- talks - lists.cs.princeton.edu

Yushan Su will present her FPO "Making Neural Network Models More Efficient" on Wednesday, May 10, 2023 at 10am in Friend 202.
by Nicki Mahler 13 Apr '23

13 Apr '23

Yushan Su will present her FPO " Making Neural Network Models More Efficient" on Wednesday, May 10, 2023 at 10am in Friend 202. The members of her committee are as follows: Examiners: Kai Li (adviser), Wyatt Lloyd, and Karthik Narasimhan; Readers: Ravi Netravali and Olga Troyanskaya. Please see title and abstract below. All are welcome to attend. Complex machine learning tasks typically require large neural network models. However, training and inference on neural models require substantial compute power and large memory footprints, and incur significant cost. My thesis studies methods to make neural networks efficient at a relatively low cost. First, we explore how to utilize CPU servers for training and inference. CPU servers are more readily available, have larger memories, and cost less than GPUs or hardware accelerators. However, they are much less efficient for training and inference tasks. My thesis studies how to design efficient software kernels for sparse neural networks that allow unstructured pruning to achieve efficiency of training or inference with minimum accuracy losses. Our evaluation shows that our sparse kernels can achieve 6.4x-20.0x speedups at different sparsities over the commonly used Intel MKL sparse library. Second, we study how to achieve high-throughput inference for large models. We explore ways to combine data multiplexing with pruning. We find that combining the two approaches can achieve better throughput than using each approach alone for a given accuracy loss. We then propose an automatic method to find parameters for such combinations to maximize inference throughput given an accuracy loss budget. We show that the proposed combining can improve Transformer models throughput by 7.5-29.5X at different accuracy thresholds, which is higher than using either approach alone.

1 2

From Big Data To Equitable Health: Princeton Precision Health Spring Workshop May 2
by Emily C. Lawrence 13 Apr '23

13 Apr '23

Please save May 2, 2023, from 9:30 am - 5:00 pm, for the first Princeton Precision Health (PPH) workshop on interdisciplinary, date-driven approaches for precision medicine. We will bring together biomedical researchers, clinicians, computer scientists, ethicists, and social scientists to discuss a wide range of topics, working toward our goal of enabling more responsive, equitable, and precisely targeted health care. Registration is free but required for an accurate head count for breakfast and lunch: [ https://forms.gle/XFa4RSNhcrDTo5gp7 | https://forms.gle/XFa4RSNhcrDTo5gp7 ] Confirmed participants include: Matthias Kretzler, MD, Warner-Lambert/Parke-Davis Professor, Internal Medicine, University of Michigan. Sridhar Venkatapuram, Senior Lecturer (Associate Professor) in Global Health and Philosophy. Deputy Director, Global Health Institute. Director of Global Health Education and Training. King’s College London. Mona Singh, Wang Family Professor in Computer Science. Professor of Computer Science and the Lewis-Sigler Institute for Integrative Genomics, Princeton University. Coleen Murphy, James A. Elkins Professor in the Life Sciences. Professor of Molecular Biology and the Lewis-Sigler Institute for Integrative Genomics, Director of the Paul Glenn Laboratories for Aging Research at Princeton University. Ian Peebles, Harold T. Shapiro Postdoctoral Research Associate in Bioethics at the Center for Human Values at Princeton University. David Kinney, Postdoctoral Research Associate in Cognitive Science of Values at the Center for Human Values at Princeton University. Emily Foster-Hanson, Postdoctoral Research Associate in Cognitive Science of Values at Princeton University (Current); Assistant Professor of Psychology at Swarthmore College (Fall 2023). Hope to see you there! Olga Troyanskaya *************************************************************** Olga Troyanskaya Director, Princeton Precision Health Professor of Computer Science and the Lewis-Sigler Institute for Integrative Genomics Princeton University

1 1

ECE Seminar: Vanessa Sanchez, Stanford University TODAY at 4:30pm
by Emily C. Lawrence 13 Apr '23

