CMS Upcoming SeminarsCMS Upcoming Seminar Feed
http://www.cms.caltech.edu/seminars.rss
enCMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 27 Jan 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87001H.B. Keller Colloquium: Challenges in Reliable Machine Learningdbohler@caltech.edu (Diana Bohler)H.B. Keller Colloquium<strong>Speaker(s):</strong> Kamalika Chaudhuri (University of California, San Diego)<br><strong>Location:</strong> Annenberg 105<br><p></p><p>As machine learning is increasingly used in real applications, there is a need for reliable and robust methods. In this talk, we will discuss two such challenges that arise in reliable machine learning. The first is sample selection bias, where training data is available from a distribution conditioned on a sample selection policy, but the resultant classifier needs to be evaluated on the entire population. We will show how we can use active learning to get a small amount of labeled data from the entire population that can be used to correct this kind of sample selection bias. The second is robustness to adversarial examples -- slight strategic perturbations of legitimate test inputs that cause misclassification. We next look at adversarial examples in the context of a simple non-parametric classifier -- the k-nearest neighbor classifier, and look at its robustness properties. We provide bounds on its robustness as a function of k, and propose a more robust 1-nearest neighbor classifier.</p><p>Joint work with Songbai Yan, Tara Javidi, Yaoyuan Yang, Cyrus Rastchian, Yizhen Wang and Somesh Jha</p>Mon, 27 Jan 2020 16:00:00 -0800http://www.cms.caltech.edu/events/87559IST Lunch Bunch: "Does This Vehicle Belong to You?": Extracting Social Meaning from Language by Computerdiane@cms.caltech.edu (Diane Goodfellow)IST Lunch Bunch<strong>Speaker(s):</strong> Dan Jurafsky (Stanford University)<br><strong>Location:</strong> Annenberg 105<br><p>Police body-worn cameras have the potential to play an important role in understanding and improving police-community relations. In this talk I describe a series of studies conducted by our large interdisciplinary team at Stanford that use speech and natural language processing on body-camera recordings to model the interactions between police officers and community members in traffic stops.</p><p>We draw on linguistic models of dialogue structure and of interpersonal relations like respect to automatically quantify aspects of the interaction from the text and audio. I describe the differences we find in the language directed toward black versus white community members, and offer suggestions for how these findings can be used to help improve the relations between police officers and the communities they serve. I'll also cover a number of our results on using computational methods to uncover historical societal biases, and detect framing, agenda-setting and political polarization in the media.</p><p>Together, these studies highlight how natural language processing can help us interpret latent social content behind the words we use.</p>Tue, 28 Jan 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87415CMX Lunch Seminar: Statistical Guarantees for MAP Estimators in PDE-Constrained Regression Problemsjbrink@caltech.edu (Jolene Brink)CMX Lunch Seminar<strong>Speaker(s):</strong> Sven Wang (University of Cambridge)<br><strong>Location:</strong> Annenberg 213<br><p> </p><p>The main topic of the talk are convergence rates for penalised least squares (PLS) estimators in non-linear statistical inverse problems, which can also be interpreted as Maximum a Posteriori (MAP) estimators for certain Gaussian Priors. Under general conditions on the forward map, we prove convergence rates for PLS estimators.</p><p>In our main example, the parameter f is an unknown heat conductivity function in a steady state heat equation [a second order elliptic PDE]. The observations consist of a noisy version of the solution u[f] to the boundary value corresponding to f. The PDE-constrained regression problem is shown to be solved a minimax-optimal way.</p><p>This is joint work with S. van de Geer and R. Nickl. If time permits, we will mention some related work on the non-parametric Bayesian approach, as well as computational questions for the Bayesian posterior.