April 2022 - talks - lists.cs.princeton.edu

Princeton Robotics Seminar: Mac Schwager, TODAY at 3pm
by Emily C. Lawrence 22 Apr '22

22 Apr '22

***** Princeton Robotics Seminar: Mac Schwager, TODAY at 3pm ***** Time: 3pm EST, Friday Apr 22, 2022 Location: Zoom [ https://princeton.zoom.us/my/robotics | link ] Speaker: [ https://web.stanford.edu/~schwager/ | Mac Schwager ] , Stanford AeroAstro Title: Reimagining Robot Autonomy with Neural Environment Representations Abstract: New developments in computer vision and deep learning have led to the rise of neural environment representations: 3D maps that are stored as deep networks that spatially register occupancy, color, texture, and other physical properties. These environment models can generate photo-realistic synthetic images from unseen view points, and can store 3D information in exquisite detail. In this talk, I investigate the question: How can robots use neural environment representations for perception, motion planning, manipulation, and simulation? I will show recent work from my lab in which we build a robot navigation pipeline using a Neural Radiance Field (NeRF) map of an environment. We develop a trajectory optimization algorithm that interfaces with the NeRF model to find dynamically feasible, collision-free trajectories for a robot moving through a NeRF world. We also develop an optimization-based state estimator that uses the NeRF model to give full dynamic state estimates for a robot from only on board images. I will also discuss preliminary results using NeRF models for grasp planning, and for tracking the poses of multiple 3D objects in a scene. Finally, I will discuss a new differentiable robot physics simulator called Dojo that can use NeRFs as a geometry description for objects, leading to physically interpretable motion prediction from NeRF models. I will conclude with future opportunities and challenges in integrating neural environment representations into the robot autonomy stack. Bio: Mac Schwager is an Associate Professor of Aeronautics and Astronautics at Stanford University. He directs the Multi-robot Systems Lab (MSL) where he studies distributed algorithms for control, perception, and learning in groups of robots and autonomous systems. He is interested in a range of applications including cooperative surveillance with teams of UAVs, autonomous driving in traffic, cooperative robotic manipulation, distributed SLAM, distributed trajectory planning, and autonomous drone and car racing. He obtained his BS degree from Stanford, and his MS and PhD degrees from MIT. He was a postdoctoral researcher at the University of Pennsylvania and at MIT. Prior to joining Stanford, he was an assistant professor at Boston University from 2012 to 2015. He received the NSF CAREER award in 2014, the DARPA YFA in 2018, and has received numerous best paper awards in conferences and journals including the IEEE Transactions on Robotics best paper award in 2016, the Best Paper Award in Robot Manipulation in ICRA 2018, and the Best Paper Award in Multi-Robot Systems in ICRA 2020.

1 0

Nikunj Saunshi will present his Pre FPO "Towards understanding self-supervised representation learning" on Monday, April 25, 2022 at 1:30pm via Zoom.
by Nicki Mahler 21 Apr '22

21 Apr '22

Nikunj Saunshi will present his Pre FPO "Towards understanding self-supervised representation learning" on Monday, April 25, 2022 at 1:30pm via Zoom. Zoom link: https://princeton.zoom.us/j/93697691837 His committee is as follows: Prof. Sanjeev Arora (adviser), Prof. Danqi Chen (examiner), Prof. Chi Jin (examiner), Prof. Elad Hazan (reader), Prof. Jason Lee (reader). Title: Towards understanding self-supervised representation learning Abstract: While supervised learning sparked the deep learning boom, it has some critical pitfalls: (1) it requires an abundance of expensive labeled data, and (2) it solves tasks from scratch rather than the human-like approach of leveraging knowledge and skills acquired from prior experiences. Pre-training has emerged as an alternative and effective paradigm, where a model is first trained using easily acquirable data, and later used to solve downstream tasks of interest with much fewer labeled data than supervised learning. Pre-training using unlabeled data, a.k.a. self-supervised learning, has been especially revolutionary, with successes in diverse domains (text, vision, speech, etc.). This raises an interesting and challenging question: why should pre-training on unlabeled data help with seemingly unrelated downstream tasks? In this talk I will present my works that initiate and build theoretical frameworks to study self-supervised learning methods like contrastive learning, language modeling and self-prediction based methods. Central to the framework is the idea that pre-training helps learn low-dimensional representations of data that help solve downstream tasks of interest with linear classifiers, thus requiring little labeled data. A common theme is to mathematically show how appropriate pre-training objectives can extract the downstream signal that is implicitly encoded in the unlabeled data distribution, and how this signal can be decoded from the learned representations using linear classifiers, thus providing a formalization for transference of “skills and knowledge” across tasks. All are welcome to attend.

