Triangle Computer Science Distinguished Lecturer Series

4:00 PM – 5:00 PM ET

Sitterson 011

Watch a recording

Title: Nonparametric Language Models: Trading Data for Parameters (and Compute) in Large Language Models

Host School: UNC – Chapel Hill

Abstract

 Large language models (LLMs) such as ChatGPT have taken the world by storm, but are incredibly expensive to train, requiring significant amounts of data and computational resources. They also hallucinate, e.g. by regularly introducing made up facts, and are difficult to keep up to date over time, as the world around them changes. In this talk, I will survey some of our recent work on non-parametric and retrieval-based language models, which are instead designed to be easily extensible and provide much more careful provenience for their predictions. The key idea is to trade parameters for data; rather than attempting to memorize all the worlds facts and knowledge in the learned parameters of a single monolithic LM, we instead provide the model an explicit knowledge store (e.g. a collection of web pages from Wikipedia) that can be used to look up information in real time. This is a relatively new research direction where best practices are still forming, but I will argue retrieval augmentation is a very general idea that can lead to much more efficient training, can provide fundamentally new insights into how LLMs work, and is broadly applicable to a range of settings, including e.g. models that do text-to-image generation. I will also provide, to the best of my ability, a guess about where things are going and what it would take to convince every major LLM to go non-parametric in the near future.

Short Biography

Luke Zettlemoyer is a Professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington, and a Research Director at Meta. His research focuses on empirical methods for natural language semantics, and involves designing machine learning algorithms and models, introducing new tasks and datasets, and, most recently, studying how to best develop self-supervision signals for pre-training. His honors include being named an ACL Fellow as well as winning a PECASE award, an Allen Distinguished Investigator award, and multiple best paper awards. Luke was an undergrad at NC State (over 20 years ago…), received his PhD from MIT, and was a postdoc at the University of Edinburgh.

4:00 PM – 5:00 PM ET

Title: Beyond 5G Networks: Strategic Roadmap and R&D Prospects

Host School: Duke University

Abstract

During 2020-2022, the speaker served as Program Lead for Innovate Beyond 5G, DoD OUSD Office of Research & Engineering, managing an annual portfolio for research and prototyping in support of significant investment by DoD/USG with respect to adopting 5G for the Enterprise and Innovating Beyond (i. e. pushing technology boundaries beyond current 5G implementation). Accordingly, the talk will be divided into 2 components:

1. An Initial Overview of DoD 5G-to-xG R&D strategy (towards enabling future Integrated Tactical Network concepts based on emerging 5G New Radio (NR) network features such as direct Device-2-Device (sidelink) modes, non-terrestrial networking, impact of RAN disaggregation and virtualization, dynamic spectrum sharing and management between 5G/B5G and DoD/Federal networks (current emphasis on 3.1-3.45 GHz)

followed by a deeper dive into

2. Cellular V2X, notably performance aspects of Sidelink Mode-2 enabled ad-hoc networking, current status of Integrated Access & Backhaul (IAB) highlighting feature gaps and R&D opportunities.

The talk will conclude with distilling the impacts of the above as potential vectors for 5G+/6G and implications for 3GPP standards evolution beyond Rel-18.

Short Biography

Sumit Roy (Fellow, IEEE 2007) received the B. Tech. degree from the Indian Institute of Technology (Kanpur) in 1983, and the M. S. and Ph. D. degrees from the University of California (Santa Barbara), all in Electrical & Comp. Engineering (1985 and 1988 respectively), as well as an M. A. in Statistics and Applied Probability (1988). His previous academic appointments were at the Moore School of Electrical Engineering, University of Pennsylvania, and at the University of Texas, San Antonio. His research interests and technology expertise spans analysis/design and prototyping of wireless communication systems/networks, with an emphasis on various technologies: 5G wireless LANs (802.11ax), 5G New Radio and emerging 5G/beyond 5G standards for vehicular (terrestrial and airborne) networks, multi-standard inter-networking/coexistence and dynamic spectrum access solutions for spectrum sharing. He was elevated to IEEE Fellow by Communications Society for “contributions to multi-user communications theory and cross-layer design of wireless networking standards” and held the ECE-CoE Integrated Systems Term Professorship (2014- 19) at Univ. of Washington in recognition of his international reputation in the area.

