- talks - lists.cs.princeton.edu

Orestis Plevrakis will present his Pre-FPO talk "New Exploration and Identification schemes for Linear Dynamics: the cases of Convex costs and Censored Data" on Wednesday, June 9, 2021 at 2PM via zoom.
by jfarquer＠cs.princeton.edu 04 Jun '21

04 Jun '21

Orestis Plevrakis will present his Pre-FPO talk "New Exploration and Identification schemes for Linear Dynamics: the cases of Convex costs and Censored Data" on Wednesday, June 9, 2021 at 2PM via Zoom. Zoom link: https://princeton.zoom.us/j/99277010909 Committee: Examiners: Sanjeev Arora (advisor), Mark Braverman, and Chi Jin (ECE) Readers: Elad Hazan and Jason D. Lee (ECE) All are welcome to attend. Title: "New Exploration and Identification schemes for Linear Dynamics: the cases of Convex costs and Censored Data" Abstract: Adaptive control, i.e., learning to control an unknown linear dynamical system, has a long history in statistics and control theory. However, sharp non-asymptotic bounds were only recently obtained, and are outputs of an ongoing research program aiming to apply tools from learning theory and optimization, to control theory. A common feature of all the major results of the program has been the learning-part of the proposed algorithms, i.e., the exploration strategy. All these algorithms apply the so-called "certainty-equivalence", where the learner spends the first steps applying random controls (e.g. Gaussian noise), then estimates the dynamics via least-squares, and thereafter optimizes the control-policy using these estimates. Even though this is the simplest exploration strategy, it is optimal for the classical linear quadratic regulator (LQR). In this thesis, I study 1) the problem of adaptive control when the cost can be any convex function (thus generalizing LQR), and 2) the problem of learning a linear dynamical system (LDS) from "censored" observations, i.e., the state is observed only when it falls inside some set (e.g., the state can be the position of an object and the set can be the camera-frame). For the first problem, certainty-equivalence is suboptimal. For the second, the least-squares solution is not even consistent. The contribution of the thesis is the development of new computationally and statistically efficient algorithms for both problems. Specifically, 1. For adaptive control with convex costs, we consider the objective of regret with respect to the benchmark of stabilizing linear control policies. Leveraging ideas from convex geometry, we design an exploration strategy that achieves optimal regret, in terms of the dependence on the time-horizon. Our result improves upon the previous best known bounds, which correspond to algorithms applying certainty-equivalence. 2. In the problem of learning an LDS from censored observations, the learner observes the state $x_t \in \R^d$ if and only if $x_t$ belongs to some set $S_t \subseteq \R^d$. This setting was first considered by Lee and Maddala (1985), and Zeger and Brookmeyer (1986). We develop the first computationally and statistically efficient algorithm for learning the system, assuming only membership-oracle access to the sets $S_t$ (which can be arbitrarily complex sets, e.g., non-convex). Our algorithm, Stochastic Online Newton with Switching Gradients, is a novel second-order method that builds on the Online Newton Step of Hazan et al. (2007).

1 0

Ari Seff will present his FPO "Learning-Aided Design with Structured Generative Modeling" on June 4, 2021 at 2PM via Zoom.
by jfarquer＠cs.princeton.edu 29 May '21