13 Apr '23

ECE SEMINAR Speaker: Vanessa Sanchez, Stanford University Day: Thursday, April 13, 2023 Time: 4:30 PM Location: B205 Engineering Quadrangle Host: Jaime Fernández Fisac Title: Textiles for Soft Wearable Robotics Abstract: Wearable robots and devices-garments with embedded elements that actuate to change shape or apply forces to the wearer, typically based on signals from integrated sensors-offer promise for assistive and augmentative applications including rehabilitative gloves, haptic devices, and dynamically thermoregulating clothing. Early iterations of wearables from the 50s and 60s primarily took the form of rigid exoskeletons; however, in the past twenty years, a growing subset of this field has transitioned to the use of soft components and materials to improve portability, accessibility, fit, and comfort, guided in part by advances in the related field of soft robotics. Based on the unique requirements for wearables, including personalization for varied bodies and low cost for accessibility, automated and highly customizable textile-compatible manufacturing strategies must be developed to support the fabrication and integration of all the necessary components (sensors, actuators, interconnections). This seminar will explore the intersection of knowledge from the field of textile manufacturing with the needs of soft robots and devices, specifically focusing on wearable applications, including performance metrics, material and component choices, and fabrication strategies. Several integrated design and fabrication platforms will be presented in the context of their ability to create constituent components for wearable robots and devices. Bio: Vanessa is a fashion-designer-turned-engineer working to make our clothing smarter and more assistive through focusing on the intersection of materials, manufacturing, and robotics. She currently conducts research on supramolecular shape memory polymer fibers as an NSF MPS-Ascend Postdoctoral Research Fellow in the Department of Chemical Engineering at Stanford University. She attended The Fashion Institute of Technology (FIT) for fashion design and later graduated with a BS from the Fiber Science program at Cornell University, after which she obtained her PhD at Harvard University in Materials Science and Mechanical Engineering with a focus on soft robotic textiles as part of the Harvard Microrobotics Laboratory. Her PhD was supported by the DoD NDSEG Fellowship and the GEM Fellowship, and her resulting work has been featured in Vice, Wired, and Engadget. She has been named to the list of 50 Women in Robotics You Need to Know, recognized as an ACS CAS Future Leader, and selected for Forbes 30 Under 30.

1 0

CS Colloquium Speaker Rika Antonova TODAY at 12:30pm
by Emily C. Lawrence 13 Apr '23

13 Apr '23

Speaker: Rika Antonova, Stanford University Date: Tuesday, April 4 Time: 12:30pm EST Location: CS 105 Host: Szymon Rusinkiewicz Event page: https://www.cs.princeton.edu/events/26378 Title: Enabling Self-Sufficient Robot Learning Abstract: Autonomous exploration and data-efficient learning are important ingredients for helping machine learning handle the complexity and variety of real-world interactions. In this talk, I will describe methods that provide these ingredients and serve as building blocks for enabling self-sufficient robot learning. First, I will outline a family of methods that facilitate active global exploration. Specifically, they enable ultra data-efficient Bayesian optimization in reality by leveraging experience from simulation to shape the space of decisions. In robotics, these methods enable success with a budget of only 10-20 real robot trials for a range of tasks: bipedal and hexapod walking, task-oriented grasping, and nonprehensile manipulation. Next, I will describe how to bring simulations closer to reality. This is especially important for scenarios with highly deformable objects, where simulation parameters influence the dynamics in unintuitive ways. The success here hinges on finding a good representation for the state of deformables. I will describe adaptive distribution embeddings that provide an effective way to incorporate noisy state observations into modern Bayesian tools for simulation parameter inference. This novel representation ensures success in estimating posterior distributions over simulation parameters, such as elasticity, friction, and scale, even for scenarios with highly deformable objects and using only a small set of real-world trajectories. Lastly, I will share a vision of using distribution embeddings to make the space of stochastic policies in reinforcement learning suitable for global optimization. This research direction involves formalizing and learning novel distance metrics on this space and will support principled ways of seeking diverse behaviors. This can unlock truly autonomous learning, where learning agents have incentives to explore, build useful internal representations and discover a variety of effective ways of interacting with the world. Bio: Rika is a postdoctoral scholar at Stanford University and a recipient of the NSF/CRA Computing Innovation Fellowship. Rika completed her Ph.D. work on data-efficient simulation-to-reality transfer at KTH. Earlier, she obtained a research Master's degree from the Robotics Institute at Carnegie Mellon University. Before that, Rika was a software engineer at Google, first in the Search Personalization group, then in the Character Recognition team (developing open-source OCR engine Tesseract).

