October 2018 - talks - lists.cs.princeton.edu

Yushan Su will present her General Exam on Thursday, October 25, 2018 at 12:30pm in CS 401.
by Nicki Gotsis 19 Oct '18

19 Oct '18

Yushan Su will present her General Exam on Thursday, October 25, 2018 at 12:30pm in CS 401. The members of her committee are Kai Li (advisor), Kyle Jamieson, and David August. Everyone is invited to attend her talk, and those faculty wishing to remain for the oral exam following are welcome to do so. Her abstract and reading list follow below. Title: FPGA Accelerator for 2D Iterative Compressed Sensing Reconstruction Abstract: During the past decade, Compressed Sensing (CS) has emerged as an important method in signal acquisition and processing, allowing the reconstruction of original data from sparse signals with a significantly lower sampling rate than traditional sampling techniques. Among the CS reconstruction algorithms, iterative algorithms can achieve high accuracy, but require intensive computation. As quite a few applications of CS such as biomedical imaging and radar imaging have real-time constraints, how to accelerate the performance of iterative CS reconstruction algorithms is a challenging problem. In this research, we explore how to build an accelerator of 2D iterative CS reconstruction algorithms for FPGAs, since they can potentially achieve higher performance to power ratio than CPUs and GPUs, and they are more programmable than ASICs. Our approach is based on the methodology to mapping to spatial computing fabrics. We have implemented 2D Iterative Soft-shrinkage Threshold Algorithm (ISTA) on Intel/Altera Stratix 10 FPGA. Comparing with the performance of a CPU and a GPU with similar technologies, the projected performance to power ratio is 24.2x than CPU, 2.4x than GPU running the open-source BART MRI toolkit. Reading List: Candès, Emmanuel J., and Michael B. Wakin. "An introduction to compressive sampling." IEEE signal processing magazine 25.2 (2008): 21-30. Lustig, Michael, David Donoho, and John M. Pauly. "Sparse MRI: The application of compressed sensing for rapid MR imaging." Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine 58.6 (2007): 1182-1195. Compton, Katherine, and Scott Hauck. "Reconfigurable computing: a survey of systems and software." ACM Computing Surveys (csuR) 34.2 (2002): 171-210. Parashar, Angshuman, et al. "Triggered instructions: a control paradigm for spatially-programmed architectures." ACM SIGARCH Computer Architecture News. Vol. 41. No. 3. ACM, 2013. Lee, Edward A., and David G. Messerschmitt. "Synchronous data flow." Proceedings of the IEEE 75.9 (1987): 1235-1245. Chen, Ren, Hoang Le, and Viktor K. Prasanna. "Energy efficient parameterized FFT architecture." Field Programmable Logic and Applications (FPL), 2013 23rd International Conference on. IEEE, 2013. Zeng, Hanqing, et al. "A framework for generating high throughput CNN implementations on FPGAs." Proceedings of the 2018 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays. ACM, 2018. Bai, Lin, et al. "High-speed compressed sensing reconstruction on FPGA using OMP and AMP." Electronics, Circuits and Systems (ICECS), 19th IEEE International Conference on. 2012. Stanislaus, Jerome LVM, and Tinoosh Mohsenin. "Low-complexity FPGA implementation of compressive sensing reconstruction." Computing, Networking and Communications (ICNC), 2013 International Conference on. IEEE, 2013. Quan, Yinghui, et al. "FPGA implementation of real-time compressive sensing with partial Fourier dictionary." International Journal of Antennas and Propagation 2016 (2016). Book List: Hennessy, John L., and David A. Patterson. Computer architecture: a quantitative approach. Elsevier, 2011. Cormen, Thomas H., et al. Introduction to algorithms. MIT press, 2009.