29 May '21

Ari Seff will present his FPO "Learning-Aided Design with Structured Generative Modeling" on June 4, 2021 at 2PM via Zoom. Zoom link: https://princeton.zoom.us/j/96743727531 The members of Ari's committee are as follows: Readers: Ryan Adams(Adviser), Szymon Rusinkiewicz Examiners: Tom Griffiths, Olga Russakovsky, Abigail Doyle (Chemistry) A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu <mailto:gradinfo@cs.princeton.edu> if you would like a copy of the thesis. Everyone is invited to attend his talk. Title: Learning-Aided Design with Structured Generative Modeling Abstract follows below: Humans exhibit a remarkable ability to design new tools and systems for solving problems. At the heart of this capability lies a rich suite of evolved pattern recognition modules, enabling the reuse and modification of previously existing solutions. Machine learning offers the intriguing possibility of automatically capturing the patterns that arise in various design domains, in particular via generative modeling. By leveraging the general tool of distribution approximation over data, we may hope to computationally encode common design patterns, leading to applications that enable more efficient workflows in engineering, design, and even scientific discovery. But this promising route faces challenges. Integral to the above fields are discrete domains, such as molecules and graphs. In contrast to the Euclidean spaces where generative modeling has recently thrived (e.g., images and audio), discrete domains often exhibit irregular structure, strict validity constraints, and non-canonical representations, presenting unique challenges to modeling. In addition, machine learning as applied to engineering still lacks maturity, and as a result there is a need for curated datasets, benchmarks, and shared pipelines. First, we discuss a novel generative model for discrete domains, known as reversible inductive construction. Building off of generative interpretations of denoising autoencoders, the model employs a Markov chain where transitions are restricted to a set of local operations that preserve validity. This approach constrains the generative model to only produce valid objects, requires the learner to only discover local modifications to the objects, and avoids marginalization over an unknown and potentially large space of construction histories. Next, we present SketchGraphs, a dataset and processing pipeline that contains 15 million real-world CAD sketches paired with their ground truth geometric constraint graphs. Unlike previously available CAD data, explicit supervision is provided regarding the designer-imposed geometric relationships between primitives, making the typically latent construction operations directly accessible. We demonstrate and establish benchmarks for two use cases of the dataset: generative modeling of sketches and conditional generation of likely constraints given unconstrained geometry. Lastly, we propose a method for CAD sketch inference from images, leveraging a conditional generative model that consumes a raster image of a hand-drawn sketch infers the depicted primitives and constraints. Evaluation of this model on SketchGraphs demonstrates its potential to enable new design workflows.

1 0

Anunay Kulshrestha will present his General Exam "Measuring Incidental Collection in Foreign Intelligence Surveillance" on Monday, May 31, 2021 at 10AM via zoom.
by jfarquer＠cs.princeton.edu 24 May '21

24 May '21

Anunay Kulshrestha will present his General Exam "Measuring Incidental Collection in Foreign Intelligence Surveillance" on Monday, May 31, 2021 at 10AM via zoom. Title: Measuring Incidental Collection in Foreign Intelligence Surveillance Zoom link: https://princeton.zoom.us/j/93184542846 <https://princeton.zoom.us/j/92520464322> The members of his committee are as follows: Jonathan Mayer (adviser), Ed Felten, Seny Kamara (Associate Professor of Computer Science at Brown University). Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so. Link to Abstract: https://docs.google.com/document/d/1FZJkA2ypjZ7vd_qcP5Gshrt9aTWBJiJk4KFqef1g pIo/edit <https://docs.google.com/document/d/1FZJkA2ypjZ7vd_qcP5Gshrt9aTWBJiJk4KFqef1 gpIo/edit> Link to Reading List: https://docs.google.com/document/d/1v08sEz7_5KW0Vs_jMvdas-C0GGU2NaTsaixYyJB6 KXw/edit <https://docs.google.com/document/d/1v08sEz7_5KW0Vs_jMvdas-C0GGU2NaTsaixYyJB 6KXw/edit>

1 0

Gregory Gundersen will present his FPO "Practical Algorithms for Latent Variable Models" on Friday, May 28, 2021 at 2:30PM via Zoom
by jfarquer＠cs.princeton.edu 21 May '21

