- talks - lists.cs.princeton.edu

Colloquium speaker Kristen Grauman, Wed Dec 3, 4:30pm
by Nicole E. Wagenblast 01 Dec '14

01 Dec '14

C olloquium Speaker Kristen Grauman, University of Texas at Austin Wednesday, December 3, 4:30pm Computer Science 105 Searching and Browsing Visual Data Widespread visual sensors and unprecedented connectivity have left us awash with visual data---from online photo collections, home videos, news footage, medical images, or surveillance feeds. How can we efficiently browse image and video collections based on semantically meaningful criteria? How can we bring order to the data, beyond manually defined keyword tags? I will present work exploring these questions in the context of interactive visual search and summarization. In particular, I’ll first introduce attribute representations that connect visual properties to human describable terms. I’ll show how these attributes enable both fine-grained content-based retrieval as well as new forms of human supervision for recognition problems. Then, I’ll overview our recent work on video summarization, where the goal is to automatically transform a long video into a short one. Using videos captured with egocentric wearable cameras, we’ll see how hours of data can be distilled to a succinct visual storyboard that is understandable in just moments. Together, these ideas are promising steps towards widening the channel of communication between humans and computer vision algorithms, which is critical to facilitate efficient browsing of large-scale image and video collections. This is work done with Adriana Kovashka, Yong Jae Lee, Devi Parikh, Lu Zheng, Bo Xiong, and Dinesh Jayaraman. Kristen Grauman is an Associate Professor in the Department of Computer Science at the University of Texas at Austin. Her research in computer vision and machine learning focuses on visual search and object recognition. Before joining UT-Austin in 2007, she received her Ph.D. in the EECS department at MIT, in the Computer Science and Artificial Intelligence Laboratory. She is an Alfred P. Sloan Research Fellow and Microsoft Research New Faculty Fellow, a recipient of NSF CAREER and ONR Young Investigator awards, the Regents' Outstanding Teaching Award from the University of Texas System in 2012, the PAMI Young Researcher Award in 2013, the 2013 Computers and Thought Award from the International Joint Conference on Artificial Intelligence, and a Presidential Early Career Award for Scientists and Engineers (PECASE) in 2013. She and her collaborators were recognized with the CVPR Best Student Paper Award in 2008 for their work on hashing algorithms for large-scale image retrieval, and the Marr Best Paper Prize at ICCV in 2011 for their work on modeling relative visual attributes.

1 0

Colloquium speaker Edoardo Airoldi- Tuesday, Dec 2, 4:30pm
by Nicole E. Wagenblast 01 Dec '14

01 Dec '14

Colloquium Speaker Edoardo Airoldi, Harvard University Tuesday, December 2, 4:30pm Computer Science 105 Statistical and machine learning challenges in the analysis of large networks Network data --- i.e., collections of measurements on pairs, or tuples, of units in a population of interest --- are ubiquitous nowadays in a wide range of machine learning applications, from molecular biology to marketing on social media platforms. Surprisingly, assumptions underlying popular statistical methods are often untenable in the presence of network data. Established machine learning algorithms often break when dealing with combinatorial structure. And the classical notions of variability, sample size and ignorability take unintended connotations. These failures open to door to a number of technical challenges, and to opportunities for introducing new fundamental ideas and for developing new insights. In this talk, I will discuss open statistical and machine learning problems that arise when dealing with large networks, mostly focusing on modeling and inferential issues, and provide an overview of key technical ideas and recent results and trends. Edoardo M. Airoldi is an Associate Professor of Statistics at Harvard University, where he leads the Harvard Laboratory for Applied Statistical Methodology. He holds a holds Ph.D. in Computer Science and an M.Sc. in Statistics from Carnegie Mellon University, and a B.Sc. in Mathematical Statistics and Economics from Bocconi University. His current research focuses on statistical theory and methods for designing and analyzing experiments in the presence of network interference, and on inferential issues that arise in models of network data. He works on applications in molecular biology and proteomics, and in social media analytics and marketing. Airoldi is the recipient several research awards including the ONR Young Investigator Award, the NSF CAREER Award, and the Alfred P. Sloan Research Fellowship, and has received several outstanding paper awards including the Thomas R. Ten Have Award for his work on causal inference, and the John Van Ryzin Award for his work in biology. He has recently advised the Obama for America 2012 campaign on their social media efforts, and serves as a technical advisor at Nanigans and Maxpoint.

