:7902-1536922800-1536926400@idss-stage.mit.edu SUMMARY:An Information-Geometric View of Learning in High Dimensions DESCRIPTION:Abstract: We consider the problem of data feature selection prior to inference task specification\, which is central to high-dimensional learning. Introducing natural notions of universality for such problems\, we show a local equivalence among them. Our analysis is naturally expressed via information geometry\, and represents a conceptually and practically useful learning methodology. The development reveals the key roles of the singular value decomposition\, Hirschfeld-Gebelein-Renyi maximal correlation\, canonical correlation and principle component analyses\, Tishby’s information bottleneck\, Wyner’s common information\, Ky Fan k-norms\, and Brieman and Friedman’s alternating conditional expectation algorithm. As we’ll discuss\, this framework provides a basis for understanding and optimizing aspects of learning systems\, including neural network architectures\, matrix factorization methods for collaborative filtering\, rank-constrained multivariate linear regression\, and semi-supervised learning\, among others.\nJoint work with Shao-Lun Huang\, Anuran Makur\, and Lizhong Zheng\n\n Biography: Gregory W. Wornell received the B.A.Sc. degree (with honors) from the University of British Columbia\, Canada\, and the S.M. and Ph.D. degrees from the Massachusetts Institute of Technology\, all in Electrical Engineering and Computer Science\, in 1985\, 1987 and 1991\, respectively.\nHis research interests and publications span the areas of signal processing\, information theory\, statistical inference\, digital communication\, and information security\, and include architectures for sensing\, learning\, computing\, communication\, and storage; systems for computational imaging\, vision\, and perception; aspects of computational biology and neuroscience; and the design of wireless networks. He has been involved in the Information Theory and Signal Processing societies of the IEEE in a variety of capacities\, and maintains a number of close industrial relationships and activities. He has won a number of awards for both his research and teaching\, including the IEEE Leon K. Kirchmayer Graduate Teaching Award\, and is a Fellow of the IEEE. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-10/ LOCATION:32-155\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180907T110000 DTEND;TZID=America/New_York:20180907T120000 DTSTAMP:20260517T070710 CREATED:20180621T183631Z LAST-MODIFIED:20180626T140733Z UID:7874-1536318000-1536321600@idss-stage.mit.edu SUMMARY:Dejan Slepcev DESCRIPTION:MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-9/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180525T110000 DTEND;TZID=America/New_York:20180525T120000 DTSTAMP:20260517T070710 CREATED:20180510T154321Z LAST-MODIFIED:20180801T190714Z UID:7554-1527246000-1527249600@idss-stage.mit.edu SUMMARY:Fitting a putative manifold to noisy data DESCRIPTION:Abstract: We give a solution to the following question from manifold learning.\nSuppose data belonging to a high dimensional Euclidean space is drawn independently\, identically distributed from a measure supported on a low dimensional twice differentiable embedded compact manifold M\, and is corrupted by a small amount of i.i.d gaussian noise. How can we produce a manifold $M_o$ whose Hausdorff distance to M is small and whose reach (normal injectivity radius) is not much smaller than the reach of M?\nThis is joint work with Charles Fefferman\, Sergei Ivanov\, Yaroslav Kurylev\, and Matti Lassas. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-8/ LOCATION:E18-304\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180511T110000 DTEND;TZID=America/New_York:20180511T120000 DTSTAMP:20260517T070710 CREATED:20171215T163823Z LAST-MODIFIED:20180801T190611Z UID:7150-1526036400-1526040000@idss-stage.mit.edu SUMMARY:Dynamic Incentive-aware Learning: Robust Pricing in Contextual Auctions DESCRIPTION:MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-3/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180504T110000 DTEND;TZID=America/New_York:20180504T120000 DTSTAMP:20260517T070710 CREATED:20171215T163500Z LAST-MODIFIED:20180801T190448Z UID:7148-1525431600-1525435200@idss-stage.mit.edu SUMMARY:Size-Independent Sample Complexity of Neural Networks DESCRIPTION:MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-2/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180427T110000 DTEND;TZID=America/New_York:20180427T120000 DTSTAMP:20260517T070710 CREATED:20171215T163016Z LAST-MODIFIED:20180426T181058Z UID:7146-1524826800-1524830400@idss-stage.mit.edu SUMMARY:Inference\, Computation\, and Visualization for Convex Clustering and Biclustering DESCRIPTION:Abstract: Hierarchical clustering enjoys wide popularity