March 2019 - talks - lists.cs.princeton.edu

REMINDER: Colloquium Speaker-Jiajun Wu, tomorrow at 12:30pm
by Emily C. Lawrence 31 Mar '19

31 Mar '19

Colloquium Speaker Jiajun Wu from the Massachusetts Institute of Technology Monday, April 1 - 12:30pm Computer Science - Room 105 Host: Tom Griffiths Learning to See the Physical World [ http://www.cs.princeton.edu/events/25840 | http://www.cs.princeton.edu/events/25840 ] Human intelligence is beyond pattern recognition. From a single image, we're able to explain what we see, reconstruct the scene in 3D, predict what's going to happen, and plan our actions accordingly. In this talk, I will present our recent work on physical scene understanding---building versatile, data-efficient, and generalizable machines that learn to see, reason about, and interact with the physical world. The core idea is to exploit the generic, causal structure behind the world, including knowledge from computer graphics, physics, and language, in the form of approximate simulation engines, and to integrate them with deep learning. Here, deep learning plays two major roles: first, it learns to invert simulation engines for efficient inference; second, it learns to augment simulation engines for constructing powerful forward models. I'll focus on a few topics to demonstrate this idea: building scene representation for both object geometry and physics; learning expressive dynamics models for planning and control; perception and reasoning beyond vision. Bio: Jiajun Wu is a Ph.D. student in Electrical Engineering and Computer Science at Massachusetts Institute of Technology. He received his B.Eng. from Tsinghua University in 2014. His research interests lie in the intersection of computer vision, machine learning, robotics, and computational cognitive science. His research has been recognized through the IROS Best Paper Award on Cognitive Robotics and fellowships from Facebook, Nvidia , Samsung, Baidu, and Adobe, and his work has been covered by major media outlets including CNN, BBC, WIRED, and MIT Tech Review.

1 0

Colloquium Speaker-Henry Corrigan-Gibbs, Thursday, April 4 at 12:30pm
by Emily C. Lawrence 29 Mar '19

29 Mar '19

Colloquium Speaker Henry Corrigan-Gibbs from Stanford University Thursday, April 4 - 12:30pm Computer Science - Room 105 Host: Arvind Narayanan [ https://www.cs.princeton.edu/events/25830 | https://www.cs.princeton.edu/events/25830 ] Protecting Privacy by Splitting Trust When the maker of my phone, smart-watch, or web browser collects data about how I use it, must I trust the manufacturer to protect that sensitive information from theft? When I use the cryptographic hardware module in my laptop, need I trust that it will keep my secrets safe? When I use a messaging app to chat with friends, must I trust the app vendor not to sell the details of my messaging activity for profit? This talk will show that we can get the functionality we want from our systems without having to put blind faith in the correct behavior of the companies collecting our data, building our hardware, or designing our apps. The principle is to split our trust -- among organizations, or devices, or users. I will introduce new cryptographic techniques and systems-level optimizations that make it practical to split trust in a variety of settings. Then, I will present three built systems that employ these ideas, including one that now ships with the Firefox browser. Bio: Henry Corrigan-Gibbs is a Ph.D. candidate at Stanford, advised by Dan Boneh . His research interests are in computer security, applied cryptography, and online privacy. Henry and his collaborators have received the Best Young Researcher Paper Award at Eurocrypt 2018, the 2016 Caspar Bowden Award for Outstanding Research in Privacy Enhancing Technologies, and the 2015 IEEE Security and Privacy Distinguished Paper Award, and Henry's work has been cited by IETF and NIST.

1 0

Distinguished Colloquium Speaker-Corey Sanders '04, Wednesday, April 3 at 4:30pm
by Emily C. Lawrence 28 Mar '19