21 May '21

Gregory Gundersen will present his FPO "Practical Algorithms for Latent Variable Models" on Friday, May 28, 2021 at 2:30PM via Zoom Zoom link: https://princeton.zoom.us/j/92503349718 The members of Greg's committee are as follows: Barbara Engelhardt (Adviser), Readers: Tom Griffiths and Brandon Stewart, Examiners: Barbara Engelhardt, Ryan Adams, Jonathan Pillow A copy of his thesis is available upon request. Please email jfarquer(a)cs.princeton <mailto:jfarquer@cs.princeton> if you would like a copy of the thesis. Everyone is invited to attend the talk. Abstract: Latent variables allow researchers and engineers to encode assumptions into their sta[1]tistical models. A latent variable might, for example, represent an unobserved covariate, measurement error, or a missing class label. Inference is challenging because one must account for the conditional dependence structure induced by these variables, and marginal[1]ization is often intractable. In this thesis, I present several practical algorithms for inferring latent structure in probabilistic models used in computational biology, neuroscience, and time-series analysis. First, I present a multi-view framework that combines neural networks and probabilistic canonical correlation analysis to estimate shared and view-specific latent structure of paired samples of histological images and gene expression levels. The model is trained end-to-end to estimate all parameters simultaneously, and we show that the latent variables capture interpretable structure, such as tissue-specific and morphological variation. Next, I present a family of nonlinear dimension-reduction models that use random features to support non[1]Gaussian data likelihoods. By approximating a nonlinear relationship between the latent variables and observations with a function that is linear with respect to random features, we induce closed-form gradients of the posterior distribution with respect to the latent variables. This allows for gradient-based nonlinear dimension-reduction models for a variety of data likelihoods. Finally, I discuss lowering the computational cost of online Bayesian filtering of time series with abrupt changes in structure, called changepoints. We consider settings in which a time series has multiple data sources, each with an associated cost. We trade the cost of a data source against the quality or fidelity of that source and how its fidelity affects the estimation of changepoints. Our framework makes cost-sensitive decisions about which data source to use based on minimizing the information entropy of the posterior distribution over changepoints.

1 0

Fermi Ma will present his Pre-FPO talk "Secure Computation in a Quantum World" on Friday, May 28, 2021 at 1PM via Zoom.
by jfarquer＠cs.princeton.edu 21 May '21

21 May '21

Fermi Ma will present his Pre-FPO talk "Secure Computation in a Quantum World" on Friday, May 28, 2021 at 1PM via Zoom. Zoom Link: https://princeton.zoom.us/j/99360445283 Fermi's committee is as follows: Examiners: Mark Zhandry (advisor), Sanjeev Arora, and Gillat Kol Readers: Ran Raz and Zeev Dvir All are welcome to attend. Title: Secure Computation in a Quantum World Abstract: This talk will cover new results on two central cryptographic tools - succinct arguments and multi-party computation - and novel aspects that arise given the emergence of quantum computers. (1) Quantum-Secure Succinct Arguments. A succinct argument enables an untrusted (but computationally bounded) prover to convince a verifier of any NP statement with total communication independent of the witness length. The security of all known succinct argument systems has only been established against classical attackers, as security relies on the ubiquitous "rewinding technique" from classical cryptography, which is not applicable to quantum adversaries. The lack of a general-purpose quantum rewinding technique has been a long-standing source of frustration in developing quantum-secure cryptosystems. In the first part of this talk, I will present the first general-purpose technique for rewinding a quantum attacker. This technique allows us to transfer classical rewinding-based proofs to the quantum setting, and proves that succinct arguments secure against quantum attackers exist under standard cryptographic assumptions. (2) Multi-Party Quantum Computation from One-Way Functions. Another central primitive in classical cryptography is multi-party computation, which enables a group of mutually distrusting parties to compute a shared function over their inputs while revealing no other information. In the second part of this talk, I will describe new results showing that when all parties are quantum, secure multi-party computation can be based solely on the existence of the minimal cryptographic primitive: one-way functions. This is in stark contrast to the classical setting where no such implication is known.