1 0

Peng Sun's Pre FPO: Monday, December 8 at 10am in CS 402, Title: "Network Management Architecture for Delivering Cloud Services"
by Nicki Gotsis 01 Dec '14

01 Dec '14

Peng Sun will present his Pre FPO on Monday, December 8 at 10am in CS 402 His committee members are: Jennifer Rexford (advisor), Dave Walker and Nick Feamster (readers), Margaret Martonosi and Kai Li (non-readers) Everyone is invited to attend his talk. The talk title and abstract follow below: Title: Network Management Architecture for Delivering Cloud Services Abstract: Efficient and reliable networks are critical to good end-to-end performance of cloud services. In both the datacenters of cloud providers and the ISPs connecting clients, many network management applications were built to improve efficiency and reliability. However, the traditional way of building management applications is not sustainable, relying on heavy human involvement. Software-defined networking (SDN) promises to simplify programming management applications. Yet SDN's current scope is limited to routing control on network devices within a single domain. My research works explore expanding the concept of SDN for better end-to-end performance of cloud services. In the exploration, I take a research approach of achieving and balancing three objectives: 1) design simple and expressive programming abstraction, 2) build efficient and scalable system for execution, and 3) deploy and run in real world. Using this research approach, I have explored three directions. The first direction is deep into network devices to make infrastructure easy to program. The network operating system I built, called Statesman, introduces a three-view-of-network-state abstraction for programming a wide range of management applications. Besides the abstraction, Statesman enables automatic and safe operation of multiple applications in parallel. Published in SIGCOMM'14, Statesman has been deployed and operational worldwide in Microsoft Azure for over one year. The second direction is expanding into servers in datacenters to leverage the fine-grained knowledge of cloud services in the server OS. The system, called HONE, offers a three-stage programming framework to simplify traffic management. HONE's runtime executes the programs across servers and network devices in an efficient and scalable fashion. HONE has been integrated into products for Verizon Business Cloud, and appeared in JNSM July'14. The final direction is reaching out to ISPs for better delivery of cloud services. This work-in-progress aims at a direct, fine-grained, and self-contained solution to inbound traffic engineering (TE) for enterprises. The solution employs three tiers of abstractions to simplify the inbound-TE policies, and it scales up with a distributed enforcement system to control the inbound ISPs for traffic flows. This ongoing work is in collaboration with OIT on our campus network.

1 0

Colloquium Speaker Sham Kakade Monday, today, 4:30pm
by Nicole E. Wagenblast 01 Dec '14