28 Mar '19

Distinguished Colloquium Speaker Corey Sanders '04 from Microsoft Solutions Wednesday, April 3 - 4:30pm Computer Science - Room 104 (large auditorium) AI and Reshaping the Industry [ http://www.cs.princeton.edu/events/25839 | http://www.cs.princeton.edu/events/25839 ] Artificial Intelligence has quickly moved from being a thing of science fiction to driving business forward across a wide variety of industries including retail, manufacturing, oil and gas, health, and financial services. It is becoming a ubiquitous component of normal customer IT, whether to improve end-customer experience or reduce back-end costs. In this talk, Corey will walk through some real and technical examples of AI and how it is reshaping every industry. Corey will also delve into the responsibility we all have in considering the impact of this technology as it continues to advance and improve. Bio: Corey is the Corporate Vice President for Microsoft Solutions. He is responsible for the sales strategy and corporate technical sales team for Azure, M365, and Dynamics. Previously, Corey was the Head of Product for Azure Compute, responsible for product, strategy, and technical vision for core Azure compute services Azure Virtual Machines (both Windows and Linux), Containers, Kubernetes, OSS, Service Fabric, Event Grid, Service Bus, Management, Migration and serverless computing such as Functions. Since joining Microsoft in 2004 Corey has led program management for multiple Azure services, was a developer in Windows Serviceability team owning areas across the networking and kernel stack for Windows and was the founder of Azure’s Infrastructure-as-a -service business. Corey graduated from Princeton in 2004 with a BSE in Computer Science.

1 0

TODAY March 27 at 4:30pm in B205 E-Quad - Min Ye, Princeton University
by Emily Lawrence 27 Mar '19

27 Mar '19

EE SEMINAR SERIES Speaker: Min Ye, Princeton University Title: Long-standing problems and new challenges in coding theory Day: Wednesday, March 27, 2019 Time: 4:30 pm Room: B205 Engineering Quadrangle Host: Prof. Peter Ramadge Abstract: Emerging problems in data analysis and management create a demand for new methods of data storage, aggregation, and transmission. Systems as diverse as data centers or wireless networks have put forward new requirements to the encoding and representation of digital information. In this talk, we discuss three problems arising from this modern coding-theoretic context: (i) We present coding schemes that minimize communication costs in distributed storage and learning systems. Our schemes were evaluated on Amazon EC2 clusters and were shown to improve upon the state-of-the-art solutions. (ii) We address long-standing coding-theoretic problems regarding Reed-Muller codes: We present advances on the capacity-achieving conjecture as well as a new decoder of Reed-Muller codes which significantly outperforms the state-of-the-art codes adopted in the 5G standard. (iii) We discuss a new method of using neural networks with certain permutation-invariant properties to decode codes with symmetric structures, which suggests a way to construct and efficiently decode the codes with the best possible finite length performance. Bio: Min Ye received his B.S. in Electrical Engineering from Peking University, Beijing, China in 2012, and his Ph.D. in the Department of Electrical and Computer Engineering, University of Maryland, College Park in 2017. He is currently a postdoctoral researcher at Princeton University. His research interests include coding theory, information theory, differential privacy, and machine learning.

1 0

Yichen Chen will present his Pre FPO on Monday, April 1, 2019 at 2pm in CS 105.
by Nicki Mahler 26 Mar '19

26 Mar '19

Yichen Chen will present his Pre FPO on Monday, April 1, 2019 at 2pm in CS 105. The members of his committee are as follows: Mengdi Wang (advisor), Sanjeev Arora (nonreader), Robert Schapire (nonreader), Karthik Narasimhan (reader), Yuxin Chen (reader) The talk title and abstract follow below. All are welcome to attend. Title: Sample-efficient reinforcement learning via utilizing state features Abstract: The field of reinforcement learning has made enormous progress in the last few years, largely due to deep neural networks. Despite success on many practical problems, our reinforcement learning algorithms require huge amount of data. In many applications such as healthcare, data of this large scale is simply not available. In this talk, I explore novel approaches from stochastic optimization that open up opportunities for designing sample-efficient reinforcement learning algorithms. First, I will show how state features, which might come from machine learning algorithms or expert domain knowledge, can facilitate the design of data-efficient reinforcement learning algorithms. Next, I will show that reinforcement learning problems are easier to solve compared to Markov decision processes with full knowledge due to the powerful sampling mechanism in reinforcement learning.