1 2

Nikhil Pimpalkhare will present an MSE talk "Context-Free Counter-Based Procedure Summarization" on April 26, 2023 at 4pm in CS 402
by Nicki Mahler 12 Apr '23

12 Apr '23

Nikhil Pimpalkhare will present an MSE talk entitled "Context-Free Counter-Based Procedure Summarization" on April 26, 2023 at 4pm in CS 402. The committee is as follows: Zachary Kincaid (adviser) and David Walker (reader) All are welcome to attend. Title: "Context-Free Counter-Based Procedure Summarization". My abstract is as follows: Abstract: Program analyses aim to formally reason about program behavior without ever running them. A typical structure of a program analysis is to compute a logical summary of an input program that over-approximates its behavior as narrowly as possible, then use automated reasoning algorithms to prove statements about program behavior. This paper addresses procedure summarization for a particularly challenging domain, nonlinear recursive programs, in a way that is both compositional and monotone. We summarize program transformations using vector addition systems with resets (VASR), a model that is simultaneously strong enough to capture interesting numerical behavior and weak enough for its reachability to remain computable. We show how to compute a VASR that is a best abstraction of an input program, how to compute the reachability of a VASR-weighted pushdown system, and how to combine these algorithms to produce summaries of programs. We additionally introduce a novel method to synthesize linear bounds on recursive depth, increasing the specificity of our summary. Together, these techniques combine for a summarization procedure which is performant, powerful, and predictable.

1 0

Sacheth Sathyanarayanan will present an MSE talk "On the Manipulability of Tournament Rules by Throwing Matches" on April 18, 2023 at 12:30pm in CS 301.
by Nicki Mahler 11 Apr '23

11 Apr '23

Sacheth Sathyanarayanan will present an MSE talk "On the Manipulability of Tournament Rules by Throwing Matches" on April 18, 2023 at 12:30pm in CS 301. The talk details follow below. Date: 04/18/2023 Time: 12.30 - 1.00 pm Location: COS 301 ( [ https://csguide.cs.princeton.edu/node/6653 | https://csguide.cs.princeton.edu/node/6653 ] ) Advisor: Matt Weinberg Reader: Mark Braverman Title: On the Manipulability of Tournament Rules by Throwing Matches Abstract: A Tournament Rule takes as input a set of {n choose 2} pairwise match results on n teams and (possibly randomly) selects a ranking of teams, with a prize associated to each place in the ranking. Two teams may manipulate their match outcomes to try and improve their expected prize winnings. [SSW 17] and subsequent works seek to understand the minimum manipulability among all " reasonabl e" ' tournament rules (i.e. tournament rules so that if team i beats team j , and also all teams that j beats, then i appears above j in the ranking with probability 1 ), when these two teams fix the match between them. Our work instead considers the possibility that two teams may both fix the match between them, and additionally throw matches to outside teams (that is, they can intentionally lose a match to a non-colluding team that they could have won). Our main result establishes that no two teams can gain more than 1/3 in expected prize winnings in Nested Randomized King of the Hill (introduced in [DFRSW 22], similar to QuickSort), by together fixing their match and throwing matches to external teams. This is optimal, as any " reasonable'' tournament admits the possibility of two teams gaining 1/3 in expected prize-winnings just by fixing the match between them.

1 0

Christopher Hodsdon will present his FPO "Stronger Abstractions and Performance Guarantees for Building Strongly Consistent Distributed Services" on May 9th, 2023 at 11am in CS 302.
by Gradinfo 11 Apr '23