</p>Wed, 29 Jan 2020 12:00:00 -0800http://www.cms.caltech.edu/events/86831EE Systems Seminar: "Finite-Length Performance of Spatially Coupled Codes and Its Applications"lchavarr@caltech.edu (Liliana Chavarria)EE Systems Seminar<strong>Speaker(s):</strong> Fredrik Brännström (Chalmers University of Technology)<br><strong>Location:</strong> Moore B280<br><p><b>ABSTRACT</b> Frame asynchronous coded slotted ALOHA (FA-CSA) is an uncoordinated multiple access scheme, but in some sense equivalent to spatially coupled low-density parity-check (SC-LDPC) codes. We analyze the performance of FA-CSA in terms of packet loss rate and delay, where we derive tight approximations of the error floor (EF) for the finite frame length regime. We show that, in general, FA-CSA provides better performance in both the EF and waterfall regions as compared to its frame synchronous version. We also analyze the finite-length performance in the waterfall region of SC-LDPC codes under window decoding over the binary erasure channel. In particular, we propose a refinement of the scaling law proposed by Olmos and Urbanke for both the frame and bit error rate of terminated SC-LDPC ensembles under full belief propagation decoding, and also extend the analysis under the more practical sliding window decoding.</p><p><b>BIO</b> Fredrik Brännström is Professor and Head of the Communication Systems Group, Department of Electrical Engineering, Chalmers University of Technology, Gothenburg, Sweden. He received the M.Sc. degree from Luleå University of Technology, Luleå, Sweden, and the Ph.D. degree in Communication Theory from the Department of Computer Engineering, Chalmers University of Technology, Gothenburg, Sweden. From 2006 to 2010, he was a Principal Design Engineer with Quantenna Communications, Inc., Fremont, CA. He was a recipient of the 2013 IEEE Communication Theory Workshop Best Poster Award. In 2014, he received the Department of Signals and Systems Best Teacher Award. Together with his students he has co-authored the papers that received the 2016 and 2017 Best Student Conference Paper and the 2018 Best Student Journal Paper, all awarded by the IEEE Sweden Joint VT-COM-IT Chapter. His current research interests include algorithms, resource allocation, synchronization, antenna concepts, and protocol design for vehicular communication systems, as well as different applications of coding.</p>Wed, 29 Jan 2020 16:00:00 -0800http://www.cms.caltech.edu/events/87632Caltech + Finance Symposium 2020: TBDsabrina@hss.caltech.edu (Sabrina Hameister)Caltech + Finance Symposium 2020<strong>Speaker(s):</strong> <br><strong>Location:</strong> Dabney Hall, Lounge<br><p><b><i>Featuring distinguished Caltech faculty<br></i></b><a href="http://www.hss.caltech.edu/people/federico-m-echenique"><b>Federico Echenique</b></a>, Allen and Lenabelle Davis Professor of Economics<br><a href="http://www.hss.caltech.edu/people/michael-j-ewens"><b>Michael Ewens</b></a>, Professor of Finance and Entrepreneurship<br><a href="http://www.hss.caltech.edu/people/lawrence-j-jin"><b>Lawrence Jin</b></a>, Assistant Professor of Finance<br><b><i>and Keynote Speaker<br></i></b><a href="https://som.yale.edu/faculty/nicholas-c-barberis"><b>Nicholas C. Barberis</b></a>, Stephen and Camille Schramm Professor of Finance, Yale School of Management<br></p><p><b>Schedule<br></b>1:30 PM <b>||</b> Opening Remarks<br>1:40 PM <b>||</b> "Markets for Centralized Allocation Problems: Fairness, Efficiency, and Property Rights," Federico Echenique<br>2:20 PM <b>||</b> "Governance and Compensation in Startups," Michael Ewens<br>3:00 PM <b>||</b> Break<br>3:15 PM <b>||</b> "Prospect Theory and Stock Market Anomalies," Lawrence Jin<br>4:05 PM <b>||</b> Keynote Address, "What's Going on in Behavioral Finance? A Survey of the Latest Ideas," Nicholas Barberis<br>5:30 PM <b>||</b> Reception</p>Fri, 31 Jan 2020 13:30:00 -0800http://www.cms.caltech.edu/events/87557CMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 03 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87002H.B. Keller Colloquium: Algorithms for Eliciting Machine Learning Metricsdbohler@caltech.edu (Diana Bohler)H.B. Keller Colloquium<strong>Speaker(s):</strong> Sanmi Koyejo (University of Illinois at Urbana-Champaign)<br><strong>Location:</strong> Annenberg 105<br><p> </p><p>Given a prediction problem with