because of its fast computation\, ease of interpretation\, and appealing visualizations via the dendogram and cluster heatmap. Recently\, several have proposed and studied convex clustering and biclustering which\, similar in spirit to hierarchical clustering\, achieve cluster merges via convex fusion penalties. While these techniques enjoy superior statistical performance\, they suffer from slower computation and are not generally conducive to representation as a dendogram. In the first part of the talk\, we present new convex (bi)clustering methods and fast algorithms that inherit all of the advantages of hierarchical clustering. Specifically\, we develop a new fast approximation and variation of the convex (bi)clustering solution path that can be represented as a dendogram or cluster heatmap. Also\, as one tuning parameter indexes the sequence of convex (bi)clustering solutions\, we can use these to develop interactive and dynamic visualization strategies that allow one to watch data form groups as the tuning parameter varies. In the second part of this talk\, we consider how to conduct inference for convex clustering solutions that addresses questions like: Are there clusters in my data set? Or\, should two clusters be merged into one? To achieve this\, we develop a new geometric representation of Hotelling’s T^2-test that allows us to use the selective inference paradigm to test multivariate hypotheses for the first time. We can use this approach to test hypotheses and calculate confidence ellipsoids on the cluster means resulting from convex clustering. We apply these techniques to examples from text mining and cancer genomics. This is joint work with John Nagorski\, Michael Weylandt\, and Frederick Campbell. \nBiography:  Genevera Allen is an Associate Professor of Statistics\, Computer Science\, and Electrical and Computer Engineering at Rice University. She is also a member of the Jan and Dan Duncan Neurological Research Institute at Texas Children’s Hospital and Baylor College of Medicine where she holds a joint appointment. Dr. Allen received her PhD in statistics from Stanford University (2010)\, under the mentorship of Prof. Robert Tibshirani\, and her bachelors\, also in statistics\, from Rice University (2006).\nDr. Allen’s research focuses on developing statistical methods to help scientists make sense of their ‘Big Data’ in applications such as high-throughput genomics and neuroimaging. Her work lies in the areas of modern multivariate analysis\, graphical models\, statistical machine learning\, and data integration or data fusion. She is the recipient of several honors including a National Science Foundation CAREER award\, the International Biometric Society’s Young Statistician Showcase award\, and the George R. Brown School of Engineering’s Research and Teaching Excellence Award at Rice University. In 2013 and 2014\, she represented the American Statistical Association (ASA) at the Coalition for National Science Funding on Capitol Hill and has had her research highlighted on the House floor in a speech by Congressman McNerney (D-CA). In 2014\, Dr. Allen was named to the “Forbes ’30 under 30′: Science and Healthcare” list. Dr. Allen currently serves as an Associated Editor for Biometrics\, the Secretary / Treasurer for the ASA Section on Statistical Computing\, and the Program Chair for the ASA Section on Statistical Learning and Data Science.\nOutside of work\, Dr. Allen is a patron of the Houston Symphony and Houston Grand Opera and is involved with several arts organizations throughout Houston. She also enjoys traveling\, Texas craft beers\, and playing viola. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180413T110000 DTEND;TZID=America/New_York:20180413T120000 DTSTAMP:20260517T070710 CREATED:20171215T161019Z LAST-MODIFIED:20180801T190032Z UID:7143-1523617200-1523620800@idss-stage.mit.edu SUMMARY:Testing degree corrections in Stochastic Block Models DESCRIPTION:Abstract:  The community detection problem has attracted signi cant attention in re- cent years\, and it has been studied extensively under the framework of a Stochas- tic Block Model (SBM). However\, it is well-known that SBMs fit real data very poorly\, and various extensions have been suggested to replicate characteristics of real data. The recovered community assignments are often sensitive to the model used\, and this naturally begs the following question:  Given a network with community structure\, how to decide whether to fi t a vanilla SBM\, or a more complicated model?  In this talk\, we will formulate this problem as a classical goodness of fi t question\, and try to provide some principled answers in this direction. \nThis is based on joint work with Rajarshi Mukherjee. \nBio:  Subhabrata Sen is Schramm Postdoctoral Fellow at Microsoft Re- search NE and MIT Mathematics. He graduated from the Stanford Statistics Department in 2017\, where he was advised by Amir Dembo and Andrea Mon- tanari. He was awarded the “Probability Dissertation Award” for his thesis on “Random graphs\, optimization\, and spin glasses”.  