11 Apr '23

Christopher Hodsdon will present his FPO "Stronger Abstractions and Performance Guarantees for Building Strongly Consistent Distributed Services" on May 9th, 2023 at 11am in CS 302. His committee is as follows: Examiners: Wyatt Lloyd (advisor), Ethan Katz-Bassett (advisor), Michael Freedman Readers: Siddhartha Sen, Ravi Netravali Everyone is invited to attend the talk. Title: Stronger Abstractions and Performance Guarantees for Building Strongly Consistent Distributed Services Abstract: Strongly consistent distributed services require complex coordination among the services’ machines. Building these services thus requires designing and reasoning about complex coordination. To make building strongly consistent distributed services easier, designers build services atop abstractions that simplify reasoning about the service. Two abstractions used today are the multi-sequence abstraction and replicated state machines. The multi-sequence abstraction simplifies ordering operations on sharded data; it explicitly orders the operations on each shard it accesses. Existing implementations have two shortcomings: failures can result in some multi-sequence numbers never being assigned, exposing a noncontiguous multi-sequence, which requires designing and implementing complex coordination in the form of consensus to handle, and existing implementations do not scale The replicated state machine (RSM) abstraction is used by many of today’s services to ensure correctness and resilience despite failures. It provides the abstraction of “a single machine that does not fail”. However, with RSM protocols used today, an individual slow replica can slow the entire RSM. RSMs are commonly geo-distributed, but current state-of-the-art protocols that provide the stronger slowdown-tolerant abstraction do not have comparable latency to the state-of-practice in a wide-area network or are unable to handle certain types of slowdowns. This dissertation shows that it is possible to build systems that strengthen both abstractions while providing performance guarantees. First, we posit that sequencers should expose our new contiguous multi-sequence abstraction. Contiguity guarantees every sequence number is assigned an operation, simplifying and strengthening the abstraction. Without scalability in the implementation of the contiguous multi-sequence, a service would need to be redesigned once it has outgrown the implementation. We design and implement MASON, the first system to expose the contiguous multi-sequence abstraction and the first to provide a scalable multi-sequence. MASON is thus an ideal building block for consistent, scalable services. Our evaluation shows MASON unlocks scalable throughput for two strongly consistent services built on it. Second, we seek to provide the stronger RSM abstraction of “a single machine that does not slow down”, with AVICENNA, a slowdown-tolerant replicated state machine protocol that has comparable latency to the current state-of-practice in the wide-area.

1 0

Promise Ekpo will present her MSE talk "Investigating Emergent Persuasiveness in Large Language Models" on Wednesday, April 19, 2023 at 12:30pm in CS 302.
by Nicki Mahler 11 Apr '23

11 Apr '23

Promise Ekpo will present her MSE talk "Investigating Emergent Persuasiveness in Large Language Models" on Wednesday, April 19, 2023 at 12:30pm in CS 302. The committee is as follows: Adviser: Prof. Jaime Fernandez Fisac Reader: Prof Karthik Narasimhan All are welcome to attend. Abstract: While the rate of progress and innovation in Artificial intelligence (AI) has many potential benefits, the potential for accidental deleterious effects cannot be overemphasized. It has been empirically demonstrated that large language models (LLMs) can learn to perform a wide range of natural language processing (NLP) tasks in a self-supervised setting. However, these models might unintentionally produce convincing arguments for false statements. There has been recent interest in improving LLM performance by fine-tuning in a reinforcement learning framework through interaction with human users. One could raise the concern that even seemingly benign reward functions can lead to goal-driven manipulation of user responses as an instrumental goal to achieve higher overall performance. This thesis seeks to investigate this possibility by evaluating the persuasiveness of self-supervised-only and reinforcement-learning-fine-tuned LLMs. In this work, we will discuss three approaches to investigating the degree of emergent persuasiveness in LLMs by searching for qualitative failures through a direct query, quantifying the persuasiveness of generated outputs, and training on this persuasiveness metric as a reward signal with reinforcement learning. Through our investigation, we find that state-of-the-art LLMs fail when prompted with statements about less popular misconceptions or domain-specific myths. With this investigation of the safety-critical related failures of LLMs, we hope to further inform the public of the degree of reliability of these models and guide their use.

1 0

Theano Stavrinos will present her FPO "Leopard: Unlocking Better Cache Performance at Lower Cost with Expiration Time-based Flash Caching" on Tuesday, May 9, 2023 at 1pm in CS 302.
by Gradinfo 11 Apr '23