1 0

ECE and CS Seminar - Song Han, MIT: TODAY at 10:30am
by Emily C. Lawrence 19 May '21

19 May '21

CS & ECE SEMINAR Speaker: Song Han, MIT Title: Putting AI on a Diet: TinyML and Efficient Deep Learning Day: Wednesday, May 19, 2021 Time: 10:30 am Zoom registration: [ https://princeton.zoom.us/meeting/register/tJ0tcu2tpzkqHtP8kEDbsDJLGM25atnx… | https://princeton.zoom.us/meeting/register/tJ0tcu2tpzkqHtP8kEDbsDJLGM25atnx… ] Hosts: Kai Li and Naveen Verma Abstract: Today’s AI is too big. Deep neural networks demand extraordinary levels of data and computation, and therefore power, for training and inference. This severely limits the practical deployment of AI in edge devices. We aim to improve the efficiency of neural network design. First, I’ll present MCUNet [1] that brings deep learning to IoT devices. MCUNet is a framework that jointly designs the efficient neural architecture (TinyNAS) and the light-weight inference engine (TinyEngine), enabling ImageNet-scale inference on micro-controllers that have only 1MB of Flash. Next I will introduce Once-for-All Network [2], an efficient neural architecture search approach, that can elastically grow and shrink the model capacity according to the target hardware resource and latency constraints. From inference to training, I’ll present TinyTL [3] that enables tiny transfer learning on-device, reducing the memory footprint by 7-13x. Finally, I will describe data-efficient GAN training techniques [4] that can generate photo-realistic images using only 100 images, which used to require tens of thousands of images. We hope such TinyML techniques can make AI greener, faster, more efficient and more sustainable. [1] MCUNet: Tiny Deep Learning on IoT Devices, (NeurIPS’20 spotlight) [2] Once-for-All: Train One Network and Specialize it for Efficient Deployment (ICLR’19) [3] Tiny Transfer Learning: Reduce Memory, not Parameters for Efficient On-Device Learning (NeurIPS’20) [4] Differentiable Augmentation for Data-Efficient GAN Training (NeurIPS’20) Bio: Song Han is an assistant professor at MIT’s EECS. He received his PhD degree from Stanford University. His research focuses on efficient deep learning computing. He proposed “deep compression” technique that can reduce neural network size by an order of magnitude without losing accuracy, and the hardware implementation “efficient inference engine” that first exploited pruning and weight sparsity in deep learning accelerators. His team’s work on hardware-aware neural architecture search that bring deep learning to IoT devices was highlighted by MIT News, Wired, Qualcomm News, VentureBeat, IEEE Spectrum, integrated in PyTorch and AutoGluon, and received many low-power computer vision contest awards in flagship AI conferences (CVPR’19, ICCV’19 and NeurIPS’19). Song received Best Paper awards at ICLR’16 and FPGA’17, Amazon Machine Learning Research Award, SONY Faculty Award, Facebook Faculty Award, NVIDIA Academic Partnership Award. Song was named “35 Innovators Under 35” by MIT Technology Review for his contribution on “deep compression” technique that “lets powerful artificial intelligence (AI) programs run more efficiently on low-power mobile devices.” Song received the NSF CAREER Award for “efficient algorithms and hardware for accelerated machine learning” and the IEEE “AIs 10 to Watch: The Future of AI” award.

1 0

Colloquium Speaker Umesh Vazirani, TODAY at 12:30pm
by Emily C. Lawrence 19 May '21