1 0

Colloquium Speaker-Jiajun Wu, Monday, April 1 at 12:30pm
by Emily C. Lawrence 26 Mar '19

26 Mar '19

Colloquium Speaker Jiajun Wu from the Massachusetts Institute of Technology Monday, April 1 - 12:30pm Computer Science - Room 105 Host: Tom Griffiths Learning to See the Physical World [ http://www.cs.princeton.edu/events/25840 | http://www.cs.princeton.edu/events/25840 ] Human intelligence is beyond pattern recognition. From a single image, we're able to explain what we see, reconstruct the scene in 3D, predict what's going to happen, and plan our actions accordingly. In this talk, I will present our recent work on physical scene understanding---building versatile, data-efficient, and generalizable machines that learn to see, reason about, and interact with the physical world. The core idea is to exploit the generic, causal structure behind the world, including knowledge from computer graphics, physics, and language, in the form of approximate simulation engines, and to integrate them with deep learning. Here, deep learning plays two major roles: first, it learns to invert simulation engines for efficient inference; second, it learns to augment simulation engines for constructing powerful forward models. I'll focus on a few topics to demonstrate this idea: building scene representation for both object geometry and physics; learning expressive dynamics models for planning and control; perception and reasoning beyond vision. Bio: Jiajun Wu is a Ph.D. student in Electrical Engineering and Computer Science at Massachusetts Institute of Technology. He received his B.Eng. from Tsinghua University in 2014. His research interests lie in the intersection of computer vision, machine learning, robotics, and computational cognitive science. His research has been recognized through the IROS Best Paper Award on Cognitive Robotics and fellowships from Facebook, Nvidia , Samsung, Baidu, and Adobe, and his work has been covered by major media outlets including CNN, BBC, WIRED, and MIT Tech Review.

1 0

TODAY March 21 at 4:30pm in B205 E-Quad - Rajalakshmi Nandakumar, University of Washington
by Emily Lawrence 21 Mar '19

21 Mar '19

Speaker: Rajalakshmi Nandakumar, University of Washington Title: Computational wireless sensing at scale Day: Thursday, March 21, 2019 Time: 4:30 pm Room: B205 Engineering Quadrangle Host: Prof. Niraj Jha Abstract: Computational wireless sensing is an exciting field of research where we use wireless signals from everyday computing devices to enable sensing. The key challenge is to enable new sensing capabilities that can be deployed at scale and have an impact in the real world. In this talk, I will show how to enable computational wireless sensing at scale by leveraging ubiquitous hardware such as smartphones. Specifically, I will present core technology that can wirelessly sense motion and physiological signals such as breathing using just a smartphone, in a contactless manner. To achieve this, we transform smartphones into active sonar systems. I will show how we can use this technology to detect potentially life-threatening conditions such as opioid overdoses as well as sleep apnea. Finally, I will briefly talk about my work that leverages new hardware trends in micro-controllers to enable wireless sensing applications ranging from object tracking to sensing using live insects such as bees. Bio: Rajalakshmi Nandakumar is a Ph.D. candidate at the Paul G. Allen School of Computer Science and Engineering at the University of Washington. Her research focuses on developing wireless sensing technologies that enable novel applications in various domains including mobile health, user interfaces and IoT networks. She developed the first contactless smartphone-based sleep apnea diagnosis system that was licensed by ResMed Inc. and now used by millions of users for sleep staging. She was recognized with the UW Medicine Judy Su Clinical Research award, Paul Baran Young Scholar award by the Marconi Society and also named as the rising star in EECS by MIT. She has first author papers in top medical journals including Science translational medicine as well as computing venues (CHI, SIGCOMM, SenSys, MobiCom, MobiSys). Her research was awarded multiple accolades and nominations including MobiSys 2015 best paper nominee, CHI 2016 Honorable mention award and SenSys 2018 best paper award.