real-world tradeoffs, which cost function should the machine learning model be trained to optimize? Unfortunately, typical default metrics in machine learning, such as accuracy applied to binary classifiers, may not capture tradeoffs relevant to the problem at hand. This talk proposes metric elicitation as a formal strategy to address the metric selection problem, specifically by automatically discovering implicit preferences from an expert or an expert panel via interactive feedback. I will primarily focus on algorithms for eliciting classification metrics, showing that simple algorithms are efficient for metric elicitation under broad assumptions. Finally, I will briefly outline early work on metric selection for measuring group fairness in classification problems with sensitive groups.</p>Mon, 03 Feb 2020 16:00:00 -0800http://www.cms.caltech.edu/events/87570IST Lunch Bunch: Online Optimization with Preference Feedback, Structured Decision Spaces, and Safety Constraintsdiane@cms.caltech.edu (Diane)IST Lunch Bunch<strong>Speaker(s):</strong> Yanan Sui (Tsinghua University)<br><strong>Location:</strong> Annenberg 105<br><p> Bandit algorithms tackle the fundamental challenge of balancing exploration (collecting data for learning better models) and exploitation (using the estimates to make decisions). In this talk, I will formalize bandit problems with preference feedback, with structured decision spaces, and with safety constraints (when bad samples are not allowed). These constraints commonly exist in many applications. In particular, we are motivated by online decision-making for clinical treatment and robotic control. This talk will exhibit several algorithms for these constrained optimization problems. Theoretical guarantees and empirical efficiencies of our algorithms will be presented. I will also show our clinical practices of online decision-making for neuromodulation. </p>Tue, 04 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87644IQIM Postdoctoral and Graduate Student Seminar: TBDmarciab@caltech.edu (Marcia Brown)IQIM Postdoctoral and Graduate Student Seminar<strong>Speaker(s):</strong> Christopher White (University of Maryland)<br><strong>Location:</strong> East Bridge 114<br><p><b>Abstract:</b> TBD</p>Fri, 07 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87633CMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 10 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87003Special CMX Seminar: TBAjbrink@caltech.edu (Jolene Brink)Special CMX Seminar<strong>Speaker(s):</strong> Carolina Osorio (Massachusetts Institute of Technology)<br><strong>Location:</strong> Annenberg 213<br><p></p>Thu, 13 Feb 2020 16:30:00 -0800http://www.cms.caltech.edu/events/87416EE Systems Seminar: What is the role of curvature in complexity of optimization on manifolds?lchavarr@caltech.edu (Liliana Chavarria)EE Systems Seminar<strong>Speaker(s):</strong> Nicolas Boumal (Princeton University)<br><strong>Location:</strong> Moore B280<br><p><b>ABSTRACT</b> The talk is about solving optimization problems of the form: min f(x), where x lives on a (known) smooth manifold. For example, the manifold could be a sphere, a set of orthonormal matrices, complex phases, fixed-rank matrices or tensors, rigid motions, or a more abstract quotient space owing to symmetry. This comes up in signal and image processing, computer vision, machine learning, inverse problems etc.<br></p><p>After a brief description of how Riemannian geometry enables efficient algorithms, I'll discuss which properties of the cost function and of the manifold affect the worst-case complexity of computing approximate stationary points (both first and second order). In particular, I'll share some thoughts about the role of Riemannian curvature on that front.</p><p><b>BIO</b> Nicolas Boumal is an assistant professor in the mathematics department at Princeton University, where he was also an instructor with the Program in Applied and Computational Mathematics 2016-2018. He studies non-convex optimization, numerical analysis and statistical estimation, exploiting mathematical structures such as smooth geometry, convex geometry and low rank. He is the author of a popular Riemannian optimization toolbox called Manopt.