1 0

Yinda Zhang will present his FPO "From Pixels to Scenes: Recovering 3D Geometry and Semantics for Indoor Environments" on Thursday, 10/25/2018 at 3:00 pm in CS 302.
by Nicki Gotsis 18 Oct '18

18 Oct '18

Yinda Zhang will present his FPO "From Pixels to Scenes: Recovering 3D Geometry and Semantics for Indoor Environments" on Thursday, 10/25/2018 at 3:00 pm in CS 302. The members of his committee are as follows: Thomas Funkhouser (adviser); Examiners: Ryan Adams, Olga Russakovsky, and Thomas Funkhouser; Readers: Szymon Rusinkiewicz and James Hays (Georgia Institute of Technology) A copy of his thesis is available upon request. Everyone is invited to attend her talk. The talk title and abstract follow below: Understanding the 3D geometry and semantics of real environments is in critically high demand for many applications, such as autonomous driving, robotics, and augmented reality. However, it is extremely challenging due to imperfect and noisy measurements from real sensors, limited access to ground truth data, and cluttered scenes exhibiting heavy occlusions and intervening objects. To address these issues, this thesis introduces a series of works that produce a geometric and semantic understanding of the scene in both pixel-wise and holistic 3D representations. Starting from estimating a depth map, which is a fundamental task in many approaches for reconstructing the 3D geometry of the scene, we introduce a learning-based active stereo system that is trained in a self-supervised fashion and reduces the disparity error to 1/10th of other canonical stereo systems. To handle a more common case where only one input image is available for scene understanding, we create a high-quality synthetic dataset facilitating pre-training of data-driven approaches, and demonstrating that we can improve the surface normal estimation and improve raw depth measurements from commodity RGBD sensors. Lastly, we pursue holistic 3D scene understanding by estimating a 3D representation of the scene, in which objects and room layout are represented using 3D bounding box and planar surfaces respectively. We propose methods to produce such a representation from either a single color panorama or a depth image, leveraging scene context. On the whole, these proposed methods produce understanding of both 3D geometry and semantics from the most ne-grained pixel level to the holistic scene scale, building foundations that support future work in 3D scene understanding.

1 0

TODAY 10/17 at 4:30pm in B205 - Negar Kiyavash, Georgia Institute of Technology
by Emily Lawrence 17 Oct '18

17 Oct '18

EE SEMINAR SERIES Speaker: Negar Kiyavash, Georgia Institute of Technology Title: Network Inference: from Passive to Active Learning Day: Wednesday, October 17, 2018 Time: 4:30 pm Room: B205 Engineering Quadrangle Host: Prof. H. Vincent Poor Abstract: One of the paramount challenges of this century is that of understanding complex, dynamic, large-scale networks. Such high-dimensional networks, including social, financial, and biological networks, cover the planet and dominate modern life. In this talk, we propose novel approaches to inference in such networks, for both active (interventional) and passive (observational) learning scenarios. We highlight how timing could be utilized as a degree of freedom that provides rich information about the dynamics. This information allows resolving direction of causation even when only a subset of the nodes is observed (latent setting). In the presence of large data, we propose algorithms that identify optimal or near-optimal approximations to the topology of the network often by harnessing side information about the model class. Bio: Negar Kiyavash is a joint Associate Professor in the H. Milton Stewart School of Industrial & Systems Engineering (ISyE) and the School of Electrical and Computer Engineering (ECE) at Georgia Institute of Technology (Gatech). Prior to joining Gatech, when was a Willett Faculty Scholar and a joint Associate Professor of Industrial and Enterprise Engineering (IE) and Electrical and Computer Engineering (ECE) at the University of Illinois . Her research interests are in design and analysis of algorithms for network inference and security. She is a recipient of NSF CAREER and AFOSR YIP awards and the Illinois College of Engineering Dean's Award for Excellence in Research.