He spent 2001-03 on academic leave at Intel Wireless Technology Lab as a Senior Researcher engaged in systems architecture definition and IEEE standards contributions for ultra- wideband systems (Wireless PANs) and next generation high-speed (pre-802.11n) wireless LANs. He served as Science Foundation of Ireland Isaac Walton Fellow during a sabbatical at University College, Dublin (Jan-Jun 2008) and was the recipient of a Royal Acad. Engineering (UK) Distinguished Visiting Fellowship during summer 2011. During 2014-15, he spent a sabbatical year at Microsoft Research, Bangalore, India, as Erskine Fellow at University of Canterbury, New Zealand and as Short Term Visiting Foreign Expert at Shanghai JiaoTung University. His research has been consistently funded by various US agencies and industrial organizations leading to 10 awarded US patents and his work on Radar-WiFi Spectrum Sharing published in 2016 was recognized by IEEE Trans. Aersosp. Elect. Systems with Barry Carlton (Best Paper) Award. He also currently represents US on the NATO SET-302 Technical Working Group on `Cognitive Radar’.

He continues to be professionally active in IEEE Communications Society (ComSoc) – notably IEEE Future Networks Initiative (https://futurenetworks.ieee.org/) for which he currently serves as Distinguished Lecturer. He has served as Associate Editor for all the major ComSoc publications (IEEE Trans. Communications, IEEE J. Sel. Areas of Communications, IEEE Trans. on Wireless Communications, IEEE Trans. Mobile Computing) at various times and was previously selected for two stints as IEEE ComSoc Distinguished Lecturer (2013-2015, 2017- 18). He was also elected to Executive Comm. Member for the National Spectrum Consortium (www.nationalspectrumconsortium.org). Between Sep. 2020-22, he served as Program Lead for Innovate Beyond 5G program within US DoD Office of Under Secy. R&E’s 5G-to-xG initiative https://www.cto.mil/5g/ .

4:00 PM – 5:00 PM ET

Title: Balancing multiple objectives in contextual multi-armed bandits

Host School: N.C. State University

Abstract

Contextual multi-armed bandits are a popular framework for learning to make good decisions under uncertainty, and are applicable to areas ranging from ad placement to optimizing flu shot reminders. The majority of work in this space assumes the goal is to learn a decision policy to map from contexts to decisions in a way that maximizes the cumulative sum of outcomes of interest, such as total clicks or flu shot appointments. However in many real world settings there are multiple objectives of interest: for instance, a stakeholder may have a limited budget, care about fairness to subpopulations, or may wish to balance the experience for those participating in a study with the generalizable knowledge that might be learned and of use for other situations. In this talk I’ll discuss some of our work on learning to make decisions under uncertainty given multiple objectives of interest, and highlight motivating settings in education, healthcare and public policy.

Short Biography

Emma Brunskill is an associate professor in the Computer Science Department at Stanford University where she and Brunskill’s lab aim to create AI systems that learn from few samples to robustly make good decisions. Their work spans algorithmic and theoretical advances to experiments, inspired and motivated by the positive impact AI might have in education and healthcare. Brunskill’s lab is part of the Stanford AI Lab, the Stanford Statistical ML group, and AI Safety @Stanford. Brunskill has received a NSF CAREER award, Office of Naval Research Young Investigator Award, a Microsoft Faculty Fellow award and an alumni impact award from the computer science and engineering department at the University of Washington. Brunskill and her lab have received multiple best paper nominations and awards both for their AI and machine learning work (UAI best paper, Reinforcement Learning and Decision Making Symposium best paper twice) and for their work in Ai of education (Intelligent Tutoring Systems Conference, Educational Data Mining conference x3, CHI).

