Events

Hide Filters Show Filters
Hide Filters Show Filters

Event Type

Impact Area

Research Area

Seminar Series

Reset Filters

December 06, 2024

No events scheduled

December 09, 2024

Daniel Genkin: Side Channel Attacks: Lessons Learned or Troubles Ahead?

Daniel Genkin (Georgia Tech)
4:00P
- 6:00P

Location

32-G575
Add to Calendar 2024-12-09 16:00:00 2024-12-09 18:00:00 America/New_York Daniel Genkin: Side Channel Attacks: Lessons Learned or Troubles Ahead? The security and architecture communities will remember the past five years as the era of side channels. Starting from Spectre and Meltdown, time and again we have seen how basic performance-improving features can be exploited to violate fundamental security guarantees. Making things worse, the rise of side channels points to a much larger problem, namely the presence of large gaps in the hardware-software execution contract on modern hardware. In this talk, I will give an overview of this gap, in terms of both security and performance. First, I will give a high-level survey on speculative execution attacks such as Spectre and Meltdown. I will then talk about how speculative attacks are still a threat to both kernel and browser isolation primitives, despite numerous mitigation attempts. Next, going beyond speculation, I will discuss security issues involving violation of constant time guarantees due to data-memory prefetching. Finally, taking the performance perspective, I will discuss new techniques for microarchitectural code optimizations, with an emphasis on compute-heavy workloads such as cryptographic protocols.The talk will be interactive and include attack demonstrations. Bio:Daniel Genkin is an Alan and Anne Taetle Early Career Associate Professor at the School of Cybersecurity and Privacy at Georgia Tech. Daniel’s research interests are in hardware and system security, with particular focus on side channel attacks and defenses. Daniel’s work has been recognized by best paper awards in multiple academic and industry venues, as well as covered by national and scientific press. Recently, Daniel has been part of the team performing the first analysis of speculative and transient execution, resulting in the discovery of Spectre, Meltdown and follow ups. Daniel has a PhD in Computer Science from the Technion Israel’s Institute of Technology and has been awarded the 2024 Alfred P. Sloan Research Fellowship. 32-G575

December 10, 2024

Tech Talk [Bayer]: Data-Centric Revolution: Enhancing AI with Croissant Metadata

Steffen Vogler (Bayer)
2:00P
- 3:00P

Location

32-G449 KIVA
Add to Calendar 2024-12-10 14:00:00 2024-12-10 15:00:00 America/New_York Tech Talk [Bayer]: Data-Centric Revolution: Enhancing AI with Croissant Metadata The deployment of AI-based medical software has yielded mixed results, highlighting the urgent need for a data-centric approach to machine learning. This talk investigates the transformative potential of the community tool called "Croissant", a standardized metadata format developed to enhance the ML-readiness of datasets. By shifting the focus from model architecture to the quality and structure of input data, Croissant facilitates the creation of more robust and accurate AI systems. We examine a proposal of an ecosystem supporting Croissant, emphasizing its integration with data providers, tech enablers, and governance frameworks. Furthermore, we explore how this innovative tool can streamline the development process, improve data discoverability, and promote responsible AI practices, ultimately leading to improved outcomes in healthcare and other domains.Steffen Vogler (he/him) is a Principal Imaging Technology Scientist at Bayer Radiology-R&D, leading research and product development on AI in medical computer vision with special focus on the Radiology domain. His interest is in data-centric machine learning, ethical AI and health equity. He has been member of the ITU-WHO Focus Group "Artificial Intelligence for Health" and currently curator of the discovery track “Data-centric Machine Learning for Good” at “AI for Good”. Prior to joining Bayer, he did a PhD in Neurobiology and worked on basic research question around memory formation in the mammal brain.- new meta data exchange format https://github.com/mlcommons/croissantBayer’s AI Innovation Platform https://app.innovationplatform.ai/

Coboundary Expansion Inside Chevalley Coset Complex HDXs

Ryan O'Donnell

Part Of

4:15P
- 5:15P

Location

32-G449 (Kiva/Patil)
Add to Calendar 2024-12-10 16:15:00 2024-12-10 17:15:00 America/New_York Coboundary Expansion Inside Chevalley Coset Complex HDXs Refreshments at 4:00 PMAbstract:Recent major results in property testing and PCPs were unlocked by moving to high-dimensional expanders (HDXs) constructed from C_d-type buildings, rather than the long-known A_d-type ones. At the same time, these building quotient HDXs are not as easy to understand as the more elementary (and more symmetric/explicit) coset complex HDXs constructed by Kaufman-Oppenheim [KO18] (of A-d-type) and O’Donnell-Pratt [OP22] (of Bd-, Cd-, Dd-type). Motivated by these considerations, we study the B_3-type generalization of a recent work of Kaufman-Oppenheim [KO21], which showed that the A_3-type coset complex HDXs have good 1-coboundary expansion in their links, and thus yield 2-dimensional topological expanders. The crux of Kaufman-Oppenheim’s proof of 1-coboundary expansion was: (1) identifying a group-theoretic result by Biss and Dasgupta on small presentations for the A_3-unipotent group over F_q; (2) "lifting" it to an analogous result for an A_3-unipotent group over polynomial extensions F_q[x]. For our B_3-type generalization, the analogue of (1) appears to not hold. We manage to circumvent this with a significantly more involved strategy: (1) getting a computer-assisted proof of vanishing 1-cohomology of B_3-type unipotent groups over F_5; (2) developing significant new "lifting" technology to deduce the required quantitative 1-cohomology results in B_3-type unipotent groups over F_{5^k}[x].Joint work with Noah G. Singer (Carnegie Mellon). 32-G449 (Kiva/Patil)