1 0

Tomorrow, October 18, 4:30pm, E-Quad B205, Ayfer Ozgur Aydin, Stanford
by Emily Lawrence 17 Oct '18

17 Oct '18

TOMORROW Department of Electrical Engineering Seminar Series Speaker: Prof. Ayfer Ozgur Aydin Stanford University Title: A Geometric Approach to two Problems in Networks: Learning High-Dimensional Distributions and Communication with Relays Date: Thursday, October 18, 2018 Time: 4:30 pm Room: Engineering Quadrangle B205 Host: Prof. Yuxin Chen Abstract: We investigate the pivotal role geometry can play in characterizing the fundamental limits of information flow in two different network problems. We first consider a classical problem in network information theory: characterizing the capacity of a so-called relay channel. Even though the relay channel has been one of the central problems in network information theory, its capacity remains unknown for almost four decades. We solve an open problem posed by Cover in 1987. This problem, which Cover calls the "Capacity of the Relay Channel", corresponds to characterizing the capacity of this channel at one special operating point. Our approach is geometric and builds on an extension we develop for the isoperimetric inequality on a high-dimensional sphere. We then turn to another network problem, where the end goal is not to communicate a message but learn a high-dimensional distribution or parameter from its distributed samples under communication constraints. We first provide a geometric characterization of Fisher information from quantized samples. We then use this characterization to prove tight minimax bounds for distributed estimation of common statistical models (such as the product Bernoulli model, multinomial model, dense/sparse Gaussian location models). Our results show that the impact of the communication constraint can be drastically different depending on the tail behavior of the score function of the model. Some of our results recover or strengthen existing results in this area with simpler and more transparent proofs. We conclude with a discussion on future directions in the intersection of geometry and information theory. First part of the talk is joint work with Xiugang Wu and Leighton Barnes. Second part is joint work with Leighton Barnes, Yanjun Han and Tsachy Weissman. Bio: Ayfer Ozgur received her Ph.D. degree in 2009 from the Information Processing Group at EPFL, Switzerland. In 2010 and 2011, she was a post-doctoral scholar at the same institution. She is an Assistant Professor in the Electrical Engineering Department at Stanford University since 2012. Her research interests include distributed communication and learning, wireless systems, and information theory. Dr. Ozgur received the EPFL Best Ph.D. Thesis Award in 2010, an NSF CAREER award in 2013, the Okawa Foundation Research Grant and the IEEE 2018 Communication Theory Technical Committee (CTTC) Early Achievement Award in 2018. This seminar is supported with funds from the Korhammer Lecture Series.

1 0

Fwd: NEXT PACM Colloquium: Oct 22 - Matthew Salganik, Princeton University
by Jennifer Rexford 16 Oct '18

16 Oct '18

> From: "Gina M. Holland" <gholland(a)Princeton.EDU> > Subject: NEXT PACM Colloquium: Oct 22 - Matthew Salganik, Princeton University > Date: October 16, 2018 at 3:21:01 PM EDT > To: "pacmdistribution(a)math.princeton.edu" <pacmdistribution(a)math.princeton.edu> > > The next PACM Colloquium will be given by Matthew Salganik, Princeton University > > Title and abstract are below. > > ================= > DATE/LOCATION: Monday, October 22, 2018, 4:00 – 5:00 PM / 214 Fine Hall > > SPEAKER: Matthew Salganik, Princeton University > > TITLE: The Fragile Families Challenge: A Scientific Mass Collaboration > > ABSTRACT: > > Sociologists have long theorized about the processes through which childhood experiences shape life outcomes. However, statistical models that use data on family background and childhood experiences to predict life outcomes often have poor predictive performance. These empirical results have lead pioneers of the field muse that random chance must play an important role. In this talk, we present results from the Fragile Families Challenge, a scientific mass collaboration based on the Fragile Families and Child Wellbeing Study, a birth cohort study of about 5,000 families from large US cities. The study began with a probability sample of newborns, and for more than 15 years, researchers have followed these families to collect information related to child and family development as reported by the child as well as the child’s mother, father, primary caregiver, and teachers. During the Fragile Families Challenge, 159 research teams from 68 institutions in 7 countries used this high-dimensional survey data to produce statistical and machine learning models to predict six life outcomes. Results suggest that (a) modern machine learning methods enabled predictive performance that outpaced approaches more common in social science, but (b) overall predictive performance was poor. The talk will include a discussion of the potential reasons for poor predictive performance in this setting, open methodological questions raised by our results, and the potential for mass collaboration to address other scientific and policy questions. > > > Matthew Salganik is Professor of Sociology at Princeton University, and he is affiliated with several of Princeton's interdisciplinary research centers: the Office for Population Research, the Center for Information Technology Policy, the Center for Health and Wellbeing, and the Center for Statistics and Machine Learning. His research interests include social networks and computational social science. He is the author of Bit by Bit: Social Research in the Digital Age <http://www.bitbybitbook.com/>. Salganik's research has been published in journals such as Science, PNAS, Sociological Methodology, and Journal of the American Statistical Association. His papers have won the Outstanding Article Award from the Mathematical Sociology Section of the American Sociological Association and the Outstanding Statistical Application Award from the American Statistical Association. Popular accounts of his work have appeared in the New York Times, Wall Street Journal, Economist, and New Yorker. Salganik's research is funded by the National Science Foundation, National Institutes of Health, Joint United Nations Program for HIV/AIDS (UNAIDS), Russell Sage Foundation, Sloan Foundation, Facebook, and Google. During sabbaticals from Princeton, he has been a Visiting Professor at Cornell Tech and a Senior Research at Microsoft Research. > >

