May 2017 - talks - lists.cs.princeton.edu

Andrej Risteski will present his pre-FPO on Tuesday, May 16th, 2017 at 1pm in CS 302
by Nicki Gotsis 11 May '17

11 May '17

Andrej Risteski will present his pre-FPO on Tuesday, May 16th, 2017 at 1pm in CS 302. The members of his committee are: Readers: Sanjeev Arora (advisor), Rong Ge (Duke University), Elad Hazan Non-readers: Bernard Chazelle, Mark Braverman The talk title and abstract follow below. All are welcome to attend. Title : New techniques for learning and inference in probabilistic graphical models Abstract : A common theme in machine learning is succinct modeling of distributions over large domains. Probabilistic graphical models are one of the most expressive frameworks for doing this. The two major tasks involving graphical models are learning and inference. Learning is the task of calculating the best fit graphical model from raw data, while inference is the task of answering probabilistic queries for a known graphical model (for example what is the marginal distribution of one of the variables or what is the distribution of a subset of variables, after conditioning on the values of some other subset of variables). Learning can be thought of as finding a graphical model that “explains” the raw data, while the inference queries extract the “knowledge” the graphical model contains. This talk will concern new provable techniques for performing both of these tasks. I will focus on a few vignettes: in the context of inference, I will talk about a new understanding of a class of algorithms for calculating partition functions, called variational methods through a combination of convex programming hierarchies — a recent area of interest in the area of combinatorial optimization — and entropy approximations based on low-order moments of a distribution. In the context of learning, I will talk about alternating-minimization and method-of-moments techniques for learning noisy-or Bayesian networks and topic modeling. The former are used for modeling the causal structure of diseases and symptoms, and the later for modeling the latent structure of text.

1 0

Ohad Fried will present his FPO on Wednesday, May 17, 2017 at 1:30pm in CS 401.
by Nicki Gotsis 10 May '17

10 May '17

Ohad Fried will present his FPO on Wednesday, May 17, 2017 at 1:30pm in CS 401. The members of his committee are: Adam Finkelstein (adviser); readers: Tom Funkhouser and Eli Shechtman (Adobe); examiners: Szymon Rusinkiewicz and David Dobkin. A copy of his thesis is available in Room 310. Everyone is invited to attend his talk. The talk title and abstract follow below: PHOTO MANIPULATION, THE EASY WAY The typical smartphone user has many thousands of photos in their personal collection. Photo acquisition is effortless, and the next challenge is in devising methods to easily edit such large collections. Specifically, we need manipulation algorithms that are powerful enough for experts, yet simple for novices to master. We identified three key directions to empower novice users with expert-level editing capabilities while maintaining an overall simplicity in the process. Those directions are (1) better selection masks, (2) high-level goal-centric algorithms and (3) domain specific algorithms. In this thesis we give examples from each category. Given a photo, a novice user will typically either not edit it at all, or apply a simple global operation such as exposure correction. In contrast, a professional photo editor might perform local edits, specifying a selection mask to limit the operation to specific photo regions, or combining regions from several photos into a single composition. To ease selection mask creation we present a new patch embedding technique that allows for single-click selection masks. Novices often think in terms of goals (e.g. improve lighting, de-clutter photo) and less in technical terms such as color spaces and image layers. One example of a high-level goal is the removal of distracting elements from photos. The task is motivated by the way professional photographers operate. They carefully frame the scene and might move objects around in order to stage the perfect photo. We define “photo distractors” as the elements that, if removed, would improve the photo. Using a simple slider interaction we allow users to automatically remove such distractors from photos. It is at times useful to tailor solutions to specific photo types. As an example we show that, specifically for human heads, simple controls can induce sophisticated edits. Given a single portrait photo as input, we can change the pose of the head and the camera distance. This allows users to correct the “selfie effect”, i.e. big noses and small ears or to transform distant photos into selfies. We conclude by discussing how each of these directions can be further explored to enable better image editing tools for novices.