<br></p><p>He obtained his PhD in mathematical engineering from the Université catholique de Louvain in Belgium in 2014, and was a postdoc with the computer science department of the Ecole Normale Supérieure de Paris in 2015. His research is partially supported by a grant of the National Science Foundation.</p>Fri, 21 Feb 2020 16:00:00 -0800http://www.cms.caltech.edu/events/87469CMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 24 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87004CMX Lunch Seminar: An Optimal Transport Perspective on Uncertainty Propagationjbrink@caltech.edu (Jolene Brink)CMX Lunch Seminar<strong>Speaker(s):</strong> Amir Sagiv (Columbia University)<br><strong>Location:</strong> Annenberg 213<br><p> In many scientific areas, a deterministic model (e.g., a differential equation) is equipped with parameters. In practice, these parameters might be uncertain or noisy, and so an honest model should account for these uncertainties and provide a statistical description of the quantity of interest. Underlying this computational problem is a fundamental question - If two "similar" functions push-forward the same measure, are the new resulting measures close, and if so, in what sense? In this talk, I will first show how the probability density function (PDF) can be approximated, and present applications to nonlinear optics. We will then discuss the limitations of PDF approximation, and present an alternative Wasserstein-distance formulation of this problem, which through optimal-transport theory yields a simpler theory. </p>Wed, 26 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/86832Center for Social Information Sciences (CSIS) Seminar: Multiple Imputation for Large Multiscale Data with Linear Constraintsmmartin@caltech.edu (Mary Martin)Center for Social Information Sciences (CSIS) Seminar<strong>Speaker(s):</strong> Jian Cao (Caltech)<br><strong>Location:</strong> Baxter B125<br><p>Abstract: We present a new method that is capable of handling both missing and suppressed value problems for large multiscale data sets, such as the Quarterly Census of Employment and Wages (QCEW) from the U.S. Bureau of Labor Statistics. Existing multiple imputation methods are hard to scale for such data sets. This particularly acute in the case of QCEW, with as many as 1.5 billion observations aggregated along three different scales (industry structure, geographic levels, and time). Our method incorporates three innovations. First, we improve the accuracy of the Bootstrapping-based Expectation Maximization method (King et al. 2010), a state-of-the-art multiple imputation method, by utilizing the extra information from the singular covariance matrix and taking into account of the multiscale data structure. Second, we introduce a quasi-Monte Carlo technique to accelerate convergence. Third, we develop a parallel sequential approach that partitions the large data set into quasi-independent small data sets according to the data structure and patterns of suppressed and missing observations. We demonstrate that our new method improves speed and accuracy. Moreover, it can be applied to large data sets with complicated multiscale structures.</p>Fri, 28 Feb 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87078CMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 02 Mar 2020 12:00:00 -0800http://www.cms.caltech.edu/events/87005Special CMX Seminar: TBAjbrink@caltech.edu (Jolene Brink)Special CMX Seminar<strong>Speaker(s):</strong> Liliana Borcea (University of Michigan)<br><strong>Location:</strong> Annenberg 213<br><p></p>Tue, 03 Mar 2020 16:30:00 -0800http://www.cms.caltech.edu/events/87511Finance Seminar: Topic to be announcedsabrina@hss.caltech.edu (Sabrina Hameister)Finance Seminar<strong>Speaker(s):</strong> Eric Zwick (University of Chicago)<br><strong>Location:</strong> Baxter B125<br><p>Please check later for additional details</p><p><i>Finance Seminars at Caltech are funded through the generous support of The Ronald and Maxine Linde Institute of Economic and Management Sciences (lindeinstitute.caltech.edu).</i></p>Thu, 05 Mar 2020 16:00:00 -0800http://www.cms.caltech.edu/events/86147CMS-EE Partners Tech Talk: TBDclairer@caltech.edu (Claire Ralph)CMS-EE Partners Tech Talk<strong>Speaker(s):</strong> <br><strong>Location:</strong> Annenberg 105<br><p></p>Mon, 09 Mar 2020 12:00:00 -0700http://www.cms.caltech.edu/events/87006