1 0

ACM/IEEE Computer Graphics Film Show - October 18, 2018
by Szymon Rusinkiewicz 16 Oct '18

16 Oct '18

The Princeton ACM / IEEE-CS chapters kick off our season of meetings with the annual computer graphics film show, featuring the latest and greatest computer-animated short films direct from the ACM SIGGRAPH conference held this summer. These films were selected for the juried Computer Animation Festival at the conference, to demonstrate advancements in graphics research, highlight creative visual effects, educate with scientific visualizations, and show the many uses of real-time game rendering engines. These films were created by large organizations and individuals, professional animators and students, on high-end clusters and ordinary personal computers. Prof. Szymon Rusinkiewicz will be providing commentary on the computer graphics techniques used in these films. Come and enjoy our most popular meeting of each year! As in the past, we will be co-celebrating Martin Gardner's 104th birthday by hosting a Celebration of Mind party (http://www.celebrationofmind.org/) as part of our meeting. Gardner (1914-2010) was the editor of the Mathematical Games column of Scientific American from 1956-1981, and we will be celebrating the things that delighted him - puzzles, games, math, magic and more! Plan to arrive at 7:30pm for refreshments and a special guest speaker, with the film show starting around 8:15pm. All Princeton ACM / IEEE-CS meetings are open to the public. Students and their parents are welcome. There is no admission charge, and refreshments are served. Date: Thursday, October 18, 2018, 8:15 pm (Refreshments, networking, and G4G at 7:30 pm) Place: Friend Center Auditorium, Room 101, Princeton University, Olden & William Streets On-line info: http://PrincetonACM.acm.org/

1 0

Shuran Song will present her FPO "Data-Driven 3D Scene Understanding" on Tuesday, 10/23/2018 at 10am in CS 402.
by Nicki Gotsis 16 Oct '18