1 0

Mona Wang will present her General Exam on April 27, 2022 at 3pm via Zoom.
by Nicki Mahler 19 Apr '22

19 Apr '22

Mona Wang will present her General Exam on April 27, 2022 at 3pm via Zoom. Zoom link: [ https://princeton.zoom.us/my/monamon?pwd=UDhNd3RQVmlISGZXeDhBSCsyVU9iUT09 | https://princeton.zoom.us/my/monamon?pwd=UDhNd3RQVmlISGZXeDhBSCsyVU9iUT09 ] The members of her committee are as follows: Jonathan Mayer (adviser), Prateek Mittal, and Roya Ensafi Abstract: [ https://docs.google.com/document/d/1h22DDjrwMzaGZs477DBm3ho3ayT7Yx04x68SMKj… | https://docs.google.com/document/d/1h22DDjrwMzaGZs477DBm3ho3ayT7Yx04x68SMKj… ] Censorship and surveillance events continue to increase at an unprecedented scale, enabled by the proliferation of censorship-capable network devices that can quickly inspect and interfere with large amounts of network traffic. Advancements in networking device technology and the commoditization of deep packet inspection (DPI) devices for the purpose of enterprise network monitoring have made this censorship capability available to a growing number of countries, including those with poor human rights records. While these censorship technologies have proliferated, techniques to identify and monitor them are still relatively limited. In particular, many of these techniques focus on what is blocked, and discovering how the blocking occurs, but there are very few large-scale surveys on the censorship network devices themselves, and their vendors. This work aims to develop techniques to identify and monitor the deployment of devices that perform network censorship at Internet-scale. We perform large-scale probing of network censorship-capable middleboxes to develop accurate network-level fingerprints of censorship middleboxes, and discover how many network devices on the Internet are built on devices capable of network censorship. Reading List: [ https://docs.google.com/document/d/1_MG8cxDLGJ8XP88M2kFe0C456LFM7DnaOHrVXyp… | https://docs.google.com/document/d/1_MG8cxDLGJ8XP88M2kFe0C456LFM7DnaOHrVXyp… ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Zhou Lu will present his General Exam on Tuesday, April 26, 2022 at 11am via Zoom and in CS 301
by Nicki Mahler 19 Apr '22

19 Apr '22

Zhou Lu will present his General Exam on Tuesday, April 26, 2022 at 11am via Zoom and in CS 301. Zoom link: [ https://princeton.zoom.us/j/8104170458 | https://princeton.zoom.us/j/8104170458 ] The members of his committee are as follows: Elad Hazan (adviser), Sanjeev Arora, and Jason Lee Abstract: [ https://docs.google.com/document/d/1iu2XuYnC48Lk9_qdl3sDDOYTGLrcRP9GMk2FQEX… | https://docs.google.com/document/d/1iu2XuYnC48Lk9_qdl3sDDOYTGLrcRP9GMk2FQEX… ] Adaptive gradient methods are the most popular method for optimization in machine learning, but existing bounds didn't take (possible) changing environments into consideration. We study the problem of learning a local preconditioner, that can change as the data is changing along the optimization trajectory. We propose an adaptive gradient method that has provable adaptive regret guarantees vs. the best local preconditioner. To derive this guarantee, we prove a new adaptive regret bound in online learning that improves upon previous adaptive online learning methods. We demonstrate the robustness of our method in automatically choosing the optimal learning rate schedule for popular benchmarking tasks in vision and language domains. Without the need to manually tune a learning rate schedule, our method can, in a single run, achieve comparable and stable task accuracy as a fine-tuned optimizer. Reading List: [ https://sites.google.com/view/intro-oco/ver-2-0?authuser=0 | https://sites.google.com/view/intro-oco/ver-2-0?authuser=0 ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Howard Chen will present his General Exam "Can Rationalization Improve Robustness?" on Wednesday, April 20, 2022 at 3:00 PM via Zoom.
by Louis W. Riehl 19 Apr '22