01 Dec '14

Colloquium Speaker Sham Kakade, Microsoft Research, New England Monday, December 1, 4:30pm Computer Science 105 Some Algorithmic Challenges in Statistics Machine learning is seeing tremendous progress in its impact on society. Along with this progress comes an increasing role for both scalable algorithms and theoretical foundations; the hope being that the such progress can facilitate further breakthroughs on core AI problems. This will talk will survey recent progress and future challenges at the intersection of computer science and statistics, with a focus on three areas: How can we learn the interactions between observed variables, where there exist certain latent (or hidden) causes which help to explain the correlations in the observed data. Such settings where latent variable models have seen successes include document (or topic) modeling, hidden Markov models (say for modeling time series of acoustic signals), and discovering communities of individuals in social networks. The second is that of stochastic optimization. Many problems that arise in science and engineering are those in which we only have a stochastic approximation to the underlying problem at hand (e.g. linear regression or other problems where our objective function is a sample average). Such problems highlight some of the challenges we face at the interface of computer science and statistics: should we use a highly (numerically) accurate algorithm (with costly time and space requirements) or a crude stochastic approximation scheme like stochastic gradient descent (which is light on memory and simple to implement, yet has a poor convergence rate)? Finally, I will provide a brief discussion with regards to future challenges inspired by the impressive successes of deep learning. A recurring theme is that algorithmic advances can provide new practical techniques for statistical estimation. Sham Kakade is a principal research scientist scientist at Microsoft Research, New England. His research focus is on designing scalable and efficient algorithms for machine learning and artificial intelligence; he has worked (and has continued interests) in areas such as statistics, optimization, probability theory, algorithms, economics, and neuroscience. Previously, Dr. Kakade was an associate professor at the Department of Statistics, Wharton, University of Pennsylvania (from 2010-2012) and was an assistant professor at the Toyota Technological Institute at Chicago. Before this, he did a postdoc in the Computer and Information Science department at the University of Pennsylvania under the supervision of Michael Kearns. Dr. Kakade completed his PhD at the Gatsby Unit where his advisor was Peter Dayan. Before Gatsby, Dr. Kakade was an undergraduate at Caltech where he did his BS in physics.

1 0

Fwd: EE Seminar - December 1, 2014, 4:30 pm, E-Quad B205 - Prof. Johan Pouwelse
by Jennifer Rexford 01 Dec '14

01 Dec '14

Reminder... > ---------- Forwarded message ---------- > From: Lori A. Bailey <lbailey(a)princeton.edu> > Date: Wed, Nov 26, 2014 at 11:14 AM > Subject: EE Seminar - December 1, 2014, 4:30 pm, E-Quad B205 - Prof. Johan Pouwelse > To: "ee-seminar(a)princeton.edu" <ee-seminar(a)princeton.edu> > > > > > > Speaker: Prof. Johan Pouwelse > > Delft University of Technology > > Title: Anonymous HD Video Streaming and Reputations > > Date: Monday, December 1, 2014 > > Time: 4:30 p.m. > > Room: E-Quad, Room B205 > > Host: Prof. Prateek Mittal > > > > Abstract: Privacy is non-existent for 99% of all web traffic. We provide Tor-like privacy protection for the most demanding class of Internet traffic: HD-quality video streams. To date, no viable solution has ever been presented for privacy-enhanced streaming. The challenge is to scale to one billion users, the current usage level of Youtube. > > > > Youtube tracks what you like, when you view it, where you watch it and what your friends are viewing. Scaling privacy-enhancing technology while retaining security has proving to be insurmountable. Many proposals exists, but no system has any significant usage (e.g. 1+ million users) or has even been fully implemented. We present the first functional fully self-organising darknet for HD-quality video streaming. The key building block is our BarterCast reputation system, indicating which users help others by relaying their encrypted traffic. Our darknet, called Tribler, evolved over a period of nine years, has a thriving user community, comes with proper unit tests, coding tutorials, and is designed to be fault-tolerant by not relying on any central server, tracker or website. > > > > Biography: Dr. ir. J.A. Pouwelse is an associate professor at Delft University of Technology, specialized in self-organizing systems and privacy-enhancing technology. His research group conducts experimental research in the field of self-organizing systems, reputation systems, crowdsourcing, onion routing and peer-to-peer technology. The methodology of the group is proving the validity of novel scientific ideas with thousands of real-world Internet users. Our Tribler software is installed by 1.6 million users, it serves as a living laboratory and testing ground for next-generation self-organizing systems. Previously Dr. Pouwelse delivered a statement for the FTC in Washington, was a visiting scientist at MIT, and spent several summers at Harvard to study mechanisms for cooperation. >

1 0

Colloquium Speaker Sham Kakade Monday, Dec 1, 4:30pm
by Nicole E. Wagenblast 30 Nov '14