16 Oct '18

Shuran Song will present her FPO "Data-Driven 3D Scene Understanding" on Tuesday, 10/23/2018 at 10am in CS 402. The members of her committee are as follows: Thomas Funkhouser (adviser); Examiners: Adam Finkelstein, Szymon Rusinkiewicz, and Thomas Funkhouser; Readers: Olga Russakovsky and Alberto Rodriguez (MIT) A copy of her thesis is available upon request. Everyone is invited to attend her talk. The talk title and abstract follow below: Intelligent robots require advanced vision capabilities to perceive and interact with the real physical world. While computer vision has made great strides in recent years, its predominant paradigm still focuses on analyzing image pixels to infer two dimensional outputs (e.g. 2D bounding boxes, or labeled 2D pixels.), which remain far from sufficient for real-world robotics applications. This dissertation presents the use of amodal 3D scene representations that enable intelligent systems to not only recognize what is seen (e.g. Am I looking at a chair?), but also predict contextual information about the complete 3D scene beyond visible surfaces (e.g. What could be behind the table? Where should I look to find an exit?). More specifically, it presents a line of work that demonstrates the power of these representations: First it shows how 3D amodal scene representation can be used to improve the performance of a traditional tasks such as object detection. We present SlidingShapes and DeepSlidingShapes for the task of amodal 3D object detection, where the system is designed to fully exploit the advantage of 3D information provided by depth images. Second, we introduce the task of semantic scene completion and our approach SSCNet, whose goal is to produce a complete 3D voxel representation of volumetric occupancy and semantic labels for a scene from a single-view depth map observation. Third, we introduce the task of semantic-structure view extrapolation and our approach Im2Pano3D, which aims to predict the 3D structure and semantic labels for a full 360 panoramic view of an indoor scene when given only a partial observation. Finally, we present two large-scale datasets (SUN RGB-D and SUNCG) that enable the research on data-driven 3D scene understanding. This dissertation demonstrates that leveraging a complete 3D scene representations not only significantly improves algorithm's performance for traditional computer vision tasks, but also paves the way for new scene understanding tasks that have previously been considered ill-posed given only 2D representations.

1 0

Thomas Macrina will present his General Exam on Monday, October 22, 2018 at 2pm in PNI 159
by Nicki Gotsis 15 Oct '18

15 Oct '18

Thomas Macrina will present his General Exam on Monday, October 22, 2018 at 2pm in PNI 159. The members of his committee are as follows: Sebastian Seung (adviser), Jonathan Pillow (Neuro), and Barbara Engelhardt. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. Title Aligning petascale serial-section electron microscopy images, and analyzing structure + function of networks of real neurons Abstract Routinely reconstructing neural wiring diagrams based on petascale serial section electron microscopy would be a boon for neuroscience. The major obstacle is producing a low-error alignment of 2D images into a 3D stack. I will discuss how we have used elastic models based on normalized cross correlation correspondences, and now optical flow models based on deep learning to overcome this obstacle. With routine reconstructions on the horizon, focus can turn to analysis of these wiring diagrams. I will also discuss my work investigating these real neural networks: correlating them with neural activity, and using contacts between neurons as a basis for connection properties . Reading List Braitenberg, Valentino, and Almut Schuez. 1998. Cortex Statistics Geometry Neuronal Connectivity. Springer. Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. 2016. Deep Learning (Adaptive Computation and Machine Learning). Edited by Francis Bach. The MIT Press. Jaderberg, Max, Karen Simonyan, Andrew Zisserman, and Koray Kavukcuoglu. 2015. “Spatial Transformer Networks.” In Advances in Neural Information Processing Systems 28, edited by C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett, 2017–25. Curran Associates, Inc. Kasthuri, Narayanan, Kenneth Jeffrey Hayworth, Daniel Raimund Berger, Richard Lee Schalek, José Angel Conchello, Seymour Knowles-Barley, Dongil Lee, et al. 2015. “Saturated Reconstruction of a Volume of Neocortex.” Cell 162 (3): 648–61. Kubota, Yoshiyuki. 2014. “Untangling GABAergic Wiring in the Cortical Microcircuit.” Current Opinion in Neurobiology 26 (June): 7–14. Lee, Wei-Chung Allen, Vincent Bonin, Michael Reed, Brett J. Graham, Greg Hood, Katie Glattfelder, and R. Clay Reid. 2016. “Anatomy and Function of an Excitatory Network in the Visual Cortex.” Nature 532 (7599): 370–74. Luo, Liqun. 2015. Principles of Neurobiology. 1 edition. Garland Science. Mishchenko, Yuriy, Tao Hu, Josef Spacek, John Mendenhall, Kristen M. Harris, and Dmitri B. Chklovskii. 2010. “Ultrastructural Analysis of Hippocampal Neuropil from the Connectomics Perspective.” Neuron 67 (6): 1009–20. Ranjan, Anurag, and Michael J. Black. 2016. “Optical Flow Estimation Using a Spatial Pyramid Network.” arXiv [cs.CV]. arXiv. http://arxiv.org/abs/1611.00850. Saalfeld, Stephan, Richard Fetter, Albert Cardona, and Pavel Tomancak. 2012. “Elastic Volume Reconstruction from Series of Ultra-Thin Microscopy Sections.” Nature Methods 9 (7): 717–20. Yoo, I., D. G. C. Hildebrand, W. F. Tobin, and W. C. A. Lee. 2017. “ssEMnet: Serial-Section Electron Microscopy Image Registration Using a Spatial Transformer Network with Learned Features.” Deep Learning in. https://link.springer.com/chapter/10.1007/978-3-319-67558-9_29.