1 0

Colloquium Speaker: Nima Anari, Tuesday, May 9th, 12:30pm
by Nicole E. Wagenblast 09 May '17

09 May '17

Colloquium Speaker Dr. Nima Anari, Stanford Univeristy Tuesday, May 9- 12:30pm Computer Science 105 Algorithm Design Using Polynomials In this talk I will present new methods for algorithm design and showcase results and breakthroughs obtained by looking through the lens of polynomials. I will start with the best known approximation algorithm for estimating the solution of the asymmetric traveling salesman problem, one of the oldest problems in computer science and optimization. Then, I will discuss a general framework based on computing inner products of polynomials that is used to design approximation algorithms for several important computational problems: volume maximization, counting matchings in bipartite graphs, Nash welfare maximization, and computing the permanent of positive semidefinite matrices. Nima Anari is a postdoctoral researcher at Stanford University working with Amin Saberi. His research interests are in the design and analysis of algorithms, with a recent focus on applications of polynomials in this endeavor. He recently completed his Ph.D. in computer science at UC Berkeley, where he was advised by Satish Rao and was part of the theory group. Prior to that he received his B.Sc. in computer engineering and mathematics from Sharif university of technology.

2 1

Alexander Beatson will present his general exam on Wednesday, May 10, 2017 at 3pm in CS 301.
by Nicki Gotsis 09 May '17

09 May '17

Alexander Beatson will present his general exam on Wednesday, May 10 , 2017 at 3pm in CS 301. The members of his committee are: Han Liu (ORFE, adviser), Elad Hazan, and Arvind Narayanan. 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: Deep learning in some non-iid settings In real-world settings observed data may often be non-iid. There might be differences between the training and deployment distributions, which we can model in the worst case as perturbation by an adversary [1], or the training data may be drawn from different distributions at different times or at different locations, and we might wish to learn from the global distribution without storing or communicating the data [12]. I will discuss some work on attacks on machine learners and on non-iid distributed deep learning with communication constraints. Attacks on machine learners: One threat model for machine learning involves an adversary perturbing the training set. We show that even a “blind” adversary with no knowledge of the training data can harm learning (meaningfully lower-bound the learner’s effective sample size) via simple data injection strategies. Another threat model involves an adversary perturbing the examples observed in deployment. Linear and deep models are particularly vulnerable to such adversarial examples [2-4]. We show that deep generative models can be used to correctly classify adversarial examples or detect them as being far from the manifold of real data. Non-iid distributed deep learning: “Federated learning” [12] features communication constraints and data distributed non-iid across nodes. We propose an asynchronous algorithm for this setting which uses the Hessian of the likelihood at each node to weight model averaging and compute proximal constraints. This may be derived as an iterated natural proximal method or an application of the Laplace approximation [9]. Experiments show this converges faster than other asynchronous algorithms when data are non-iid across nodes. We may approximate the algorithm by sending only a small fraction of the variable updates from each node, sampled according to diagonal of that node’s Hessian. This maintains improved convergence in the non-iid setting, and in both iid and non-iid settings reduces the communication volume needed to reach a given accuracy. Reading List: Attacks on machine learners: Overview of the field: [1] Concrete problems in AI safety . Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Mané. Adversarial examples in deep learning: [2] Explaining and harnessing adversarial examples . Ian J. Goodfellow, Jonathon Shlens, Christian Szegedy. [3] Practical black-box attacks against deep learning systems using adversarial examples . Nicolas Papernot, Patrick McDaniel, Ian Goodfellow, Somesh Jha, Z. Berkay Celik, Ananthram Swami. [4] Towards Evaluating the Robustness of Neural Networks . Nicholas Carlini, David Wagner. Previously proposed defenses: [5] Biologically inspired protection of deep networks from adversarial attacks . Aran Nayebi, Surya Ganguli. [6] Parseval Networks: Improving Robustness to Adversarial Examples . Cisse Moustapha, Bojanowski Piotr, Grave Edouard, Dauphin Yann, Usunier Nicolas. Deep generative models: [7] Generative Adversarial Networks . Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio. [8] Auto-Encoding Variational Bayes . Diederik P Kingma, Max Welling. Non-iid and distributed deep learning: The Laplace approximation: [9] A practical Bayesian framework for backpropogation networks . David J. C. MacKay. Using the Laplace approximation to estimating weight importances in deep learning: [10] Optimal Brain Damage . Yann LeCun, John S. Denker, Sara A. Solla. [11] Overcoming catastrophic forgetting in neural networks . James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran, Raia Hadsell. Federated learning: [12] Communication-Efficient Learning of Deep Networks from Decentralized Data . H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, Blaise Agüera y Arcas. [13] Federated Learning: Strategies for Improving Communication Efficiency . Jakub Konečný, H. Brendan McMahan, Felix X. Yu, Peter Richtárik, Ananda Theertha Suresh, Dave Bacon. Textbook: [14] Deep Learning . Ian Goodfellow, Yoshua Bengio, Aaron Courville.