19 May '21

Colloquium Speaker Speaker: Umesh Vazirani, University of California at Berkeley Title: Theoretical Reflections on Quantum Supremacy Date: Wednesday, May 19 Time: 12:30pm ET Hosts: Sanjeev Arora & Andrew Houck Event page: [ https://www.cs.princeton.edu/events/26082 | https://www.cs.princeton.edu/events/26082 ] Register for Zoom webinar here: [ https://princeton.zoom.us/webinar/register/WN_6zJv9owaS3WOuss4Gg7_7A | https://princeton.zoom.us/webinar/register/WN_6zJv9owaS3WOuss4Gg7_7A ] Abstract: The recent demonstration of quantum supremacy by Google is a first step towards the era of small to medium scale quantum computers. In this talk I will explain what the experiment accomplished and the theoretical work it is based on, as well as what it did not accomplish and the many theoretical and practical challenges that remain. I will also describe recent breakthroughs in the design of protocols for the testing and benchmarking of quantum computers, a task that has deep computational and philosophical implications. Specifically, this leads to protocols for scalable and verifiable quantum supremacy, certifiable quantum random generation and verification of quantum computation. Bio: Umesh Vazirani is Strauch Distinguished Professor of Computer Science at UC Berkeley, and Director of the Berkeley Quantum Computing Center (BQIC). His 1993 paper with Ethan Bernstein laid the foundations of quantum complexity theory, and his research has touched on many facets of quantum computation, including Quantum Algorithms, Quantum Hamiltonian Complexity and Interactive classical testing of quantum devices. Vazirani is co-inventor of the Bid Scaling algorithm for the AdWords auction which is widely used by Internet search companies, and co-winner of the Fulkerson Prize for the ARV graph partitioning algorithm. He is member of the NAS, and co-author of two books: An Introduction to Computational Learning Theory (MIT Press) and Algorithms (McGraw-Hill).

1 0

Reminder: Colloquium Speaker Umesh Vazirani, tomorrow at 12:30pm
by Emily C. Lawrence 18 May '21

18 May '21

Colloquium Speaker Speaker: Umesh Vazirani, University of California at Berkeley Title: Theoretical Reflections on Quantum Supremacy Date: Wednesday, May 19 Time: 12:30pm ET Hosts: Sanjeev Arora & Andrew Houck Event page: [ https://www.cs.princeton.edu/events/26082 | https://www.cs.princeton.edu/events/26082 ] Register for Zoom webinar here: [ https://princeton.zoom.us/webinar/register/WN_6zJv9owaS3WOuss4Gg7_7A | https://princeton.zoom.us/webinar/register/WN_6zJv9owaS3WOuss4Gg7_7A ] Abstract: The recent demonstration of quantum supremacy by Google is a first step towards the era of small to medium scale quantum computers. In this talk I will explain what the experiment accomplished and the theoretical work it is based on, as well as what it did not accomplish and the many theoretical and practical challenges that remain. I will also describe recent breakthroughs in the design of protocols for the testing and benchmarking of quantum computers, a task that has deep computational and philosophical implications. Specifically, this leads to protocols for scalable and verifiable quantum supremacy, certifiable quantum random generation and verification of quantum computation. Bio: Umesh Vazirani is Strauch Distinguished Professor of Computer Science at UC Berkeley, and Director of the Berkeley Quantum Computing Center (BQIC). His 1993 paper with Ethan Bernstein laid the foundations of quantum complexity theory, and his research has touched on many facets of quantum computation, including Quantum Algorithms, Quantum Hamiltonian Complexity and Interactive classical testing of quantum devices. Vazirani is co-inventor of the Bid Scaling algorithm for the AdWords auction which is widely used by Internet search companies, and co-winner of the Fulkerson Prize for the ARV graph partitioning algorithm. He is member of the NAS, and co-author of two books: An Introduction to Computational Learning Theory (MIT Press) and Algorithms (McGraw-Hill).

1 0

Reminder: ECE and CS Seminar - Song Han, MIT: tomorrow at 10:30am
by Emily C. Lawrence 18 May '21

