- talks - lists.cs.princeton.edu

Colloquium Speaker, Jonathan Ragan-Kelley, March 22, 12:30pm
by Mitra D. Kelly 16 Mar '16

16 Mar '16

Colloquium Speaker Jonathan Ragan-Kelley, Stanford March 22, 12:30pm Computer Science 105 Organizing Computation for High-Performance Visual Computing Future visual computing applications-from photorealistic real-time rendering, to 4D light field cameras, to pervasive sensing and computer vision-demand orders of magnitude more computation than we currently have. >From data centers to mobile devices, performance and energy scaling is limited by locality (the distance over which data has to move, e.g., from nearby caches, far away main memory, or across networks) and parallelism. Because of this, I argue that we should think of the performance and efficiency of an application as determined not just by the algorithm and the hardware on which it runs, but critically also by the organization of computations and data. For algorithms with the same complexity-even the exact same set of arithmetic operations and data-executing on the same hardware, the order and granularity of execution and placement of data can easily change performance by an order of magnitude because of locality and parallelism. To extract the full potential of our machines, we must treat the organization of computation as a first class concern while working across all levels from algorithms and data structures, to compilers, to hardware. This talk will present facets of this philosophy in systems I have built for visual computing applications from image processing and vision, to 3D rendering, simulation, optimization, and 3D printing. I will show that, for data-parallel pipelines common in graphics, imaging, and other data-intensive applications, the organization of computations and data for a given algorithm is constrained by a fundamental tension between parallelism, locality, and redundant computation of shared values. I will focus particularly on the Halide language and compiler for image processing, which explicitly separates what computations define an algorithm from the choices of organization which determine parallelism, locality, memory footprint, and synchronization. I will show how this approach can enable much simpler programs to deliver performance often many times faster than the best prior hand-tuned C, assembly, and CUDA implementations, while scaling across radically different architectures, from ARM cores, to massively parallel GPUs, to FPGAs and custom ASICs. Jonathan Ragan-Kelley is a postdoc in computer science at Stanford. He works on high-efficiency visual computing, including systems, compilers, and architectures for image processing, vision, 3D rendering, 3D printing, physical simulation, and scientific computing. He earned his PhD in Computer Science at MIT in 2014, where he built the Halide language for high-performance image processing. Halide is now used throughout industry to deploy code to hundreds of millions of smartphones and process tens of billions of images per day. Jonathan previously built the Lightspeed preview system, which was used on over a dozen films at Industrial Light & Magic and was a finalist for an Academy Technical Achievement Award. He has worked in GPU architecture, compilers, and research at NVIDIA, Intel, and ATI. Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium Speaker: Adam Belay, Thursday, March 24- 12:30pm
by Nicole E. Wagenblast 16 Mar '16

16 Mar '16

Colloquium Speaker Adam Belay, Stanford University Thursday, March 24, 12:30pm Computer Science 105 Unleashing Hardware Potential through Better OS Abstractions Datacenter workloads have demanding performance requirements, including the simultaneous need for high throughput, low tail latency, and high server utilization. While modern hardware is compatible with these goals, modern operating systems remain a bottleneck. Better OS abstractions could significantly improve performance, yet deploying these abstractions has become intractable given the size and complexity of today's systems. I will first discuss Dune, a kernel extension that allows OS developers to sidestep software and hardware complexity by running an OS within an ordinary Linux process. With Dune, developers can both access the capabilities of raw hardware and fall back on the functionality of a full Linux environment where convenient. I will then discuss IX and Shinjuku, two generations of new datacenter-focused operating systems that were enabled by Dune. IX provides a novel system call interface that greatly improves network throughput without sacrificing latency. For example, IX improves Memcached's TCP throughput by 5x over Linux. Shinjuku, an ongoing research effort, aims to significantly increase CPU utilization through a centralized approach to intra-server load balancing. Adam Belay is a Ph.D. candidate in Computer Science at Stanford University, where he is a member of the Secure Computer Systems Group and the Multiscale Architecture and Systems Team. Previously, he worked on storage virtualization at VMware Inc. and contributed substantial power management code to the Linux Kernel project. Adam's research area is operating systems and networking. Much of his work has focused on restructuring computer systems so that developers can more easily reach the full performance potential of hardware. Adam has received a Stanford Graduate Fellowship, a VMware Graduate Fellowship, and an OSDI Jay Lepreau Best Paper Award.