30 Nov '14

Colloquium Speaker Sham Kakade, Microsoft Research, New England Monday, December 1, 4:30pm Computer Science 105 Some Algorithmic Challenges in Statistics Machine learning is seeing tremendous progress in its impact on society. Along with this progress comes an increasing role for both scalable algorithms and theoretical foundations; the hope being that the such progress can facilitate further breakthroughs on core AI problems. This will talk will survey recent progress and future challenges at the intersection of computer science and statistics, with a focus on three areas: How can we learn the interactions between observed variables, where there exist certain latent (or hidden) causes which help to explain the correlations in the observed data. Such settings where latent variable models have seen successes include document (or topic) modeling, hidden Markov models (say for modeling time series of acoustic signals), and discovering communities of individuals in social networks. The second is that of stochastic optimization. Many problems that arise in science and engineering are those in which we only have a stochastic approximation to the underlying problem at hand (e.g. linear regression or other problems where our objective function is a sample average). Such problems highlight some of the challenges we face at the interface of computer science and statistics: should we use a highly (numerically) accurate algorithm (with costly time and space requirements) or a crude stochastic approximation scheme like stochastic gradient descent (which is light on memory and simple to implement, yet has a poor convergence rate)? Finally, I will provide a brief discussion with regards to future challenges inspired by the impressive successes of deep learning. A recurring theme is that algorithmic advances can provide new practical techniques for statistical estimation. Sham Kakade is a principal research scientist scientist at Microsoft Research, New England. His research focus is on designing scalable and efficient algorithms for machine learning and artificial intelligence; he has worked (and has continued interests) in areas such as statistics, optimization, probability theory, algorithms, economics, and neuroscience. Previously, Dr. Kakade was an associate professor at the Department of Statistics, Wharton, University of Pennsylvania (from 2010-2012) and was an assistant professor at the Toyota Technological Institute at Chicago. Before this, he did a postdoc in the Computer and Information Science department at the University of Pennsylvania under the supervision of Michael Kearns. Dr. Kakade completed his PhD at the Gatsby Unit where his advisor was Peter Dayan. Before Gatsby, Dr. Kakade was an undergraduate at Caltech where he did his BS in physics.

1 0

Colloquium Speaker Sham Kakade Monday, Dec 1, 4:30pm
by Nicole E. Wagenblast 25 Nov '14

25 Nov '14

Colloquium Speaker Sham Kakade, Microsoft Research, New England Monday, December 1, 4:30pm Computer Science 105 Some Algorithmic Challenges in Statistics Machine learning is seeing tremendous progress in its impact on society. Along with this progress comes an increasing role for both scalable algorithms and theoretical foundations; the hope being that the such progress can facilitate further breakthroughs on core AI problems. This will talk will survey recent progress and future challenges at the intersection of computer science and statistics, with a focus on three areas: How can we learn the interactions between observed variables, where there exist certain latent (or hidden) causes which help to explain the correlations in the observed data. Such settings where latent variable models have seen successes include document (or topic) modeling, hidden Markov models (say for modeling time series of acoustic signals), and discovering communities of individuals in social networks. The second is that of stochastic optimization. Many problems that arise in science and engineering are those in which we only have a stochastic approximation to the underlying problem at hand (e.g. linear regression or other problems where our objective function is a sample average). Such problems highlight some of the challenges we face at the interface of computer science and statistics: should we use a highly (numerically) accurate algorithm (with costly time and space requirements) or a crude stochastic approximation scheme like stochastic gradient descent (which is light on memory and simple to implement, yet has a poor convergence rate)? Finally, I will provide a brief discussion with regards to future challenges inspired by the impressive successes of deep learning. A recurring theme is that algorithmic advances can provide new practical techniques for statistical estimation. Sham Kakade is a principal research scientist scientist at Microsoft Research, New England. His research focus is on designing scalable and efficient algorithms for machine learning and artificial intelligence; he has worked (and has continued interests) in areas such as statistics, optimization, probability theory, algorithms, economics, and neuroscience. Previously, Dr. Kakade was an associate professor at the Department of Statistics, Wharton, University of Pennsylvania (from 2010-2012) and was an assistant professor at the Toyota Technological Institute at Chicago. Before this, he did a postdoc in the Computer and Information Science department at the University of Pennsylvania under the supervision of Michael Kearns. Dr. Kakade completed his PhD at the Gatsby Unit where his advisor was Peter Dayan. Before Gatsby, Dr. Kakade was an undergraduate at Caltech where he did his BS in physics.

