Title: GASP Codes for Secure Distributed Matrix Multiplication
Abstract: We consider the problem of secure distributed matrix multiplication (SDMM) in which a user wishes to compute the product of two matrices with the assistance of honest but curious servers. We construct polynomial codes for SDMM by studying a combinatorial problem on a special type of addition table, which we call the degree table. The codes are based on arithmetic progressions, and are thus named GASP (Gap Additive Secure Polynomial) Codes. GASP Codes are shown to outperform all previously known polynomial codes for secure distributed matrix multiplication in terms of download rate.
Biography: Rafael G.L. D’Oliveira is a postdoctoral research assistant in the Department of Electrical Engineering at Rutgers University. He received his Ph.D. degree in applied mathematics from the University of Campinas in Brazil. As a Ph.D. student he did a one year research internship at Telecom ParisTech in France. His research interests lie in the areas of information theory, coding theory and their applications to privacy in distributed systems.
Title: Outage Analysis for the Open Area Mm-wave Device-to-Device Environment
Abstract: A significant portion of the 5th generation of wireless networks will operate in the mm-wave bands. One of the several challenges associated with mm-wave propagation is to overcome shadowing due to signal blockage caused by environmental objects. Particularly susceptible are nodes in a device-to-device network that typically operate at low power and in a blockage prone environment such as crowded open areas. In this talk, we provide an insight into the effect of blockages on the signal quality for an open area device-to-device scenario. We propose a blockage model based on a fundamental stochastic geometry process: the homogeneous Poisson Point Process. The model provides the average signal attenuation as a soft metric that quantifies the extent of blockage. This not only indicates whether the signal is blocked but also measures how much the signal is attenuated due to one or more blockers. The analytical results are confirmed with the help of Monte Carlo simulations for real-world blocker placement in the environment. Further, outage performance of under line-of-sight and non-line-of-sight mm-wave environments is analyzed.
Biography: Swapnil Mhaske is a Ph.D. student under the supervision of Prof. Predrag Spasojevic in the Department of Electrical and Computer Engineering, Rutgers University, New Brunswick, NJ, USA. He received the B.E. degree in electronics engineering from the University of Pune, India and the M.S. degree in electrical and computer engineering from Rutgers University, New Brunswick, NJ, USA. His research interests broadly lie in the area of wireless communications, with focus on the development of efficient hardware architectures for error control coding systems and channel modeling for mm-wave communications.
Title: Random initialization and implicit regularization in nonconvex statistical estimation
Abstract: Recent years have seen a flurry of activities in designing provably efficient nonconvex procedures for solving statistical estimation / learning problems. Due to the highly nonconvex nature of the empirical loss, state-of-the-art procedures often require suitable initialization and proper regularization (e.g. trimming, regularized cost, projection) in order to guarantee fast convergence. For vanilla procedures such as gradient descent, however, prior theory is often either far from optimal or completely lacks theoretical guarantees. This talk is concerned with a striking phenomenon arising in two nonconvex problems (i.e. phase retrieval and matrix completion): even in the absence of careful initialization, proper saddle escaping, and/or explicit regularization, gradient descent converges to the optimal solution within a logarithmic number of iterations, thus achieving near-optimal statistical and computational guarantees at once. All of this is achieved by exploiting the statistical models in analyzing optimization algorithms, via a leave-one-out approach that enables the decoupling of certain statistical dependency between the gradient descent iterates and the data. As a byproduct, for noisy matrix completion, we demonstrate that gradient descent achieves near-optimal entrywise error control. (This is joint work with Cong Ma, Kaizheng Wang, Yuejie Chi, and Jianqing Fan.)
Biography: Yuxin Chen is currently an assistant professor in the Department of Electrical Engineering at Princeton University. Prior to joining Princeton, he was a postdoctoral scholar in the Department of Statistics at Stanford University, and he completed his Ph.D. in Electrical Engineering at Stanford University. His research interests include high-dimensional statistics, convex and nonconvex optimization, statistical learning, and information theory. He received the 2019 AFOSR Young Investigator Award.
Title: Fundamental Limits in Information Networks: Communication, Inference and Learning
Abstract: Information networks surround us today in different forms and levels, ranging from neural networks, to social networks, to wireless networks and the Internet. The nodes in these networks accomplish tasks such as communication, inference and learning by exchanging information with each other. What are the fundamental laws that govern information flow in networks and how can a desired task be achieved most efficiently? This question was successfully answered by Shannon in 1948 for the case of a single point-point channel and when the desired task is the reliable communication of data, giving birth to the field of information theory. In this talk, I will demonstrate how information theory, when enriched with new tools and ideas, can be used to characterize the fundamental limits on information flow in networks more complex than a point-point channel or for tasks other than communication. To this end, I will start by presenting our recent solution to a central problem in network communication that has been open for more than 30 years and named “The Capacity of the Relay Channel.” I will then move on to establishing the fundamental limits of inference under rate constraints, and connect it to the information bottleneck method. Finally I will discuss a general principle for jointly designing the feature extractor and the inferrer based on a minimax approach to learning.