His research interests include hypothesis testing and non-parametric inference on one hand\, and combinatorial optimization and random graphs on the other. URL:https://idss-stage.mit.edu/calendar/stochastic-and-statistics-seminar/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180406T110000 DTEND;TZID=America/New_York:20180406T120000 DTSTAMP:20260517T070710 CREATED:20180311T182217Z LAST-MODIFIED:20180311T183822Z UID:7473-1523012400-1523016000@idss-stage.mit.edu SUMMARY:Optimality of Spectral Methods for Ranking\, Community Detections and Beyond DESCRIPTION:Abstract: Spectral methods have been widely used for a large class of challenging problems\, ranging from top-K ranking via pairwise comparisons\, community detection\, factor analysis\, among others.\nAnalyses of these spectral methods require super-norm perturbation analysis of top eigenvectors. This allows us to UNIFORMLY approximate elements in eigenvectors by linear functions of the observed random matrix that can be analyzed further. We first establish such an infinity-norm pertubation bound for top eigenvectors and apply the idea to several challenging problems such as top-K ranking\, community detections\, Z_2-syncronization and matrix completion. We show that the spectral methods are indeed optimal for these problems. We illustrate these methods via simulations.\n(Based on joint work with Emmanuel Abbe\, Kaizheng Wang\, Yiqiao Zhong and that of Yixin Chen\, Cong Ma and Kaizheng Wang) \n Biography: Jianqing Fan is Frederick L. Moore Professor at Princeton University. After receiving his Ph.D. from the University of California at Berkeley\, he has been appointed as assistant\, associate\, and full professor at the University of North Carolina at Chapel Hill (1989-2003)\, professor at the University of California at Los Angeles (1997-2000)\, and professor at the Princeton University (2003–). He was the past president of the Institute of Mathematical Statistics and International Chinese Statistical Association. He is co-editing Journal of Econometrics and was the co-editor of The Annals of Statistics\, Probability Theory and Related Fields and Econometrics Journal. His published work on statistics\, economics\, finance\, and computational biology has been recognized by The 2000 COPSS Presidents’ Award\, The 2007 Morningside Gold Medal of Applied Mathematics\, Guggenheim Fellow\, P.L. Hsu Prize\, Royal Statistical Society Guy medal in silver\, and election to Academician of Academia Sinica and follow of American Associations for Advancement of Science. URL:https://idss-stage.mit.edu/calendar/optimality-of-spectral-methods-for-ranking-community-detections-and-beyond/ LOCATION:E18-304\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180323T110000 DTEND;TZID=America/New_York:20180323T120000 DTSTAMP:20260517T070710 CREATED:20180205T145624Z LAST-MODIFIED:20180205T145624Z UID:7346-1521802800-1521806400@idss-stage.mit.edu SUMMARY:Statistical theory for deep neural networks with ReLU activation function DESCRIPTION:Abstract: The universal approximation theorem states that neural networks are capable of approximating any continuous function up to a small error that depends on the size of the network. The expressive power of a network does\, however\, not guarantee that deep networks perform well on data. For that\, control of the statistical estimation risk is needed. In the talk\, we derive statistical theory for fitting deep neural networks to data generated from the multivariate nonparametric regression model. It is shown that estimators based on sparsely connected deep neural networks with ReLU activation function and properly chosen network architecture achieve the minimax rates of convergence (up to logarithmic factors) under a general composition assumption on the regression function. The framework includes many well-studied structural constraints such as (generalized) additive models. While there is a lot of flexibility in the network architecture\, the tuning parameter is the sparsity of the network. Specifically\, we consider large networks with number of potential parameters being much bigger than the sample size. Interestingly\, the depth (number of layers) of the neural network architectures plays an important role and our theory suggests that scaling the network depth with the logarithm of the sample size is natural. URL:https://idss-stage.mit.edu/calendar/statistical-theory-for-deep-neural-networks-with-relu-activation-function/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180316T110000 DTEND;TZID=America/New_York:20180316T120000 