4:00 PM – 5:00 PM ET

Title: Storing and Querying Video Data

Host School: Duke University

Abstract

The proliferation of inexpensive high-quality cameras coupled with recent advances in machine learning and computer vision have enabled new applications on video data. This in turn has renewed interest in video data management systems (VDMSs). In this talk, we explore how to build a modern data management system for video data. We focus, in particular, on the storage manager and present several techniques to store video data in a way that accelerates queries over that data. We then move up the stack and discuss different types of data models that can be exposed to applications. Finally, we discuss how it’s possible to support users in expressing queries to find events of interest in a video database.

Short Biography

Magdalena Balazinska is Professor, Bill & Melinda Gates Chair, and Director of the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Magdalena’s research interests are in the field of database management systems. Her current research focuses on data management for data science, big data systems, cloud computing, and image and video analytics. Prior to her leadership of the Allen School, Magdalena was the Director of the eScience Institute, the Associate Vice Provost for Data Science, and the Director of the Advanced Data Science PhD Option. She also served as Co-Editor-in-Chief for Volume 13 of the Proceedings of the Very Large Data Bases Endowment (PVLDB) journal and as PC co-chair for the corresponding VLDB’20 conference. Magdalena is an ACM Fellow. She holds a Ph.D. from the Massachusetts Institute of Technology (2006). Shortly after her arrival at the University of Washington, she was named a Microsoft Research New Faculty Fellow (2007). Magdalena received the inaugural VLDB Women in Database Research Award (2016) for her work on scalable distributed data systems. She also received an ACM SIGMOD Test-of-Time Award (2017) for her work on fault-tolerant distributed stream processing and a 10-year most influential paper award (2010) from her earlier work on reengineering software clones.

4:00 PM – 5:00 PM ET

Title: Mitigating Technology Abuse in Intimate Partner Violence and Encrypted Messaging

Host School: Duke University

Abstract

Computer security is traditionally about the protection of technology, whereas trust and safety efforts focus on preventing technology abuse from harming people. In this talk, I’ll explore the interplay between security and tech abuse, and make the case that trust and safety represents an important frontier for computer security researchers. To do so, I’ll draw on examples from two lines of my recent work.

First, I’ll overview our work on technology abuse in the context of intimate partner violence (IPV). IPV is a widespread social ill affecting about one in our women and one in ten men at some point in their lives. Via interviews with survivors and professionals, online measurement studies, and reverse engineering of malicious tools, our research has provided the most granular view to date of technology abuse in IPV contexts. This has helped educate our efforts on intervention design, most notably in the form of what we call clinical computer security: direct, expert assistance to help survivors navigate technology abuse. Our work led to establishing the Clinic to End Tech Abuse, which has so far worked to help hundreds of survivors of IPV in New York City.

Second, I’ll discuss how basic security tools like encrypted messaging need to be adapted in light of tech abuse. Here we find a fundamental tension between the desire for messaging service providers to help moderate malicious content and the confidentiality goals of encryption, which prevent the platform from seeing content. I’ll show how we end up reconceptualizing and redesigning basic cryptographic tools to more securely support abuse mitigation.

The talk will include content on abuse, including discussion of physical, sexual, and emotional violence.

Short Biography

Thomas Ristenpart is an Associate Professor at Cornell Tech and a member of the Computer Science department at Cornell University. Before joining Cornell Tech in May, 2015, he spent four and a half years as an Assistant Professor at the University of Wisconsin-Madison. He completed his PhD at UC San Diego in 2010. His research spans a wide range of computer security topics, with recent focuses including digital privacy and safety in intimate partner violence, anti-abuse mitigations for encrypted messaging systems, improvements to authentication mechanisms including passwords, and topics in applied and theoretical cryptography. His work is routinely featured in the media and has been recognized by a number of distinguished paper awards, two ACM CCS test-of-time awards, an Advocate of New York City award, an NSF CAREER Award, and a Sloan Research Fellowship.

Dates and titles are subject to change.