1 0

Colloquium Speaker: Dimitris Papailiopoulos Today- 12:30pm
by Nicole E. Wagenblast 07 Mar '16

07 Mar '16

Please note: The Large Auditorium does not permit food, therefore lunch will not be served. Colloquium Speaker Dimitris Papailiopoulos , University of California, Berkeley Monday, March 7, 2016 - 12:30pm Computer Science Large Auditorium, 104 Less Talking, More Learning: Avoiding Coordination In Parallel Machine Learning Algorithms The recent success of machine learning (ML) in both science and industry has generated an increasing demand to support ML algorithms at scale. In this talk, I will discuss strategies to gracefully scale machine learning on modern parallel computational platforms. A common approach to such scaling is coordination-free parallel algorithms, where individual processors run independently without communication, thus maximizing the time they compute. However, analyzing the performance of these algorithms can be challenging, as they often introduce race conditions and synchronization problems. In this talk, I will introduce a general methodology for analyzing asynchronous parallel algorithms. The key idea is to model the effects of core asynchrony as noise in the algorithmic input. This allows us to understand the performance of several popular asynchronous machine learning approaches, and to determine when asynchrony effects might overwhelm them. To overcome these effects, I will propose a new framework for parallelizing ML algorithms, where all memory conflicts and race conditions can be completely avoided. I will discuss the implementation of these ideas in practice, and demonstrate that they outperform the state-of-the-art across a large number of ML tasks on gigabyte-scale data sets. Dimitris Papailiopoulos is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences at UC Berkeley and a member of the AMPLab. His research interests span machine learning, coding theory, and parallel and distributed algorithms, with a current focus on coordination-free parallel machine learning, large-scale data and graph analytics, and the use of codes to speed up distributed computation. Dimitris completed his Ph.D. in electrical and computer engineering at UT Austin in 2014. At Austin he worked under the supervision of Alex Dimakis. In 2015, he received the IEEE Signal Processing Society, Young Author Best Paper Award.

1 0

Colloquium Speaker: Dimitris Papailiopoulos Monday, March 7- 12:30pm- reminder
by Nicole E. Wagenblast 06 Mar '16

06 Mar '16

Please note: The Large Auditorium does not permit food, therefore lunch will not be served. Colloquium Speaker Dimitris Papailiopoulos , University of California, Berkeley Monday, March 7, 2016 - 12:30pm Computer Science Large Auditorium, 104 Less Talking, More Learning: Avoiding Coordination In Parallel Machine Learning Algorithms The recent success of machine learning (ML) in both science and industry has generated an increasing demand to support ML algorithms at scale. In this talk, I will discuss strategies to gracefully scale machine learning on modern parallel computational platforms. A common approach to such scaling is coordination-free parallel algorithms, where individual processors run independently without communication, thus maximizing the time they compute. However, analyzing the performance of these algorithms can be challenging, as they often introduce race conditions and synchronization problems. In this talk, I will introduce a general methodology for analyzing asynchronous parallel algorithms. The key idea is to model the effects of core asynchrony as noise in the algorithmic input. This allows us to understand the performance of several popular asynchronous machine learning approaches, and to determine when asynchrony effects might overwhelm them. To overcome these effects, I will propose a new framework for parallelizing ML algorithms, where all memory conflicts and race conditions can be completely avoided. I will discuss the implementation of these ideas in practice, and demonstrate that they outperform the state-of-the-art across a large number of ML tasks on gigabyte-scale data sets. Dimitris Papailiopoulos is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences at UC Berkeley and a member of the AMPLab. His research interests span machine learning, coding theory, and parallel and distributed algorithms, with a current focus on coordination-free parallel machine learning, large-scale data and graph analytics, and the use of codes to speed up distributed computation. Dimitris completed his Ph.D. in electrical and computer engineering at UT Austin in 2014. At Austin he worked under the supervision of Alex Dimakis. In 2015, he received the IEEE Signal Processing Society, Young Author Best Paper Award.

1 0

Colloquium Speaker: Dimitris Papailiopoulos Monday, March 7- 12:30pm
by Nicole E. Wagenblast 01 Mar '16

01 Mar '16

Please note: The Large Auditorium does not permit food, therefore lunch will not be served. Colloquium Speaker Dimitris Papailiopoulos , University of California, Berkeley Monday, March 7, 2016 - 12:30pm Computer Science Large Auditorium, 104 Less Talking, More Learning: Avoiding Coordination In Parallel Machine Learning Algorithms The recent success of machine learning (ML) in both science and industry has generated an increasing demand to support ML algorithms at scale. In this talk, I will discuss strategies to gracefully scale machine learning on modern parallel computational platforms. A common approach to such scaling is coordination-free parallel algorithms, where individual processors run independently without communication, thus maximizing the time they compute. However, analyzing the performance of these algorithms can be challenging, as they often introduce race conditions and synchronization problems. In this talk, I will introduce a general methodology for analyzing asynchronous parallel algorithms. The key idea is to model the effects of core asynchrony as noise in the algorithmic input. This allows us to understand the performance of several popular asynchronous machine learning approaches, and to determine when asynchrony effects might overwhelm them. To overcome these effects, I will propose a new framework for parallelizing ML algorithms, where all memory conflicts and race conditions can be completely avoided. I will discuss the implementation of these ideas in practice, and demonstrate that they outperform the state-of-the-art across a large number of ML tasks on gigabyte-scale data sets. Dimitris Papailiopoulos is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences at UC Berkeley and a member of the AMPLab. His research interests span machine learning, coding theory, and parallel and distributed algorithms, with a current focus on coordination-free parallel machine learning, large-scale data and graph analytics, and the use of codes to speed up distributed computation. Dimitris completed his Ph.D. in electrical and computer engineering at UT Austin in 2014. At Austin he worked under the supervision of Alex Dimakis. In 2015, he received the IEEE Signal Processing Society, Young Author Best Paper Award.