DTSTAMP:20260517T070710 CREATED:20180302T201932Z LAST-MODIFIED:20180302T201932Z UID:7461-1521198000-1521201600@idss-stage.mit.edu SUMMARY:When Inference is Tractable DESCRIPTION:Abstract:\nA key capability of artificial intelligence will be the ability to\nreason about abstract concepts and draw inferences. Where data is\nlimited\, probabilistic inference in graphical models provides a\npowerful framework for performing such reasoning\, and can even be used\nas modules within deep architectures. But\, when is probabilistic\ninference computationally tractable? I will present recent theoretical\nresults that substantially broaden the class of provably tractable\nmodels by exploiting model stability (Lang\, Sontag\, Vijayaraghavan\, AI\nStats ’18)\, structure in model parameters (Weller\, Rowland\, Sontag\, AI\nStats ’16)\, and reinterpreting inference as ground truth recovery\n(Globerson\, Roughgarden\, Sontag\, Yildirim\, ICML ’15). \nBio:\nDavid Sontag is an Assistant Professor in the Department of Electrical\nEngineering and Computer Science (EECS) at MIT\, and member of the\nInstitute for Medical Engineering and Science and the Computer Science\nand Artificial Intelligence Laboratory (CSAIL). Prior to joining MIT\,\nDr. Sontag was an Assistant Professor in Computer Science and Data\nScience at New York University from 2011 to 2016\, and a postdoctoral\nresearcher at Microsoft Research New England. Dr. Sontag received the\nSprowls award for outstanding doctoral thesis in Computer Science at\nMIT in 2010\, best paper awards at the conferences Empirical Methods in\nNatural Language Processing (EMNLP)\, Uncertainty in Artificial\nIntelligence (UAI)\, and Neural Information Processing Systems (NIPS)\,\nfaculty awards from Google\, Facebook\, and Adobe\, and a National\nScience Foundation Early Career Award. Dr. Sontag received a B.A. from\nthe University of California\, Berkeley. URL:https://idss-stage.mit.edu/calendar/when-inference-is-tractable/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180309T110000 DTEND;TZID=America/New_York:20180309T120000 DTSTAMP:20260517T070710 CREATED:20171215T165643Z LAST-MODIFIED:20180305T132412Z UID:7157-1520593200-1520596800@idss-stage.mit.edu SUMMARY:Statistical estimation under group actions: The Sample Complexity of Multi-Reference Alignment DESCRIPTION:Abstract: \nMany problems in signal/image processing\, and computer vision amount to estimating a signal\, image\, or tri-dimensional structure/scene from corrupted measurements. A particularly challenging form of measurement corruption are latent transformations of the underlying signal to be recovered. Many such transformations can be described as a group acting on the object to be recovered. Examples include the Simulatenous Localization and Mapping (SLaM) problem in Robotics and Computer Vision\, where pictures of a scene are obtained from different positions andorientations; Cryo-Electron Microscopy (Cryo-EM) imaging where projections of a molecule density are taken from unknown rotations\, andseveral others. \nOne fundamental example of this type of problems is Multi-Reference Alignment: Given a group acting in a space\, the goal is to estimate an orbit of the group action from noisy samples. For example\, in one of its simplest forms\, one is tasked with estimating a signal from noisy cyclically shifted copies. We will show that the number of observations needed by any method has a surprising dependency on the signal-to-noise ratio (SNR)\, and algebraic properties of the underlying group action. Remarkably\, in some important cases\, this sample complexity is achieved with computationally efficient methods based on computing invariants under the group of transformations. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-6/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180302T110000 DTEND;TZID=America/New_York:20180302T120000 DTSTAMP:20260517T070710 CREATED:20171215T165516Z LAST-MODIFIED:20180214T152856Z UID:7155-1519988400-1519992000@idss-stage.mit.edu SUMMARY:One and two sided composite-composite tests in Gaussian mixture models DESCRIPTION:Abstract: Finding an efficient test for a testing problem is often linked to the problem of estimating a given function of the data. When this function is not smooth\, it is necessary to approximate it cleverly in order to build good tests.