1 0

NOW in B205 E-Quad - Pat Pannuto, University of California, Berkeley
by Emily Lawrence 18 Mar '19

18 Mar '19

EE SEMINAR SERIES Speaker: Pat Pannuto, University of California, Berkeley Title: A Modular Platform for Nanopower Computing Day: Monday, March 18, 2019 Time: 4:30 pm Room: B205 Engineering Quadrangle Host: Prof. Naveen Verma Beyond the Internet of Things that is emerging today lies the Internet of Everything-the vision of ubiquitous and pervasive computing, of a digital sensory layer enveloping the world. Realizing this vision relies on expanding our capability to place computation in the wild, to then network these sensors, to manage them, and most challengingly, to power them. This talk is about a suite of solutions working in concert to usher in the nanopower computing class. We will discuss architectural innovations that enable power oblivious operation and modular design for millimeter-scale devices. Next, we will look at how to bring multiprocessing, safety, and resource isolation to the deeply embedded computing class. Then, we will consider the some of services, in particular localization, necessary to make pervasive computers useful and viable. Finally, we will look toward the future, with an eye towards what technical challenges remain and how our interactions with computational infrastructure will change as we realize the ubiquitous computing vision. Bio: Pat Pannuto is currently completing his PhD in the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley. He received his MSE in Computer Science and BSE in Computer Engineering from the University of Michigan. Pat's research is in the broad area of networked embedded systems, with contributions to computer architecture, wireless communications, mobile computing, operating systems, and development engineering. Pat's work has been recognized as a Top Pick in Computer Architecture and selected as a Best Paper Finalist at IPSN, and has been awarded NSF, NDSEG, and Qualcomm Innovation fellowships. Pat has also received teaching awards from the Computer Science Department, the College of Engineering, and the Rackham Graduate School at the University of Michigan.

1 0

Unsolved Data Problems: HAPPENING NOW!
by Emily Lawrence 13 Mar '19

13 Mar '19

Wednesday, March 13 4:30 – 5:30 pm Computer Science, Rm 105 https://cdh.princeton.edu/events/2019/03/unsolved-data-problems/ Unsolved Data Problems will introduce faculty and students in the computer and data sciences to the untapped research possibilities inherent in humanities data. A panel of Princeton faculty - Meredith Martin (English), Marina Rustow (History and Near Eastern Studies) and Dan Trueman (Music) - will discuss some of Princeton’s landmark digital humanities projects, and the challenges they’ve faced when transforming historical, multilingual and experimental source material into data and code. Projects discussed include the Princeton Prosody Archive, the Princeton Geniza Lab, and bitKlavier. Jennifer Rexford and Brian Kernighan (Computer Science) will moderate the panel. Help discover innovative algorithmic solutions to these unsolved computational problems. This panel will be of particular interest to researchers working in the fields of: computer vision, natural language processing, machine learning, and audio/music engineering. This event is collaboratively organized by the Center for Digital Humanities, the Department of Computer Science and the Center for Statistics and Machine Learning. Reception to follow immediately after the panel.

1 0

Brian Bullins will present his Pre-FPO "Efficient Higher-Order Optimization for Machine Learning" on Friday, March 15, 2019 at 10am in CS 402.
by Nicki Mahler 11 Mar '19

11 Mar '19

Brian Bullins will present his Pre-FPO "Efficient Higher-Order Optimization for Machine Learning" on Friday, March 15, 2019 at 10am in CS 402. The members of his committee are as follows: Elad Hazan (adviser), Sanjeev Arora, Mark Braverman, Yuxin Chen (ELE), and Yoram Singer All are welcome to attend. The talk title and abstract follow below. Title: Efficient Higher-Order Optimization for Machine Learning Abstract: In recent years, stochastic gradient descent (SGD) has taken center stage for training large-scale models in machine learning. Although some higher-order methods have improved iteration complexity in theory, the per-iteration costs render them unusable when faced with millions of parameters and training examples. In this talk, I will present several of my works which enable higher-order optimization to be as scalable as first-order methods. The first method is a stochastic second-order algorithm for convex optimization, called LiSSA, which uses Hessian information to construct an unbiased Newton step in time linear in the dimension. To bypass the typical efficiency barriers for second-order methods, we harness the ERM structure in standard machine learning tasks. While convex problems allow for global convergence, recent state-of-the-art models, such as deep neural networks, highlight the importance of developing a better understanding of non-convex guarantees. In order to handle this challenging setting, I will present FastCubic, a Hessian-based method which provably converges to first-order critical points faster than gradient descent, while additionally converging to second-order critical points. Finally, I will show how our algorithm FastQuartic combines even higher-order information with “highly smooth acceleration” to guarantee even faster convergence for a large class of convex quartic problems.

1 1