1 0

date correction. March 3rd, Colloquium Speaker-Rachit Agarwal, 12:30PM
by Mitra D. Kelly 01 Mar '16

01 Mar '16

Colloquium Speaker Rachit Agarwal, AMPLab at UC Berkeley March 3rd, 12:30 Computer Science 105 Title: Building Systems that Query on Compressed Data Abstract: Web services today want to support sophisticated queries, with stringent interactivity constraints. Many recent studies have argued that in-memory query execution is one of the keys to achieving query interactivity. However, as web services scale to larger data sizes, executing queries in memory becomes increasingly challenging. As a result, existing systems fall short of supporting sophisticated interactive queries at scale. In this talk, we present Succinct, a distributed data store that supports functionality comparable to state-of-the-art NoSQL stores and yet, enables query interactivity for an order of magnitude larger data sizes than what is possible today (or, alternatively, up to two orders of magnitude faster queries at scale). Succinct accomplishes this by executing a wide range of queries -- e.g., search, range, and even regular expressions -- directly on compressed data. Succinct achieves scale by storing the input data in a compressed form, and interactivity by avoiding data scans and data decompression. We will also discuss how Succinct's approach of executing queries on compressed data enables a new "lens" for exploring several classical systems problems -- e.g., failure recovery, load spikes during transient failures, skewed workloads, etc. --, and leads to previously unachievable operating points in the system design space. Succinct is open-sourced, and is already being adopted in production clusters of several large-scale web services. Bio: Rachit Agarwal is a postdoc in AMPLab at UC Berkeley, where he leads the Succinct project along with Ion Stoica. His research focuses on the core problems in distributed data-intensive systems, with the goal of building systems that not only aim for practical impact but also have a strong theoretical foundation. He completed his PhD at UIUC, working with Brighten Godfrey and Matthew Caesar, and his undergraduate from IIT Kanpur. During his PhD, he received 2012 UIUC Rambus research award and 2010 Wang-Chung research award for outstanding performance in computer engineering research, and was listed in 2010 UIUC List of Teachers ranked as excellent. Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium Speaker-Rachit Agarwal , Thursday Feb 3, 12:30PM
by Mitra D. Kelly 01 Mar '16

01 Mar '16

Colloquium Speaker Rachit Agarwal, AMPLab at UC Berkeley March 3rd, 12:30 Computer Science 105 Title: Building Systems that Query on Compressed Data Abstract: Web services today want to support sophisticated queries, with stringent interactivity constraints. Many recent studies have argued that in-memory query execution is one of the keys to achieving query interactivity. However, as web services scale to larger data sizes, executing queries in memory becomes increasingly challenging. As a result, existing systems fall short of supporting sophisticated interactive queries at scale. In this talk, we present Succinct, a distributed data store that supports functionality comparable to state-of-the-art NoSQL stores and yet, enables query interactivity for an order of magnitude larger data sizes than what is possible today (or, alternatively, up to two orders of magnitude faster queries at scale). Succinct accomplishes this by executing a wide range of queries -- e.g., search, range, and even regular expressions -- directly on compressed data. Succinct achieves scale by storing the input data in a compressed form, and interactivity by avoiding data scans and data decompression. We will also discuss how Succinct's approach of executing queries on compressed data enables a new "lens" for exploring several classical systems problems -- e.g., failure recovery, load spikes during transient failures, skewed workloads, etc. --, and leads to previously unachievable operating points in the system design space. Succinct is open-sourced, and is already being adopted in production clusters of several large-scale web services. Bio: Rachit Agarwal is a postdoc in AMPLab at UC Berkeley, where he leads the Succinct project along with Ion Stoica. His research focuses on the core problems in distributed data-intensive systems, with the goal of building systems that not only aim for practical impact but also have a strong theoretical foundation. He completed his PhD at UIUC, working with Brighten Godfrey and Matthew Caesar, and his undergraduate from IIT Kanpur. During his PhD, he received 2012 UIUC Rambus research award and 2010 Wang-Chung research award for outstanding performance in computer engineering research, and was listed in 2010 UIUC List of Teachers ranked as excellent. Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium Speaker -Greg Durrett, Tuesday March 1, 12:30pm
by Mitra D. Kelly 29 Feb '16