\nIn this talk\, we will discuss two specific testing problems in Gaussian mixtures models. In both\, the aim is to test the proportion of null means. The aforementioned link between sharp approximation rates of non-smooth objects and minimax testing rates is particularly well illustrated by these problems. \n(based on joint works with Nicolas Verzelen\, Etienne Roquain and Sylvain Delattre) \nBiography:  Alexandra Carpenter is since October 2017 chair of Mathematical Statistics and Machine Learning in the Institut für Mathematische Stochastik (IMST)\, Fakultät für Mathematik (FMA)\, in the Otto-von-Guericke-Universität Magdeburg. Prior to that\, she was between 2015 and 2017 the group leader of the DFG Emmy Noether group MuSyAD on theoretical anomaly detection in the Universitaet Potsdam\, and between 2012 and 2015 in the StatsLab in the University of Cambridge as a research associate\, working with Richard Nickl. She finished her PhD in 2012 in INRIA Lille Nord-Europe under the supervision of Remi Munos and on the topic of bandit theory. Her research interests are in machine learning and mathematical statistics with an emphasis on composite testing problems\, adaptive inference in high and infinite dimension and sequential learning (e.g. bandit theory). URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-5/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180223T110000 DTEND;TZID=America/New_York:20180223T120000 DTSTAMP:20260517T070710 CREATED:20171215T164243Z LAST-MODIFIED:20180123T190050Z UID:7152-1519383600-1519387200@idss-stage.mit.edu SUMMARY:Optimization's Implicit Gift to Learning: Understanding Optimization Bias as a Key to Generalization DESCRIPTION:Abstract: \nIt is becoming increasingly clear that implicit regularization\nafforded by the optimization algorithms play a central role in machine\nlearning\, and especially so when using large\, deep\, neural\nnetworks. We have a good understanding of the implicit regularization\nafforded by stochastic approximation algorithms\, such as SGD\, and as I\nwill review\, we understand and can characterize the implicit bias of\ndifferent algorithms\, and can design algorithms with specific\nbiases. But in this talk I will focus on implicit biases of\ndeterministic algorithms on underdetermined problem. In an effort to\nuncover the implicit biases of gradient-based optimization of neural\nnetworks\, which holds the key to their empirical success\, I will\ndiscuss recent work on implicit regularization for matrix\nfactorization and for linearly separable problems with monotone\ndecreasing loss functions. \nBio: \nProfessor Nati Srebro obtained his PhD at the Massachusetts Institute\nof Technology (MIT) in 2004\, held a post-doctoral fellowship with the\nMachine Learning Group at the University of Toronto\, and was a\nVisiting Scientist at IBM Haifa Research Labs. Since January 2006\, he\nhas been on the faculty of the Toyota Technological Institute at\nChicago (TTIC) and the University of Chicago\, and has also served as\nthe first Director of Graduate Studies at TTIC. From 2013 to 2014 he\nwas associate professor at the Technion-Israel Institute of\nTechnology. Prof. Srebro’s research encompasses methodological\,\nstatistical and computational aspects of Machine Learning\, as well as\nrelated problems in Optimization. Some of Prof. Srebro’s significant\ncontributions include work on learning “wider” Markov networks\,\nincluding introducing the use of the nuclear norm for machine learning\nand matrix reconstruction and work on fast optimization techniques for\nmachine learning\, and on the relationship between learning and\noptimization. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-4/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180216T110000 DTEND;TZID=America/New_York:20180216T120000 DTSTAMP:20260517T070710 CREATED:20171207T154519Z LAST-MODIFIED:20180118T181839Z UID:7109-1518778800-1518782400@idss-stage.mit.edu SUMMARY:User-friendly guarantees for the Langevin Monte Carlo DESCRIPTION:Abstract: \nIn this talk\, I will revisit the recently established theoretical guarantees for the convergence of the Langevin Monte Carlo algorithm of sampling from a smooth and (strongly) log-concave density. I will discuss the existing results when the accuracy of sampling is measured in the Wasserstein distance and provide further insights on relations between\, on the one hand\, the Langevin Monte Carlo for sampling and\, on the other hand\, the gradient descent for optimization. I will also present non-asymptotic guarantees for the accuracy of a version of the Langevin Monte Carlo algorithm that is based on inaccurate evaluations of the gradient. Finally\, I will propose a variable-step version of the Langevin Monte Carlo algorithm that has two advantages. First\, its step-sizes are independent of the target accuracy and\, second\, its rate provides a logarithmic improvement over the constant-step Langevin Monte Carlo algorithm.