1 0

Xiaoqi Chen will present his General Exam on Tuesday, Oct 23, 2018 at 10am in CS 302
by Nicki Gotsis 12 Oct '18

12 Oct '18

Xiaoqi Chen will present his General Exam on Tuesday, Oct 23, 2018 at 10am in CS 302. The members of his committee are Jennifer Rexford (adviser), Mark Braverman, and David Walker. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. Abstract: Network measurement is a vital tool for network operators to diagnose outages, optimize performance, and detect attacks. Recently, the development of programmable switches has enabled us to run measurement algorithms in the network switch directly, and analyze packets up to Tbps in throughput. However, the programming model of programmable switches is extremely constrained, which restricts the types of algorithms we can run in them. To achieve much-needed measurement goals of network operators, we design tailored algorithms to adapt to the programming constraints imposed by practical programmable switches. We first present PRECISON, which attempts to answer the Heavy-Hitter Flow Detection problem (find out which flows sent the largest number of packets). PRECISION can accurately identify the heavy-hitter flows using a small amount of memory, by recirculating a small number of packets probabilistically. Then, we present Snappy, which tries to solve the Heavy Flow in the Queue problem (which flows occupied a large fraction of queuing buffer). Snappy can pinpoint the bursty flows causing ephemeral long queues and potential packet loss, by maintaining multiple traffic snapshots of short time intervals. Our measurement algorithms enable network operators to perform immediate actions against these specific network flows, inhibiting congestion in real-time, while potentially improving service quality for other network flows. Reading list: Textbook: * Larry L. Peterson and Bruce S. Davie. Computer Networks: A Systems Approach . Elsevier, 2011 (5th ed). Papers: 1. [E2E ‘84] Jerome H. Saltzer, David P. Reed, and David D. Clark. "End-to-end arguments in system design." ACM Transactions on Computer Systems (TOCS) 2.4 (1984): 277-288. 2. [Ether ’99] Will E. Leland, Murad S. Taqqu, Walter Willinger, and Daniel V. Wilson. "On the self-similar nature of Ethernet traffic (extended version)." IEEE/ACM Transactions on Networking (ToN) 2.1 (1994): 1-15. 3. [Click ’00] Eddie Kohler, Robert Morris, Benjie Chen, John Jannotti, and M. Frans Kaashoek. "The Click modular router." ACM Transactions on Computer Systems (TOCS) 18.3 (2000): 263-297. 4. [SS ’05] Ahmed Metwally, Divyakant Agrawal, and Amr El Abbadi. "Efficient computation of frequent and top-k elements in data streams." International Conference on Database Theory . Springer, Berlin, Heidelberg, 2005. 5. [pFabric ’13] Mohammad Alizadh, Shuang Yang, Milad Sharif, Sachin Katti, Nick McKeown, Balaji Prabhakar, and Scott Shenker. "pfabric: Minimal near-optimal datacenter transport." ACM SIGCOMM Computer Communication Review . Vol. 43. No. 4. ACM, 2013. 6. [P4 ‘14] Pat Bosshart, Dan Daly, Glen Gibb, Martin Izzard, Nick McKeown, Jennifer Rexford, Cole Schlesinger* Dan Talayco, Amin Vahdat, George Varghese, and David Walker. "P4: Programming protocol-independent packet processors." ACM SIGCOMM Computer Communication Review 44.3 (2014): 87-95. 7. [UnivMon ‘16] Zaoxing Liu, Antonis Manousis, Gregory Vorsanger, Vyas Sekar, and Vladimir Braverman. "One sketch to rule them all: Rethinking network flow monitoring with UnivMon." Proceedings of the 2016 ACM SIGCOMM Conference . ACM, 2016. 8. [RAP ‘16] Ran Ben Basat, Gil Einziger, Roy Friedman, and Yaron Kassner. "Randomized admission policy for efficient top-k and frequency estimation." INFOCOM 2017-IEEE Conference on Computer Communications, IEEE . IEEE, 2017. 9. [StarFlow ‘18] John Sonchack, Oliver Michel, Adam J. Aviv, Eric Keller, and Jonathan M. Smith. "Scaling Hardware Accelerated Network Monitoring to Concurrent and Dynamic Queries With* Flow." 2018 USENIX Annual Technical Conference (USENIX ATC 18) . USENIX Association, 2018. 10. [AFQ ’18] Naveen Kr. Sharma, Ming Liu, Kishore Atreya, and Arvind Krishnamurthy. "Approximating Fair Queueing on Reconfigurable Switches." USENIX Symposium on Networked Systems Design and Implementation . 2018. Catching heavy network flows using programmable switches