1 1

Ghassen Jerfel will present his MSE talk on Monday, May 15th, 2017 at 8:15am in CS 302.
by Nicki Gotsis 08 May '17

08 May '17

Ghassen Jerfel will present his MSE talk on Monday, May 15th, 2017 at 8:15am in CS 302. The members of his committee are Barbara Engelhardt (adviser) and Elad Hazan. The title and abstract of his talk are below. All are welcome to attend. Title: Combining multi-modality and non-linearity for density estimation: Boosted Stochastic Backpropagation for Variational Inference Abstract: Deep Generative Models, such as Variational Autoencoders, couple the structured and regularized latent variable representations of probabilistic graphical models with nonlinear likelihoods to learn flexible representations of complex high-dimensional data. Variational inference provides fast approximation of the posterior for VAEs as it formulates the problem as an optimization problem to find the closest posterior over some parametric family of distributions. However, general VI theory and standard VAE literature restrict the approximation family such that it does not contain the true posterior regardless of how long VI is run (e.g. limited to unimodal diagonal Gaussians for VAEs). In this paper, we propose a model-free, simple to implement, and easily parallelizable meta-algorithm which may be used in conjunction with recent VI advances. Based on the functional gradient descent view of boosting, our algorithm iteratively improves the variational approximation of the posterior and is guaranteed to converge at a rate of O(1/k) to the true posterior. We derive boosted Gaussian stochastic backpropagation as a principled method to incorporate and mix new encoders into a VAE model to refine the estimation of the latent space. We study boosted stochastic backpropagation on standard benchmarks for unsupervised learning, applying it to density estimation tasks on synthetic data, the MNIST, NORB, and OMNIGLOT datasets. We demonstrate competitive results that emphasize the advantage of iterative refinement and multi-modality in DGMs.

1 0

Riley Simmons-Edler will present his general exam on Friday, May 12, 2017 in CS 401 at 10am.
by Nicki Gotsis 08 May '17