\nThis is a joint work with A. Karagulyan URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-arnak-dalalyan-enseacrest/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180209T110000 DTEND;TZID=America/New_York:20180209T120000 DTSTAMP:20260517T070710 CREATED:20171207T154146Z LAST-MODIFIED:20180119T204343Z UID:7106-1518174000-1518177600@idss-stage.mit.edu SUMMARY:Variable selection using presence-only data with applications to biochemistry DESCRIPTION:Abstract: \nIn a number of problems\, we are presented with positive and unlabelled data\, referred to as presence-only responses. The application I present today involves studying the relationship between protein sequence and function and presence-only data arises since for many experiments it is impossible to obtain a large set of negative (non-functional) sequences. Furthermore\, if the number of variables is large and the goal is variable selection (as in this case)\, a number of statistical and computational challenges arise due to the non-convexity of the objective. In this talk\, I present an algorithm (PUlasso) with provable guarantees for doing variable selection and classification with presence-only data. Our algorithm involves using the majorization-minimization (MM) framework which is a generalization of the well-known expectation-maximization (EM) algorithm. In particular to make our algorithm scalable\, our algorithm has two computational speed-ups to the standard EM algorithm. I provide a theoretical guarantee where we first show that our algorithm is guaranteed to converge to a stationary point\, and then prove that any stationary point achieves the minimax optimal mean-squared error of slogp/n\, where s is the sparsity of the true parameter. I also demonstrate through simulations that our algorithm out-performs state-of-the-art algorithms in the moderate p settings in terms of classification performance. Finally\, I demonstrate that our PUlasso algorithm performs well on a biochemistry example. URL:https://idss-stage.mit.edu/calendar/stochastic-and-statistics-seminar-garvesh-raskutti-univ-of-wisconsin/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20180202T110000 DTEND;TZID=America/New_York:20180202T120000 DTSTAMP:20260517T070710 CREATED:20171228T200551Z LAST-MODIFIED:20180123T191117Z UID:7195-1517569200-1517572800@idss-stage.mit.edu SUMMARY:Connections between structured estimation and weak submodularity DESCRIPTION:Abstract:  Many modern statistical estimation problems rely on imposing additional structure in order to reduce the statistical complexity and provide interpretability. Unfortunately\, these structures often are combinatorial in nature and result in computationally challenging problems. In parallel\, the combinatorial optimization community has placed significant effort in developing algorithms that can approximately solve such optimization problems in a computationally efficient manner. The focus of this talk is to expand upon ideas that arise in combinatorial optimization and connect those algorithms and ideas to statistical questions. We will discuss three main vignettes: Cardinality constrained optimization; low-rank matrix estimation problems; and greedy estimation of sparse fourier components. \nBio:  Professor Negahban is currently an Assistant Professor in the Department of Statistics at Yale University.  Prior to that he worked with Professor Devavrat Shah at MIT as a postdoc and Prof. Martin J. Wainwright at UC Berkeley as a graduate student. URL:https://idss-stage.mit.edu/calendar/stochastics-and-statistics-seminar-7/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171201T110000 DTEND;TZID=America/New_York:20171201T120000 DTSTAMP:20260517T070710 CREATED:20171120T201126Z LAST-MODIFIED:20180801T185333Z UID:7017-1512126000-1512129600@idss-stage.mit.edu SUMMARY:Challenges in Developing Learning Algorithms to Personalize Treatment in Real Time DESCRIPTION:Abstract:  \nA formidable challenge in designing sequential treatments is to  determine when and in which context it is best to deliver treatments.  Consider treatment for individuals struggling with chronic health conditions.  Operationally designing the sequential treatments involves the construction of decision rules that input current context of an individual and output a recommended treatment.   That is\, the treatment is adapted to the individual’s context; the context may include  current health status\, current level of social support and current level of adherence for example.  Data sets on individuals with records of time-varying context and treatment delivery can be used to inform the construction of the decision rules.    There is much interest in personalizing the decision rules\, particularly in real time as the individual experiences sequences of treatment.   Here we discuss our work in designing  online “bandit” learning algorithms for use in personalizing mobile health interventions. \nBiography: \nSusan A. Murphy is Professor of Statistics\, Radcliffe Alumnae Professor at the Radcliffe Institute and Professor of Computer Science at the Harvard John A. Paulson School of Engineering and Applied Sciences\, all at Harvard University. Her lab focuses on  improving sequential\, individualized\, decision making in health\, in particular on clinical trial design and data analysis to inform the development of just-in-time adaptive interventions in mobile health.  