29 Feb '16

Colloquium Speaker Greg Durrett, UC Berkeley Tuesday March 1, 12:30pm Computer Science 105 riven Text Analysis with Joint Models One reason that analyzing text is hard is that it involves dealing with deeply entangled linguistic variables: objects like syntactic structures, semantic types, and discourse relations depend on one another in complex ways. Our work tackles several facets of text analysis using joint modeling, combining model components both across and within the various subtasks of this analysis. This model structure allows us to pass information between these entangled subtasks and propagate high-confidence predictions rather than errors. Critically, our models have the capacity to learn key linguistic phenomena as well as other important patterns in the data; that is, linguistics tells us how to structure these models, then the data injects knowledge into them. We describe state-of-the-art systems for a range of tasks, including syntactic parsing, entity resolution, and document summarization. Bio: Greg is a Ph.D. candidate at UC Berkeley working on natural language processing with Dan Klein. He is interested in building structured machine learning models for a wide variety of text analysis problems and downstream NLP applications. His work is comprised of two broad thrusts: first, designing joint models that combine information across different tasks or different views of a problem, and second, building systems that strike a balance between being linguistically motivated and data-driven Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium speaker Yuxin Chen, TODAY, 4:30pm
by Nicole E. Wagenblast 29 Feb '16

29 Feb '16

Colloquium Speaker Yuxin Chen, Stanford University Monday, February 29, 4:30pm Computer Science 104- large auditorium The Power of Nonconvex Paradigms for High-Dimensional Estimation In various scenarios in the information sciences, one wishes to estimate a large number of parameters from highly incomplete / imperfect data samples. A growing body of recent work has demonstrated the effectiveness of convex relaxation --- in particular, semidefinite programming --- for solving many problems of this kind. However, the computational cost of such convex paradigms is often unsatisfactory, which limits applicability to large-dimensional data. This talk follows another route: by formulating the problems into nonconvex programs, we attempt to optimize the nonconvex objectives directly. To illustrate the power of this strategy, we present two concrete stories. The first involves solving a random quadratic system of equations, which spans many applications ranging from the century-old phase retrieval problem to various latent-variable models in machine learning. The second is about recovering a collection of discrete variables from noisy pairwise difference measurements, which arises when one wishes to jointly align multiple images or to retrieve the genome phases from paired sequencing reads. We propose novel nonconvex paradigms for solving these two problems. In both cases, the proposed solutions can be accomplished within linear time, while achieving a statistical accuracy that is nearly un-improvable. Yuxin Chen is currently a postdoctoral scholar in the Department of Statistics at Stanford University, supervised by Prof. Emmanuel Candès. He received the Ph.D. degree in Electrical Engineering and M.S. in Statistics from Stanford University, M.S. in Electrical and Computer Engineering from the University of Texas at Austin, and B.E. in microelectronics from Tsinghua University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.

1 0

Colloquium speaker Yuxin Chen, Monday Feb 29, 4:30pm- reminder
by Nicole E. Wagenblast 26 Feb '16

26 Feb '16

Colloquium Speaker Yuxin Chen, Stanford University Monday, February 29, 4:30pm Computer Science 104- large auditorium The Power of Nonconvex Paradigms for High-Dimensional Estimation In various scenarios in the information sciences, one wishes to estimate a large number of parameters from highly incomplete / imperfect data samples. A growing body of recent work has demonstrated the effectiveness of convex relaxation --- in particular, semidefinite programming --- for solving many problems of this kind. However, the computational cost of such convex paradigms is often unsatisfactory, which limits applicability to large-dimensional data. This talk follows another route: by formulating the problems into nonconvex programs, we attempt to optimize the nonconvex objectives directly. To illustrate the power of this strategy, we present two concrete stories. The first involves solving a random quadratic system of equations, which spans many applications ranging from the century-old phase retrieval problem to various latent-variable models in machine learning. The second is about recovering a collection of discrete variables from noisy pairwise difference measurements, which arises when one wishes to jointly align multiple images or to retrieve the genome phases from paired sequencing reads. We propose novel nonconvex paradigms for solving these two problems. In both cases, the proposed solutions can be accomplished within linear time, while achieving a statistical accuracy that is nearly un-improvable. Yuxin Chen is currently a postdoctoral scholar in the Department of Statistics at Stanford University, supervised by Prof. Emmanuel Candès. He received the Ph.D. degree in Electrical Engineering and M.S. in Statistics from Stanford University, M.S. in Electrical and Computer Engineering from the University of Texas at Austin, and B.E. in microelectronics from Tsinghua University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.

1 0