11 Apr '23

Theano Stavrinos will present her FPO "Leopard: Unlocking Better Cache Performance at Lower Cost with Expiration Time-based Flash Caching" on Tuesday, May 9, 2023 at 1pm in CS 302. The committee is as follows: Examiners: Wyatt Lloyd (adviser), Kai Li, and Amit Levy Readers: Ethan Katz-Bassett (Columbia/co-adviser) and Ravi Netravali All are welcome to attend. Title: "Leopard: Unlocking Better Cache Performance at Lower Cost with Expiration Time-based Flash Caching" Abstract: Caches are crucial building blocks of web services. They keep data close to users and other services, reducing request latencies, expensive network traversals, and requests to resource-constrained backend servers. Today’s web services need high-capacity, high-performance caches for their massive working set sizes and to meet stringent performance requirements. Flash-based SSDs meet this need by providing excellent performance and high capacity at low cost. However, caching on flash involves a fundamental tradeoff. On the one hand, caches aim for low miss ratios by keeping useful objects in the cache. On the other hand, caches must protect SSDs from write-induced wear-out, which increases when useful objects are copied forward during garbage collection. Flash caches are often forced to choose between good cache performance (i.e., low cache miss ratios) and acceptable device lifespans. This dissertation describes Leopard, a flash caching framework for static content that unlocks new positions along the Pareto frontier of the miss ratio/device lifespan tradeoff, enabling more effective caching at lower cost than existing frameworks. At the foundation of Leopard's design are expiration times, which specify an object's earliest eviction time. In particular, Leopard’s garbage collection procedure uses expiration times in a cost/benefit analysis to choose the best flash block to erase, balancing write amplification from copying forward useful objects and increased miss ratios from evictions. Leopard also uses a novel clustering algorithm to group objects together by their expiration times, increasing the likelihood of low write amplification during garbage collection. Our evaluation shows that, for a range of CDN traces, Leopard significantly improves the object miss ratio achievable at a given total write volume compared to the state-of-the-art flash caching framework. Leopard also achieves better byte miss ratios at lower write volumes at most points along the byte miss ratio/write volume Pareto frontier.

1 0

ECE Seminar: Andreea Bobu, UC Berkeley, TODAY at 4:30pm
by Emily C. Lawrence 11 Apr '23

11 Apr '23

ECE SEMINAR Speaker: Andreea Bobu, University of California Berkeley Title: Aligning Robot and Human Representations Day: Tuesday, April 11, 2023 Time: 4:30 PM Location: B205 Engineering Quadrangle Host: Chi Jin Abstract: To perform tasks that humans want in the world, robots rely on a representation of salient task features; for example, to hand me a cup of coffee, the robot considers features like efficiency and cup orientation in its behavior. Prior methods try to learn both a representation and a downstream task jointly from data sets of human behavior, but this unfortunately picks up on spurious correlations and results in behaviors that do not generalize. In my view, what's holding us back from successful human-robot interaction is that human and robot representations are often misaligned: for example, our lab's assistive robot moved a cup inches away from my face -- which is technically collision-free behavior -- because it lacked an understanding of personal space. Instead of treating people as static data sources, my key insight is that robots must engage with humans in an interactive process for finding a shared representation for more efficient, transparent, and seamless downstream learning. In this talk, I focus on a divide and conquer approach: explicitly focus human input on teaching robots good representations before using them for learning downstream tasks. This means that instead of relying on inputs designed to teach the representation implicitly, we have the opportunity to design human input that is explicitly targeted at teaching the representation and can do so efficiently. I introduce a new type of representation-specific input that lets the human teach new features, I enable robots to reason about the uncertainty in their current representation and automatically detect misalignment, and I propose a novel human behavior model to learn robust behaviors on top of human-aligned representations. By explicitly tackling representation alignment, I believe we can ultimately achieve seamless interaction with humans where each agent truly grasps why the other behaves the way they do. Bio: Andreea Bobu is a Ph.D. candidate at the University of California Berkeley in the Electrical Engineering and Computer Science Department advised by Professor Anca Dragan. Her work is at the intersection of robotics, machine learning, and mathematical human modeling. Specifically, Andreea studies algorithmic human-robot interaction, with a focus on how autonomous agents and humans can efficiently and interactively arrive at shared representations of their tasks for more seamless and reliable interaction. Prior to her Ph.D. she earned her Bachelor's degree in Computer Science and Engineering from MIT in 2017. She is the recipient of the Apple AI/ML Ph.D. fellowship, is a Rising Star in EECS and an R:SS and HRI Pioneer, has won best paper award at HRI 2020, and has interned at NVIDIA Research.

1 0