The lab’s work is funded by National Institute on Drug Abuse \, by National Institute on Alcohol Abuse and Alcoholism\, by National Heart\, Lung and Blood Institute and by National Institute of Biomedical Imaging and Bioengineering.   Susan is a Fellow of the Institute of Mathematical Statistics\, a Fellow of the College on Problems in Drug Dependence\, a former editor of the Annals of Statistics\, a member of the US National Academy of Sciences\, a member of the US National Academy of Medicine and a 2013 MacArthur Fellow. URL:https://idss-stage.mit.edu/calendar/challenges-in-developing-learning-algorithms-to-personalize-treatment-in-real-time/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171117T110000 DTEND;TZID=America/New_York:20171117T120000 DTSTAMP:20260517T070710 CREATED:20171120T205246Z LAST-MODIFIED:20180801T184930Z UID:7021-1510916400-1510920000@idss-stage.mit.edu SUMMARY:Generative Models and Compressed Sensing DESCRIPTION:Abstract:  \nThe goal of compressed sensing is to estimate a vector from an under-determined system of noisy linear measurements\, by making use of prior knowledge in the relevant domain. For most results in the literature\, the structure is represented by sparsity in a well-chosen basis. We show how to achieve guarantees similar to standard compressed sensing but without employing sparsity at all. Instead\, we assume that the unknown vectors lie near the range of a generative model\, e.g. a GAN or a VAE. We show how the problems of image inpainting and super-resolution are special cases of our general framework.  \nWe show how to generalize the RIP condition for generative models and that random gaussian measurement matrices have this property with high probability. A Lipschitz condition for the generative neural network is the key technical issue for our results.  \nTime permitting we will discuss follow-up work on how GANs can model causal structure in high-dimensional probability distributions.  (Based on joint works with Ashish Bora\, Ajil Jalal\, Murat Kocaoglu\, Christopher Snyder and Eric Price) \nCode: https://github.com/AshishBora/csgm \nHomepage: users.ece.utexas.edu/~dimakis \nBiography:   \nAlex Dimakis is an Associate Professor at the ECE department\, University of Texas at Austin. He received his Ph.D. in 2008 from UC Berkeley working with Martin Wainwright and Kannan Ramchandran. He received an NSF Career award\, a Google faculty research award and the Eli Jury dissertation award. He is the co-recipient of several best paper awards including the joint Information Theory and Communications Society Best Paper Award in 2012. He is currently serving as an associate editor for IEEE Transactions on Information Theory. His research interests include information theory\, coding theory and machine learning. URL:https://idss-stage.mit.edu/calendar/generative-models-and-compressed-sensing/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171103T110000 DTEND;TZID=America/New_York:20171103T120000 DTSTAMP:20260517T070711 CREATED:20171120T200525Z LAST-MODIFIED:20171120T200525Z UID:7014-1509706800-1509710400@idss-stage.mit.edu SUMMARY:Statistics\, Computation and Learning with Graph Neural Networks DESCRIPTION:Abstract: \nDeep Learning\, thanks mostly to Convolutional architectures\, has recently transformed computer vision and speech recognition. Their ability to encode geometric stability priors\, while offering enough expressive power\, is at the core of their success. In such settings\, geometric stability is expressed in terms of local deformations\, and it is enforced thanks to localized convolutional operators that separate the estimation into scales. \nMany problems across applied sciences\, from particle physics to recommender systems\, are formulated in terms of signals defined over non-Euclidean geometries\, and also come with strong geometric stability priors. In this talk\, I will present techniques that exploit geometric stability in general geometries with appropriate graph neural network architectures. We will show that these techniques can all be framed in terms of local graph generators such as the graph Laplacian. We will present some stability certificates\, as well as applications to computer graphics\, particle physics and graph estimation problems. In particular\, we will describe how graph neural networks can be used to reach statistical detection thresholds in community detection on random graph families\, and attack hard combinatorial optimization problems\, such as the Quadratic Assignment Problem. \nBiography: \nJoan Bruna graduated from Universitat Politecnica de Catalunya (Barcelona\, Spain) in both Mathematics and Electrical Engineering. He obtained an M.Sc. in applied mathematics from ENS Cachan (France). He then became a research engineer in an image processing startup\, developing real-time video processing algorithms. He obtained his PhD in Applied Mathematics at Ecole Polytechnique (France). He was a postdoctoral researcher at the Courant Institute\, NYU\, New York\, and a fellow at Facebook AI Research. In 2015\, he became Assistant Professor at UC Berkeley\, Statistics Department\, and starting Fall 2016 he joined the Courant Institute (NYU\, New York) as Assistant Professor in Computer Science\, Data Science and Mathematics (affiliated). His research interests include invariant signal representations\, high-dimensional statistics and stochastic processes\, deep learning and its applications to signal processing. URL:https://idss-stage.mit.edu/calendar/statistics-computation-and-learning-with-graph-neural-networks/ CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171101T110000 DTEND;TZID=America/New_York:20171101T120000 DTSTAMP:20260517T070711 CREATED:20171120T181051Z LAST-MODIFIED:20171120T192008Z UID:7007-1509534000-1509537600@idss-stage.mit.edu SUMMARY:Unbiased Markov chain Monte Carlo with couplings DESCRIPTION:Abstract: Markov chain Monte Carlo methods provide consistent approximations of integrals as the number of iterations goes to infinity. However\, these estimators are generally biased after any fixed number of iterations\, which complicates both parallel computation. In this talk I will explain how to remove this burn-in  bias by using couplings of Markov chains and a telescopic sum argument\, inspired by Glynn & Rhee (2014). The resulting unbiased estimators can be computed independently in parallel\, and averaged. I will present coupling constructions for Metropolis-Hastings\, Gibbs and Hamiltonian Monte Carlo. The proposed methodology will be illustrated on various examples. If time permits\, I will describe how the proposed estimators can approximate the “cut” distribution that arises in Bayesian inference for misspecified models made of sub-models. \nThis is joint work with John O’Leary\, Yves F. Atchade and Jeremy Heng\,\navailable at arxiv.org/abs/1708.03625 and arxiv.org/abs/1709.00404. \nBiography: Pierre Jacob is an Assistant Professor of Statistics at Harvard University since 2015. Pierre was before a postdoctoral research fellow at the University of Oxford and the National University of Singapore. His Ph.D. was from Université Paris-Dauphine on computational methods for Bayesian inference. His current research is on algorithms amenable to parallel computing for Bayesian inference and model comparison\, with a focus on time series models. \nPierre E. Jacob\nAssistant Professor of Statistics\, Harvard University\npersonal website: sites.google.com/site/pierrejacob/\nblog: statisfaction.wordpress.com/ URL:https://idss-stage.mit.edu/calendar/unbiased-markov-chain-monte-carlo-with-couplings/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171027T110000 DTEND;TZID=America/New_York:20171027T120000 DTSTAMP:20260517T070711 CREATED:20171002T194208Z LAST-MODIFIED:20171120T180403Z UID:6559-1509102000-1509105600@idss-stage.mit.edu SUMMARY:Stochastics and Statistics Seminar - Amit Daniely (Google) DESCRIPTION:Abstract:  \nCan learning theory\, as we know it today\, form a theoretical basis for neural networks. I will try to discuss this question in light of two new results — one positive and one negative. \nBased on joint work with Roy Frostig\, Vineet Gupta and Yoram Singer\, and with Vitaly Feldman \nBiography: \nAmit Daniely is an Assistant Professor at the Hebrew University in Jerusalem\, and a research scientist at Google Research\, Tel-Aviv. Prior to that\, he was a research scientist at Google Research\, Mountain-View. Even prior to that\, he was a Ph.D. student at the Hebrew University of Jerusalem\, Israel\, supervised by Nati Linial and Shai Shalev-Shwartz. His main research interest is Machine Learning Theory. URL:https://idss-stage.mit.edu/calendar/stochastic-and-statistics-seminar-amit-daniely-google/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/New_York:20171020T110000 DTEND;TZID=America/New_York:20171020T120000 DTSTAMP:20260517T070711 CREATED:20171002T193921Z LAST-MODIFIED:20171006T202431Z UID:6555-1508497200-1508500800@idss-stage.mit.edu SUMMARY:Inference in dynamical systems and the geometry of learning group actions - Sayan Mukherjee (Duke) DESCRIPTION: URL:https://idss-stage.mit.edu/calendar/inference-in-dynamical-systems-and-the-geometry-of-learning-group-actions-sayan-mukherjee-duke/ LOCATION:MIT Building E18\, Room 304\, Ford Building (E18)\, 50 Ames Street\, Cambridge\, MA\, United States CATEGORIES:Stochastics and Statistics Seminar Series END:VEVENT END:VCALENDAR