19 Apr '22

Howard Chen will present his General Exam "Can Rationalization Improve Robustness?" on Wednesday, April 20, 2022 at 3:00 PM via Zoom. Zoom link: https://princeton.zoom.us/j/94177862135 Committee Members: Danqi Chen (advisor), Karthik Narasimhan (advisor), Jia Deng Abstract: A growing line of work has investigated the development of neural NLP models that can produce rationales--subsets of input that can explain their model predictions. In this work, we ask whether such rationale models can also provide robustness to adversarial attacks in addition to their interpretable nature. Since these models need to first generate rationales (``rationalizer'') before making predictions (``predictor''), they have the potential to ignore noise or adversarially added text by simply masking it out of the generated rationale. To this end, we systematically generate various types of `AddText' attacks for both token and sentence-level rationalization tasks, and perform an extensive empirical evaluation of state-of-the-art rationale models across five different tasks. Our experiments reveal that the rationale models show the promise to improve robustness, while they struggle in certain scenarios--when the rationalizer is sensitive to positional bias or lexical choices of attack text. Further, leveraging human rationale as supervision does not always translate to better performance. Our study is a first step towards exploring the interplay between interpretability and robustness in the rationalize-then-predict framework. Reading List: https://docs.google.com/document/d/15UBbTlvJSIz3Go7ZOx6A_y_QnD8aujdEpDK82m4… Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so. Louis Riehl Graduate Administrator Computer Science Department, CS213 Princeton University (609) 258-8014

2 1

Qinshi Wang will present his Pre FPO "Formally verifiable data plane programming" on Wednesday, April 20, 2022 in CS 401 at 4pm
by Nicki Mahler 19 Apr '22

19 Apr '22

Qinshi Wang will present his Pre FPO "Formally verifiable data plane programming" on Wednesday, April 20, 2022 in CS 401 at 4pm His committee is as follows: Andrew Appel (advisor) David Walker (reader) Nate Foster, Cornell (reader) Jen Rexford (examiner) Zachary Kincaid (examiner) All are welcome to attend. Title: Formally verifiable data plane programming Older-generation network devices were not end-user programmable. Software-defined network has been successful in separating the control plane and the data plane and providing more programmability for the control plane. But the data plane still lacks programmability. In recent years, programmable data plane devices and the P4 programming language, which programs these devices, have been developed. Because of P4's special language features for high-speed packet processing, and because extra consideration is needed to write P4 programs that fit into limited hardware resources, programming in P4 is hard and error-prone. Tools have been developed to check safety properties, but those are far from guaranteeing correctness. Meanwhile, P4 compilers and hardware may also have bugs. And when the network fails, it is hard to identify the cause between the P4 program, the compiler, and the hardware. We suggest that the solution is to adopt formal verification to P4 to make all the tools in the toolchain provably correct. Toward this goal, we first define an intermediate representation called P4light, which can encode the full P4 language after processing by the front end, and define the abstract syntax tree (AST) of P4light. We then define the mechanized operational semantics for P4light in the Coq proof assistant, according to the P4 specification and reference implementations. In the semantics, we establish a phase distinction between the instantiation phase (static resource allocation) and the execution phase. This phase distinction clarifies the explanation of the language, simplifies the operational semantics, and avoids tracking complicated objects when analyzing programs' behavior. Still, proving programs correct directly with respect to the semantics is too complicated to be tractable. So we build a verifier that helps users prove correctness of P4 programs. The verifier consists of an expressive program logic proved sound with respect to the semantics and a proof automation system that applies the program logic rules to prove programs correct. The verifier is powerful enough to prove the behavior of a Bloom filter on a sequence of packets with moderate human effort.

1 1

SEAS talk: Databricks: An entrepreneur’s journey from Princeton to founder TODAY at 4:30pm
by Emily C. Lawrence 19 Apr '22

19 Apr '22

SEAS talk Speakers: Patrick Wendell, Databricks & Jen Rexford, Princeton University Date: Tuesday, April 19, 2022 Time: 4:30pm EST Location: Webinar Event page: [ https://engineering.princeton.edu/events/databricks-entrepreneurs-journey-p… | https://engineering.princeton.edu/events/databricks-entrepreneurs-journey-p… ] Patrick Wendell ’11 , co-founder and vice president of engineering at Databricks, will join Jen Rexford ’91 , chair of Princeton’s Department of Computer Science, in a conversation about his entrepreneurial journey from computer science major at Princeton to co-founder of a pioneering company used by businesses around the world to manage and analyze their data. Wendell is a co-founder of Databricks, a machine learning and data infrastructure company recently valued at $38 billion. At Databricks Patrick manages several engineering efforts in the areas of machine learning and data science. He holds an M.S. in Computer Science at UC Berkeley, where his research focused on low latency scheduling for large scale analytics workloads. He holds a B.S.E in Computer Science from Princeton University. Rexford is the Gordon Y.S. Wu Professor in Engineering and chair and professor of Computer Science at Princeton. She received her Ph.D. from the University of Michigan and earned her B.S.E. in electrical engineering from Princeton in 1991. She worked at AT&T Labs before joining the Princeton faculty in 2005. Rexford is a leader in computer networking, focusing on methods to improve and expand digital communications. This event will take place at 4:30 pm Eastern Time / 1:30 pm Pacific Time on Tuesday, April 19. No advanced registration required.