08 May '17

Riley Simmons-Edler will present his general exam on Friday, May 12, 2017 in CS 401 at 10am. The members of his committee are: Tom Funkhouser, Sebastian Seung, and Szymon Rusinkiewicz. 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: "Self-Supervised and Adversarial Image Intrinsic Decomposition" Abstract: Image decomposition into intrinsic components, such as albedo and lighting, is a highly underconstrained problem. For any given image, there are infinitely many possible predictions that satisfy the relationship albedo * lighting = color. However, as humans we can define a distribution of "plausible" predictions for a given image using a combination of our internal model of what a valid intrinsic looks like and the known content of the original image. It is therefore reasonable to think it is viable to train a deep neural network to do this task by constraining the output in a similar fashion. Here, I present my work on using self-supervision and adversarial neural networks to learn an albedo and lighting decomposition for real photographs, where no ground truth data exists nor can exist. By using an adversarial network trained on a distribution of synthetic albedo and lighting, I project a given color image into the space of possible intrinsics, while using self-supervision via a reconstruction loss from the intrinsics to the input image to constrain the predictions to reproduce the input image. This combination of loss terms makes it possible to learn an accurate transformation from color input to intrinsic factors without any paired training data, allowing for training directly on real images and thus improved performance compared to previous approaches which can only use synthetic training data. Textbook: "Deep Learning," Goodfellow, Bengio, and Courville, 2016 Papers: Directly Related to my project(methods and previous application works): "Direct Intrinsics: Learning Albedo-Shading Decomposition by Convolutional Regression," Narihira, Maire, and Yu, 2015 "DARN: a Deep Adversarial Residual Network for Intrinsic Image Decomposition," Lettry, Vanhoey, and Van Gool, 2016 "Wasserstein GAN," Arjovsky, Chintala, and Bottou, 2017 "Multi-Scale Context Aggregation by Dilated Convolutions," Yu and Koltun, 2016 Other Similar Influential Works(in image to image translation): "Learning from Simulated and Unsupervised Images through Adversarial Training," Shrivastava et al, 2016 "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks," Zhu, Park, Isola, and Efros, 2017 Classic Papers: "Generative Adversarial Nets," Goodfellow et al, 2014 "Imagenet Classification with Deep Convolutional Neural Networks," Krizhevsky, Sutskever, and Hinton, 2012 "Backpropagation Applied to Handwritten Zip Code Recognition," LeCun et al, 1989 (the original convolutional neural networks paper)

1 0

REMINDER - ELE Seminar Today, 4:30 pm in B205
by Nicole E. Wagenblast 08 May '17

08 May '17

Speaker: Pramod Viswanath, University of Illinois Title: Geometries of Word Embeddings Day: Monday, May 8, 2017 Time: 4:30 pm Room: B205, E-Quad Host: Prof. Sergio Verdu Abstract: Real-valued word vectors have transformed NLP applications; popular examples are word2vec and GloVe, recognized for their ability to capture linguistic regularities via simple geometrical operations. In this talk, we demonstrate further striking geometrical properties of the word vectors. First we show that a very simple, and yet counter-intuitive, postprocessing technique, which makes the vectors "more isotropic", renders off-the-shelf vectors even stronger. Second, we show that a sentence containing a target word is well represented by a low rank subspace; subspaces associated with a particular sense of the target word tend to intersect over a line (one-dimensional subspace). We harness this Grassmannian geometry to disambiguate (in an unsupervised way) multiple senses of words, specifically so on the most promiscuously polysemous of all words: prepositions. A surprising finding is that rare senses, including idiomatic/sarcastic/metaphorical usages, are efficiently captured. Our algorithms are all unsupervised and rely on no linguistic resources; we validate them by presenting new state-of-the-art results on a variety of multilingual benchmark datasets. References: 1. Geometry of Compositionality, AAAI '17, https://arxiv.org/abs/1611.09799 2. Geometry of Polysemy, ICLR, '17, https://arxiv.org/abs/1610.07569 3. Prepositions in Context, preprint, https://arxiv.org/abs/1702.01466 Bio: Pramod Viswanath received the Ph.D. degree in electrical engineering and computer science from University of California at Berkeley in 2000. From 2000 to 2001, he was a member of research staff at Flarion Technologies, NJ. Since 2001, he is on the faculty at University of Illinois at Urbana Champaign in Electrical and Computer Engineering, where he currently is a professor. Pramod has worked extensively on information theory and its applications, specifically as applied to wireless communication. His book on wireless communication, coauthored with David Tse, is used in more than 60 institutes around the world. His interest in natural language processing is recent, although he has long been interested in natural languages. Save

1 0

Andy Zeng will present his generals exam Tuesday, May 16, 2017 in CS 302 at 11am.
by Nicki Gotsis 08 May '17