1 0

Thiago Pereira will present his pre-FPO 11/25/14 at 10am in rm 402
by Nicki Gotsis 24 Nov '14

24 Nov '14

Thiago Pereira will present his Pre-FPO on Tuesday, November 25, 2014 at 10am in rm 402. The members of his committee is Szymon Rusinkiewicz (advisor), Wojciech Matusik and Tom Funkhouser (readers), Adam Finkelstein and Jianxiong Xiao (non readers). Everyone is invited to attend his talk. The talk title and abstract follow below: "New methods for the computational fabrication of appearance" Computational fabrication of appearance is an exciting and challenging new research direction. Its objective is creating objects that interact with light in a manner prescribed in a computational model. It can have great impact in a number of different fields including product prototyping and design, medical applications such as realistic prosthesis, lighting interior environments in architecture and watermarks in security. In this thesis, we propose three works on appearance fabrication: an appearance similarity metric, a light routing algorithm and new reflectance fabrication process. First, recent spatially varying reflectance (svBRDF) printing systems can reproduce an input document as a combination of matte, glossy and metallic inks. Due to the limited number of inks, this reproduction process incurs some distortion. In this work, we present an svBRDF gamut mapping algorithm that minimizes distortions in the angular and spatial domains. To preserve a material's perceived variation with lighting and view, we introduce an improved BRDF similarity metric that builds on both experimental results on reflectance perception and on the statistics of natural lighting environments. Our experiments show better preservation of object color and highlights, as validated quantitatively as well as through a perceptual study. As for the spatial domain, we show how to adapt traditional color gamut mapping methods to svBRDFs. Our solution takes into account the contrast between regions, achieving better preservation of textures and edges. Second, despite recent interest in digital fabrication, there are still few algorithms that provide control over how light propagates inside a solid object. Existing methods either work only on the surface or restrict themselves to light diffusion in volumes. We use multi-material 3D printing to fabricate objects with embedded optical fibers, exploiting total internal reflection to guide light inside an object. We introduce automatic fiber design algorithms together with new manufacturing techniques to route light between two arbitrary surfaces. Our implicit algorithm optimizes light transmission by minimizing fiber curvature and maximizing fiber separation while respecting constraints such as fiber arrival angle. We also discuss the influence of different printable materials and fiber geometry on light propagation in the volume and the light angular distribution when exiting the fiber. Our methods enable new applications such as surface displays of arbitrary shape, touch-based painting of surfaces and sensing a hemispherical light distribution in a single shot. Third, existing BRDF fabrication methods still have many limitations such as: symmetry of light scattering or restriction to small scattering angles. We propose to use magnetic reflective pigments such that we can control their orientation by applying a magnetic field. We use dynamic magnetic fields for controlling not only off-specular lobes direction but also lobe width and anisotropy. We show how this magnetic control can be coupled with a projector to achieve new spatially-varying anisotropic BRDFs.

1 0

Fwd: EE Seminar - December 1, 2014, 4:30 pm, E-Quad B205 - Prof. Johan Pouwelse
by Jennifer Rexford 24 Nov '14