Biography: Xiugang Wu is an assistant professor at the University of Delaware, jointly appointed in Electrical and Computer Engineering, and Computer and Information Sciences. Previously, he was a postdoctoral fellow in Electrical Engineering at Stanford University, and received his Ph.D. degree in Electrical and Computer Engineering from the University of Waterloo. His research interests are in information theory, networks, data science, and the interplay between them. He is a recipient of the 2017 NSF Center for Science of Information (CSoI) Postdoctoral Fellowship.
Title: Distributed Statistical Inference Under Local Information Constraints
Abstract: Independent samples from an unknown probability distribution p on a domain of size k are distributed across n players, with each player holding one sample. Each player can communicate L bits to a central referee, with the goal of resolving a prespecified inference problem. When L >= log k bits, the problem reduces to the well studied centralized case, where all the samples are available in one place. We focus on the communication-starved setting L < log k, in which as we will see the landscape changes drastically. We develop a general formulation for inference problems in this distributed setting, and instantiate it for two prototypical inference questions, learning and identity testing. We will consider and discuss the power of shared randomness in distributed inference, and show that for identity testing, schemes without public randomness can be dramatically less efficient than those with. Finally, we show that our framework is general enough to be extended to other distributed settings, in particular to local differential privacy. Based on joint works with Clement Canonne, Cody Freitag, and Himanshu Tyagi.
Biography: Jayadev Acharya is an assistant professor in school of Electrical and Computer Engineering, Cornell University, where he joined after a postdoc at MIT, hosted by Piotr Indyk. He holds a Ph.D from UC San Diego, advised by Alon Orlitsky. His research interests are in algorithms, information theory, machine learning, and statistics. His received a Jack Wolf Student paper award at ISIT’10, and a Best Paper Honorable Mention at ICML’17.
Title: Computational Challenges to Elucidate Notch-Instructed 3D Cancer Genomes
Abstract: During this presentation, I will introduce what could be misregulated in cancer cells. Using the studies from my lab, I plan to point out a number of computational challenges that could be addressed using concepts and approaches commonly used in electrical and computer engineering.
Biography: Dr. Faryabi received his PhD from the Texas A&M University at the Dougherty Lab. Before starting his independent research at the Department of Pathology of the University of Pennsylvania, he was a post-doctoral fellow at the Nussenzweig Lab at the National Cancer Institute where he studied interplay between replication and epigenetic of cancer cells. His lab works the boundaries between chromatin and cancer biology and blends in state-of-the-arts functional and chromatin conformation genomics, high- resolution imaging and computational analysis to advance our understanding of how oncogenic signals reorganize the linear and spatial cancer epigenome. He has been a recipient of NIH Fellow Award for Research Excellence, Cooper Scholar Award, Conquer Cancer Now Award, and Susan G. Komen Foundation Career Catalyst Research Award.
Title: Hybrid ARQ optimizations for wireless networks
Abstract: In this talk, we focus on three ways to improve systems relying on Hybrid ARQ. First, we show that Hybrid ARQ mechanism can be significantly enhanced by doing superposition coding especially when the feedback is delayed. Second, HARQ can be adaptively modified according to previous channel conditions. Here we propose to select the modulation and coding scheme relevantly by using Markov Decision Process. Third, we consider multi-user resource allocation when HARQ is carried out. We especially address the energy efficiency as figure of merit and the Rician channel as propagation model. We show that the optimization process casts into the fractional programming framework.
Biography: Philippe Ciblat was born in Paris, France, in 1973. He received the Engineering degree from Ecole Nationale Superieure des Telecommunications (ENST, now called Telecom ParisTech) and the M.Sc. degree in automatic control and signal processing from University Paris-Sacaly, France, both in 1996, and the Ph.D. degree from University Paris-Est, France, in 2000. He eventually received the HDR degree from University Paris-Est, France, in 2007. In 2001, he was a Postdoctoral Researcher with University of Louvain, Belgium. In 2002, he joined Telecom ParisTech, as an Associate Professor. Since 2011, he has been (full) Professor in the same institution. He served as Associate Editor for the IEEE Communications Letters from 2004 to 2007. From 2008 to 2012, he served as Associate Editor and then Senior Area Editor for the IEEE Transactions on Signal Processing. From 2014, he is member of IEEE Technical Committee “Signal Processing for Communications and Networking”. From 2018, he serves as Associate Editor for the IEEE Transactions on Signal and Information Processing over Networks.