1 0

Speaker-Sreepathi Pai, TODAY at 12:30
by Emily C. Lawrence 12 Oct '18

12 Oct '18

Speaker Sreepathi Pai from University of Rochester Friday, October 12, 2018 - 12:30pm Computer Science - Room 105 Host: Tyler Sorensen and Margaret Martonosi “High-performance Graph Processing on GPUs” Abstract: Back in 2013, implementing high-performance graph algorithms for GPUs required manual coding in CUDA, a slow and difficult process. To significantly lower the level of effort required, we created IrGL , a language and compiler specifically for generating high-performance graph algorithm implementations for GPUs. Powered by three throughput optimizations, the IrGL -generated code outperformed nearly all handwritten graph algorithms achieving speedups of up to 6x. Freed from the drudgery of writing low-level code, IrGL has allowed us to look at a number of problems revolving around graphs. We've used the high-performance implementations to identify key memory system bottlenecks that limit performance on current GPU architectures. We've also translated graph database queries to IrGL and executed them on GPUs. In the course of extending this to the general problem of subgraph isomorphism (a key primitive in graph databases), we were named GraphChallenge 2017 champions for our implementation of the triangle-counting and k-truss problems. Along the way, we also built Groute , a runtime for asynchronous multi-GPU graph analytics that has achieved order-of-magnitude improvements over existing synchronous implementations. We sped up exhaustive testing of software by traversing graphs that were too big to materialize in memory. Recently, we have also used IrGL's ability to generate hundreds of variants of the same graph algorithm to explore correctness and performance portability issues on GPUs. Many interesting questions still remain unexplored, however, and I will summarize our current efforts in this area. [Joint work with Keshav Pingali , Tal Ben-Nun, Michael Sutton, M. Amber Hassaan , Chad Voegele , Yi-Shan Lu, Ahmet Celik , Milos Gligoric , Sarfraz Khurshid , Tyler Sorensen and Alastair Donaldson] Bio: Sreepathi Pai is an Assistant Professor of Computer Science at the University of Rochester. His research interests are in compilers, programming languages and implementation, performance models and computer architecture. His most recent research has revolved around the IrGL compiler that produces high-performance GPU code for graph algorithms. He earned his PhD at the Indian Institute of Science and his B.E. in Computer Engineering at the University of Mumbai. Prior to joining the Department of Computer Science at Rochester, he was a Postdoctoral Fellow at the University of Texas at Austin.

1 0