December 12, 2024

GlucOS: Security, correctness, and simplicity for automated insulin delivery

Hari Venugopalan
UC Davis
12:00P
- 1:00P

Location

G882 (Hewlett Room)
Add to Calendar 2024-12-12 12:00:00 2024-12-12 13:00:00 America/New_York GlucOS: Security, correctness, and simplicity for automated insulin delivery Abstract:We present GlucOS, a novel system for trustworthy automated insulin delivery. Fundamentally, this paper is about a system we designed, implemented, and deployed on real humans and the lessons learned from our experiences. GlucOS combines algorithmic security, driver security, and end-to-end verification to protect against malicious ML models, vulnerable pump drivers, and drastic changes in human physiology. We use formal methods to prove correctness of critical components and incorporate humans as part of our defensive strategy. Our evaluation includes both a real-world deployment with seven individuals and results from simulation to show that our techniques generalize. Our results show that GlucOS maintains safety and improves glucose control even under attack conditions. This work demonstrates the potential for secure, personalized, automated healthcare systems. G882 (Hewlett Room)

December 16, 2024

Alexander Lex - The reVISit User Study Platform and Applications in Studying Misinformation

Alexander Lex
University of Utah-School of Computing

Part Of

4:00P
- 5:00P

Location

D463 (Star)
Add to Calendar 2024-12-16 16:00:00 2024-12-16 17:00:00 America/New_York Alexander Lex - The reVISit User Study Platform and Applications in Studying Misinformation Abstract:In this talk I introduce the reVISit framework for designing and running empirical studies online. Traditional survey tools limit the flexibility and reproducibility of online experiments. To remedy this, we introduce a domain-specific language, the reVISit Spec, that researchers can use to design complex online user studies. reVISit Spec, combined with the relevant stimuli, is compiled into a ready-to-deploy website that handles all aspects of a user study, including sophisticated provenance-based data tracking, randomization, etc. reVISit is a community focused project and ready to use! Visit https://revisit.dev/ to get started. I will then pivot to talk about data-driven misinformation in the form of charts shared on social networks. I will demonstrate that “lying with charts” doesn’t work the way we (used to) think about it, and introduce a few strategies to “protect” charts and charting tools from being abused by malicious users. I will connect back to reVISit by illustrating how we leveraged it to run a series of crowd-sourced experiments. Bio:Alexander Lex is an Associate Professor of Computer Science at the Scientific Computing and Imaging Institute and the Kahlert School of Computing at the University of Utah. He directs the Visualization Design Lab where he and his team develop visualization methods and systems to help solve today’s scientific problems. Recently he is working on visualization accessibility, visual misinformation, provenance and reproducibility, and user study infrastructure. He is the recipient of an NSF CAREER award and multiple best paper awards or best paper honorable mentions at IEEE VIS, ACM CHI, and other conferences. He also received a best dissertation award from his alma mater. He co-founded datavisyn, a startup company developing visualization solutions for the pharmaceutical industry.This talk will also be streamed over Zoom: https://mit.zoom.us/j/99703652090. D463 (Star)

January 30, 2025

Learning to Reason with LLMs

Noam Brown
OpenAI
2:00P
- 3:30P

Location

Seminar Room G449 (Patil/Kiva)
Add to Calendar 2025-01-30 14:00:00 2025-01-30 15:30:00 America/New_York Learning to Reason with LLMs Large language models (LLMs) have demonstrated remarkable capabilities in generating coherent text and completing various natural language tasks. Nevertheless, their ability to perform complex, general reasoning has remained limited. In this talk, I will describe OpenAI's new o1 model, an LLM trained via reinforcement learning to generate a hidden chain of thought before its response. We have found that the performance of o1 consistently improves with more reinforcement learning compute and with more inference compute. o1 surpasses previous state-of-the-art models in a variety of benchmarks that require reasoning, including mathematics competitions, programming contests, and advanced science question sets. I will discuss the implications of scaling this paradigm even further. Seminar Room G449 (Patil/Kiva)

February 19, 2025