24 Nov '14

> From: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > Date: November 24, 2014 at 12:42:33 PM EST > To: ee-seminar(a)Princeton.EDU > Subject: EE Seminar - December 1, 2014, 4:30 pm, E-Quad B205 - Prof. Johan Pouwelse > Reply-To: "Lori A. Bailey" <lbailey(a)Princeton.EDU> > > > Speaker: Prof. Johan Pouwelse > Delft University of Technology > Title: Anonymous HD Video Streaming and Reputations > Date: Monday, December 1, 2014 > Time: 4:30 p.m. > Room: E-Quad, Room B205 > Host: Prof. Prateek Mittal > > Abstract: Privacy is non-existent for 99% of all web traffic. We provide Tor-like privacy protection for the most demanding class of Internet traffic: HD-quality video streams. To date, no viable solution has ever been presented for privacy-enhanced streaming. The challenge is to scale to one billion users, the current usage level of Youtube. > > Youtube tracks what you like, when you view it, where you watch it and what your friends are viewing. Scaling privacy-enhancing technology while retaining security has proving to be insurmountable. Many proposals exists, but no system has any significant usage (e.g. 1+ million users) or has even been fully implemented. We present the first functional fully self-organising darknet for HD-quality video streaming. The key building block is our BarterCast reputation system, indicating which users help others by relaying their encrypted traffic. Our darknet, called Tribler, evolved over a period of nine years, has a thriving user community, comes with proper unit tests, coding tutorials, and is designed to be fault-tolerant by not relying on any central server, tracker or website. > > Biography: Dr. ir. J.A. Pouwelse is an associate professor at Delft University of Technology, specialized in self-organizing systems and privacy-enhancing technology. His research group conducts experimental research in the field of self-organizing systems, reputation systems, crowdsourcing, onion routing and peer-to-peer technology. The methodology of the group is proving the validity of novel scientific ideas with thousands of real-world Internet users. Our Tribler software is installed by 1.6 million users, it serves as a living laboratory and testing ground for next-generation self-organizing systems. Previously Dr. Pouwelse delivered a statement for the FTC in Washington, was a visiting scientist at MIT, and spent several summers at Harvard to study mechanisms for cooperation. > > > > >

1 0

Fwd: EE Seminar -TODAY at 4:30 pm, E-Quad B205 - Prof. Lakshminarayanan Subramanian4:30 pm, E-Quad B205 - Prof. Lakshminarayanan Subramanian
by Jennifer Rexford 24 Nov '14

24 Nov '14

> > SEMINAR TODAY > > > Speaker: Prof. Lakshminarayanan Subramanian > > New York University > > Title: Societal Network Science: Where Societal Networks meets Data Science > > Date: Monday, November 24, 2014 > > Time: 4:30 p.m. > > Room: E-Quad, Room B205 > > Host: Prof. Prateek Mittal > > > > Abstract: This talk focuses on a new emerging area of research, which we term "Societal Network Science", where the goal is to understand important societal, social and human indicators based on the network characteristics of how humans, physical and digital entities interact. Societal networks aim to capture meaningful human interaction patterns and the potential causal relationships between these interactions. In this talk, we will outline our recent results on three related but seemingly disparate problems in societal network science: disease surveillance, road traffic networks and event networks. In the space of disease surveillance, our group has worked on a 3-year long project to build a fine-grained dengue surveillance system in the country of Pakistan, where we can predict the incidence level of dengue using citizen driven data at a sub-city granularity, 14 days ahead of time. In the space of road traffic networks, we will illustrate how one can predict sudden traffic jams in NY city with high confidence 3 minutes before they happen; surprisingly, 3 minutes is sufficient to potentially avoid the jam too! In the space of event networks, we will show how one can use NLP techniques to learn networks of real world events from news corpora and how such event networks can be used in the prediction of real-world social phenomena like drought, price variations and disease outbreaks. Our longer term vision is to build an automated analytics engine for societal network science that can learn and infer socio-economic phenomena from highly diverse and noisy data sources. > > > > Biography: Lakshminarayanan Subramanian is an Associate Professor in the Courant Institute of Mathematical Sciences at NYU. His research interests are in the areas of networked systems, big data analytics and computing for development. He is a member of the NYU Systems group and co-established the Center for Technology and Economic Development (CTED) at NYU. He is also affiliated with the Center for Data Science and the Global Institute of Public Health (GIPH) at NYU. He is the co-founder of Entrupy Inc, a startup on surface fingerprinting. He completed his PhD from UC Berkeley and undergraduate studies from IIT-Madras. >

1 0