18 May '21

CS & ECE SEMINAR Speaker: Song Han, MIT Title: Putting AI on a Diet: TinyML and Efficient Deep Learning Day: Wednesday, May 19, 2021 Time: 10:30 am Zoom registration: [ https://princeton.zoom.us/meeting/register/tJ0tcu2tpzkqHtP8kEDbsDJLGM25atnx… | https://princeton.zoom.us/meeting/register/tJ0tcu2tpzkqHtP8kEDbsDJLGM25atnx… ] Hosts: Kai Li and Naveen Verma Abstract: Today’s AI is too big. Deep neural networks demand extraordinary levels of data and computation, and therefore power, for training and inference. This severely limits the practical deployment of AI in edge devices. We aim to improve the efficiency of neural network design. First, I’ll present MCUNet [1] that brings deep learning to IoT devices. MCUNet is a framework that jointly designs the efficient neural architecture (TinyNAS) and the light-weight inference engine (TinyEngine), enabling ImageNet-scale inference on micro-controllers that have only 1MB of Flash. Next I will introduce Once-for-All Network [2], an efficient neural architecture search approach, that can elastically grow and shrink the model capacity according to the target hardware resource and latency constraints. From inference to training, I’ll present TinyTL [3] that enables tiny transfer learning on-device, reducing the memory footprint by 7-13x. Finally, I will describe data-efficient GAN training techniques [4] that can generate photo-realistic images using only 100 images, which used to require tens of thousands of images. We hope such TinyML techniques can make AI greener, faster, more efficient and more sustainable. [1] MCUNet: Tiny Deep Learning on IoT Devices, (NeurIPS’20 spotlight) [2] Once-for-All: Train One Network and Specialize it for Efficient Deployment (ICLR’19) [3] Tiny Transfer Learning: Reduce Memory, not Parameters for Efficient On-Device Learning (NeurIPS’20) [4] Differentiable Augmentation for Data-Efficient GAN Training (NeurIPS’20) Bio: Song Han is an assistant professor at MIT’s EECS. He received his PhD degree from Stanford University. His research focuses on efficient deep learning computing. He proposed “deep compression” technique that can reduce neural network size by an order of magnitude without losing accuracy, and the hardware implementation “efficient inference engine” that first exploited pruning and weight sparsity in deep learning accelerators. His team’s work on hardware-aware neural architecture search that bring deep learning to IoT devices was highlighted by MIT News, Wired, Qualcomm News, VentureBeat, IEEE Spectrum, integrated in PyTorch and AutoGluon, and received many low-power computer vision contest awards in flagship AI conferences (CVPR’19, ICCV’19 and NeurIPS’19). Song received Best Paper awards at ICLR’16 and FPGA’17, Amazon Machine Learning Research Award, SONY Faculty Award, Facebook Faculty Award, NVIDIA Academic Partnership Award. Song was named “35 Innovators Under 35” by MIT Technology Review for his contribution on “deep compression” technique that “lets powerful artificial intelligence (AI) programs run more efficiently on low-power mobile devices.” Song received the NSF CAREER Award for “efficient algorithms and hardware for accelerated machine learning” and the IEEE “AIs 10 to Watch: The Future of AI” award.

1 0

Chaitanya Aluru will present his Pre FPO "Reconciliation based methods for Understanding Transcription Factor Evolution" on Thursday, May 27, 2021 at 3PM via Zoom.
by jfarquer＠cs.princeton.edu 17 May '21

17 May '21

Chaitanya Aluru will present his Pre FPO "Reconciliation based methods for Understanding Transcription Factor Evolution" on Thursday, May 27, 2021 at 3PM via Zoom. Zoom link: https://princeton.zoom.us/j/91687324537 Committee: Examiners: Mona Singh, Yuri Prytikin, Barbara Englehardt Readers: Bernard Chazelle, Ben Raphael All are welcome to attend. Title: Reconciliation based methods for Understanding Transcription Factor Evolution Abstract: Regulatory networks are thought to underlie the biological differences between organisms. However, the evolutionary mechanisms by which these networks evolve is still an active area of research. Proteins called transcription factors form a central component of regulatory networks and are thought to evolve through repeated duplications and losses of functional subunits known as domains. Previous work has captured single duplication events, but fails to capture tandem duplications of multiple domains, a key evolutionary event in certain transcription factor families. In this talk, I will discuss two parsimony frameworks I have designed to incorporate tandem duplications. I will show how to write these problems as constraint satisfaction problems and find optimal solutions using ILPs. Finally, I will show example protein families in which our methods have successfully uncovered ancestral duplication patterns, and discuss future large-scale studies of zinc finger genes, the largest family of transcription factors in humans.

1 0