08 May '17

BQ_BEGIN Andy Zeng will present his generals exam Tuesday, May 16, 2017 in CS 302 at 11am. The members of his committee are: Tom Funkhouser (adviser), Szymon Rusinkiewicz, and Elad Hazan. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. Title: Self-Supervised Local 3D Geometric Descriptors and RGB-D Object Segmentation Abstract: In this talk, I will present two projects where we proposed unique methods for labeling massive amounts of data automatically. These methods enabled us to train strong data-driven models for RGB-D object segmentation and local 3D descriptors. In the first project, we present a vision system to recognize objects and their poses under noisy and cluttered environments. More specifically, our approach first segments and labels multiple RGB-D views of a scene with a fully convolutional neural network, and then fits pre-scanned 3D object models to the resulting segmentation to get the 6D object poses. However, training a deep neural network for segmentation typically requires a large amount of training data with manual labels. To alleviate this difficulty, we present an automatic method to generate a large labeled segmentation dataset without tedious manual annotations that could be scaled up to more object categories easily. Our system is reliable under a variety of scenarios including warehouse automation, and is the 3rd place winning system in the world-wide Amazon Robotics Challenge 2016. In the second project, we present a data-driven local 3D volumetric patch descriptor for matching local geometric features on noisy, low-resolution, and incomplete depth images from commodity range sensors. To amass training data for our model, we present an automatic method that leverages the millions of correspondence labels found in existing RGB-D reconstructions. Experiments show that our descriptor is not only able to match local geometry in new scenes for reconstruction, but also generalize to different tasks and spatial scales (e.g. instance-level object model alignment for the Amazon Picking Challenge, and mesh surface correspondence). Results show that 3DMatch consistently outperforms other state-of-the-art approaches by a significant margin. Textbook: Computer Vision: Algorithms and Applications [FPFH] Fast point feature histograms (fpfh) for 3d registration [SIFT] Object Recognition from Local Scale-Invariant Features. [SpinImage] Using spin images for efficient object recognition in cluttered 3d scenes [KinectFusion] KinectFusion: Real-Time Dense Surface Mapping and Tracking [PoseEstimation] Learning 6D Object Pose Estimation using 3D Object Coordinates [MatchNet] MatchNet: Unifying Feature and Metric Learning for Patch-Based Matching [Egomotion] Learning to See by Moving [FCN] Fully Convolutional Networks for Semantic Segmentation [Survey3DDescriptors] Performance Evaluation of 3D Local Feature Descriptors [Grasping] Supersizing Self-supervision: Learning to Grasp from 50K Tries and 700 Robot Hours BQ_END

1 0

Kiran Vodrahalli will present his MSE thesis talk on Wednesday, May 10, 2017 at 1pm in CS 401.
by Nicki Gotsis 05 May '17

05 May '17

Kiran Vodrahalli will present his MSE thesis talk on Wednesday, May 10, 2017 at 1pm in CS 401. The members of his committee are Sanjeev Arora (adviser) and Ken Norman (Princeton Neuroscience Institute). Everyone is invited to attend his talk. His abstract follows below. Several research groups have shown how to correlate fMRI responses to the meanings of presented stimuli. This paper presents new methods for doing so when only a natural language annotation is available as the description of the stimulus. We study fMRI data gathered from subjects watching an episode of BBCs Sherlock, and learn bidirectional mappings between fMRI responses and natural language representations. We show how to leverage data from multiple subjects watching the same movie to improve the accuracy of the mappings, allowing us to succeed at a scene classification task with 72% accuracy (random guessing would give 4%) and at a scene ranking task with average rank in the top 4% (random guessing would give 50%). The key ingredients are (a) the use of the Shared Response Model (SRM) and its variant SRM-ICA to aggregate fMRI data from multiple subjects, both of which are shown to be superior to standard PCA in producing low-dimensional representations for the tasks in this paper; a sentence embedding technique adapted from the natural language processing (NLP) literature [4] that produces semantic vector representation of the annotations; using previous timestep information in the featurization of the predictor data.