1 0

Christina Shatford will present her MSE talk "Simultaneous Pattern and Detector Optimization for Visual Tags" on Thursday, April 21, 2022 at 3:00pm via Zoom
by Nicki Mahler 18 Apr '22

18 Apr '22

Christina Shatford will present her MSE talk "Simultaneous Pattern and Detector Optimization for Visual Tags" on Thursday, April 21, 2022 at 3:00pm via Zoom. Zoom link: [ https://princeton.zoom.us/j/91816794623 | https://princeton.zoom.us/j/91816794623 ] Committee members: Rusinkiewicz (Adviser) and Ryan Adams (Reader) Abstract: Accurate and robust object localization has been a long studied problem in computer vision. In this talk, we will propose a new approach for localizing visual tags in images. Instead of optimizing a detector to locate a tag with a static pattern, this approach involves the simultaneous optimization of both the detector and the pattern of the visual tag. We will show how this joint learning leads to increased localization accuracy when compared to only training the detector and using a selection of standard non-optimized tags, such as QR codes. All are welcome to attend.

1 0

Huihan Li will present her MSE Talk "Re-evaluating Conversational Question Answering" on Thursday, April 21st, 2022 at 3:30 PM via Zoom.
by Louis W. Riehl 14 Apr '22

14 Apr '22

Huihan Li will present her MSE Talk "Re-evaluating Conversational Question Answering" on Thursday, April 21st, 2022 at 3:30 PM via Zoom. Zoom Link: [ https://princeton.zoom.us/j/92763452837 | https://princeton.zoom.us/j/92763452837 ] Committee Members: Danqi Chen (advisor), Karthik Narasimhan (reader) All are welcome to attend. Re-evaluating Conversational Question Answering Conversational question answering aims to provide natural-language answers to users in information-seeking conversations. Existing conversational QA benchmarks compare models with pre-collected human-human conversations, using ground-truth answers provided in conversational history. It remains unclear whether we can rely on this static evaluation for model development and whether current systems can well generalize to real-world human-machine conversations. In this work, we conduct the first large-scale human evaluation of state-of-the-art conversational QA systems, where human evaluators converse with models and judge the correctness of their answers. We find that the distribution of human-machine conversations differs drastically from that of human-human conversations, and there is a disagreement between human and gold-history evaluation in terms of model ranking. We further investigate how to improve automatic evaluations, and propose a question rewriting mechanism based on predicted history, which better correlates with human judgments. Finally, we demonstrate that training conversational question answering systems using their own predictions benefits the models under the new automatic evaluation protocol. Louis Riehl Graduate Administrator Computer Science Department, CS213 Princeton University (609) 258-8014

1 0

Juan Duque will present his MSE talk, "Transformer Representations for Efficient Reinforcement Learning" on Thursday, April 21st at 3:00 PM (EDT) via Zoom.
by Louis W. Riehl 14 Apr '22

14 Apr '22

Juan Duque will present his MSE talk, "Transformer Representations for Efficient Reinforcement Learning" on Thursday, April 21st at 3:00 PM (EDT) via Zoom. Zoom Link: [ https://princeton.zoom.us/j/93980023607 | https://princeton.zoom.us/j/93980023607 ] Committee Members: Karthik Narasimhan (adviser) and Elad Hazan (reader) All are welcome to attend. Transformer Representations for Efficient Reinforcement Learning Reinforcement Learning algorithms are often trained from scratch at high computational costs. In this talk I will present a self-supervised learning framework that uses the Transformer architecture to generate useful representations for Reinforcement Learning. Our formulation uses objectives borrowed from Natural Language Processing and Imitation Learning to pre-train a Tranformer model from offline trajectories that is then fine-tuned on Atari games. We demonstrate how this setup allows for quick online convergence in under 100 thousand episodes. Louis Riehl Graduate Administrator Computer Science Department, CS213 Princeton University (609) 258-8014

1 0