1 0

Qipeng Liu will present his general exam on Friday, May 12, 2017 at 1pm in CS 301.
by Nicki Gotsis 05 May '17

05 May '17

Qipeng Liu will present his general exam on Friday, May 12, 2017 at 1pm in CS 301. The members of his committee are: Mark Zhandry (Adviser), Zeev Dvir, and Ran Raz. 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. Program obfuscation — scrambling code to hide embedded secrets — has emerged as a powerful tool that can solve a variety of cryptographic tasks. Unfortunately, there is some evidence that the precise notion of obfuscation used by cryptographers — called indistinguishability obfuscation (iO) — requires either exponentially many different cryptographic assumptions or (sub)exponentially hard assumptions. Indeed, all known ways of building obfuscation suffer one of these two limitations. As such, any application built from iO suffers from these limitations as well. However, for most applications, such limitations do not appear to be inherent to the application, just the approach using iO. Indeed, several recent works have shown how to base applications of iO instead on functional encryption (FE), which can in turn be based on the polynomial hardness of just a few assumptions. However, these constructions are quite complicated and recycle a lot of similar techniques. In this work, we unify the results of previous works in the form of a weakened notion of obfuscation, called Exploding iO. We show (1) how to build exploding iO from functional encryption, and (2) how to build a variety of applications from exploding iO, including all of the applications already known from FE. The construction in (1) hides most of the difficult techniques in the prior work, whereas the constructions in (2) are much closer to the comparatively simple constructions from iO. As such, exploding iO represents a convenient new platform for obtaining more applications from polynomial hardness. reading list: 1. Barak, B., Goldreich, O., Impagliazzo, R., Rudich, S., Sahai, A., Vadhan, S. and Yang, K., 2001, August. On the (im) possibility of obfuscating programs. In Annual International Cryptology Conference (pp. 1-18). Springer Berlin Heidelberg. 2. Garg, S., Gentry, C., Sahai, A. and Waters, B., 2013, June. Witness encryption and its applications. In Proceedings of the forty-fifth annual ACM symposium on Theory of computing (pp. 467-476). ACM. 3. Garg, S., Gentry, C., Halevi, S., Raykova, M., Sahai, A. and Waters, B., 2016. Candidate indistinguishability obfuscation and functional encryption for all circuits. SIAM Journal on Computing, 45(3), pp.882-929. 4. Sahai, A. and Waters, B., 2014, May. How to use indistinguishability obfuscation: deniable encryption, and more. In Proceedings of the 46th Annual ACM Symposium on Theory of Computing (pp. 475-484). ACM. 5. Ananth, P. and Jain, A., 2015, August. Indistinguishability obfuscation from compact functional encryption. In Annual Cryptology Conference (pp. 308-326). Springer Berlin Heidelberg. 6. Bitansky, N. and Vaikuntanathan, V., 2015, October. Indistinguishability obfuscation from functional encryption. In Foundations of Computer Science (FOCS), 2015 IEEE 56th Annual Symposium on (pp. 171-190). IEEE. 7. Garg, S., Pandey, O. and Srinivasan, A., 2016, August. Revisiting the cryptographic hardness of finding a nash equilibrium. In Annual Cryptology Conference (pp. 579-604). Springer Berlin Heidelberg. 8. Garg, S., Pandey, O., Srinivasan, A. and Zhandry, M., 2016. Breaking the Sub-Exponential Barrier in Obfustopia. IACR Cryptology ePrint Archive, 2016, p.102. 9. Garg, S. and Srinivasan, A., 2016, November. Single-key to multi-key functional encryption with polynomial loss. In Theory of Cryptography Conference (pp. 419-442). Springer Berlin Heidelberg. 10. Textbook: Introduction to Modern Cryptography (2nd Edition) by Jonathan Katz and Yehuda Lindell.

1 0