Events

March 21, 2023

EECS Special Seminar: Eunice Jun: "Data analysis tools for statistical non-experts"

Eunice Jun

University of Washington

11:00A

- 12:00P

Location

34-401A

Add to Calendar 2023-03-21 11:00:00 2023-03-21 12:00:00 America/New_York EECS Special Seminar: Eunice Jun: "Data analysis tools for statistical non-experts" Data analysis is critical to science, public policy, and business. Despite their importance, statistical analyses are difficult to author, especially for researchers with expertise outside of statistics. Existing statistical tools, prioritizing mathematical expressivity and computational control, are low-level while researchers’ motivating questions and hypotheses are high-level. Researchers need to translate their questions and hypotheses into low-level statistical code in an error-prone process that involves grappling with their domain knowledge, statistics, and programming. In this talk, I will introduce two tools that embody a new way of authoring analyses: Tea and Tisane. Researchers directly express their domain knowledge through higher level abstractions, and the tools will validate the data, select a statistical analysis, and implement it, all while educating analysts about why a statistical approach is valid. Tea helps analysts author statistical tests. Tea’s key insight is that statistical test selection can be cast as a constraint satisfaction problem. Tisane enables analysts to author generalized linear models with or without mixed effects, which are difficult for even statistical experts to author. Using Tisane, analysts can express their conceptual models using a high-level domain specific language. Tisane translates these conceptual models into causal DAGs and engages analysts in a disambiguation process to arrive at an output statistical model. Real-world researchers have already used these tools to conduct analyses in published research that push their own disciplines forward. I will also introduce “hypothesis formalization,” a series of cognitive and operational steps analysts take to translate their research questions into statistical implementations. Hypothesis formalization retrospectively explains why Tea improves statistical testing and directly inspired the design of Tisane. This talk exemplifies how combining human-computer interaction with other areas in and outside of computer science leads to software tools that impact real-world users. Tea and Tisane serve as platforms for further research into computational support for statistical analysis. In the future, an ecosystem of tools and computational representations to further incorporate domain knowledge throughout the data lifecycle will be critical to explain analyses and improve scientific reproducibility. Bio: Eunice Jun is a PhD candidate at the Paul G. Allen School of Computer Science & Engineering at the University of Washington, advised by Jeffrey Heer and René Just. Her research mission is to empower more people to understand and act on data. She conducts research in human-computer interaction. Close collaborations with data scientists and other researchers outside of computer science inspire her work. Her thesis combines HCI with techniques from programming languages/software engineering and statistics to develop new interactive data analysis tools. She has received a NSF graduate research fellowship, paper awards at ACM CHI and ACM CSCW, and an honorable mention from the Barry Goldwater Foundation.https://mit.zoom.us/j/99750055409 34-401A

Ken Holstein - Designing for complementarity in AI-augmented work

Ken Holstein

Carnegie Mellon

4:00P

- 5:00P

Location

32-D463 (Star)

Add to Calendar 2023-03-21 16:00:00 2023-03-21 17:00:00 America/New_York Ken Holstein - Designing for complementarity in AI-augmented work Abstract:AI systems are increasingly used to augment human work in complex social and creative contexts, including social work, education, design, and content moderation. These technologies are typically introduced with the promise of overcoming human limitations and biases. However, AI judgments in such settings are themselves likely to be imperfect and biased, even if in different ways than humans. In this talk, I argue that today’s worker-facing AI systems often represent missed opportunities to meaningfully complement and enhance human workers’ abilities. I will present real-world case studies and conceptual frameworks from our research, which illustrate critical challenges to achieving human-AI complementarity in practice. I will then share some of our efforts to overcome these challenges, targeting various points across the AI project lifecycle—from the earliest problem formulation stages, to the design of AI system evaluations, to the development of worker-AI interfaces.Bio:Ken Holstein is an Assistant Professor in the Human-Computer Interaction Institute at Carnegie Mellon University, where he leads the Co-Augmentation, Learning, & AI (CoALA) Lab:https://www.thecoalalab.com/. Their research focuses on supporting effective forms of AI-augmented work in real-world contexts, and on scaffolding more responsible AI development practices in industry and the public sector. Throughout their work, they draw on approaches from human-computer interaction, AI, design, psychology, statistics, and the learning sciences. Their work has received awards at ACM CHI, IEEE SaTML, AIED, and ICLS, and has been covered by outlets such as PBS, Wired, Forbes, The Boston Globe, The Hechinger Report, and Marketplace Tech.The talk will also be streaming live over Zoom: https://mit.zoom.us/j/91645609513. 32-D463 (Star)

Quest | CBMM Seminar Series - Quantifying and Understanding Memorization in Deep Neural Networks

Chiyuan Zhang

Google

4:00P

- 5:00P

Location

Singleton Auditorium (46-3002)

Add to Calendar 2023-03-21 16:00:00 2023-03-21 17:00:00 America/New_York Quest | CBMM Seminar Series - Quantifying and Understanding Memorization in Deep Neural Networks Abstract: Deep learning algorithms are well-known to have a propensity for fitting the training data very well and memorize idiosyncratic properties in the training examples. From a scientific perspective, understanding memorization in deep neural networks shed light on how those models generalize. From a practical perspective, understanding memorization is crucial to address privacy and security issues related to deploying models in real world applications. In this talk, we present a series of studies centered at quantifying memorization in neural language models. We explain why in many real world tasks, memorization is necessary for optimal generalization. We also present quantitative studies on memorization, forgetting and unlearning of both vision and language models, to better understand the behaviors and implications of memorization in those models.

EECS Special Seminar: Kuan Fang, "Scalable Robot Intelligence: Self-Supervised Learning through Generation"

Kuan Fang

UC Berkeley

4:00P

- 5:00P

Location

32-G449 Kiva

Add to Calendar 2023-03-21 16:00:00 2023-03-21 17:00:00 America/New_York EECS Special Seminar: Kuan Fang, "Scalable Robot Intelligence: Self-Supervised Learning through Generation" Abstract:Rapid advances in deep learning have resulted in promising techniques for robots to boost their capabilities to perceive, reason, and act by leveraging large models and massive datasets. However, the scalability of existing robot learning methods is severely limited by the manual labor and domain knowledge that humans can provide. To acquire general-purpose skills for solving a broad range of tasks, intelligent robots need scalable methods to collect and learn from rich data without extensive human supervision. In this talk, I will present my research on scaling up robot learning through the autonomous generation of environments, goals, and tasks. I will start by describing how to leverage procedural content generation for learning robust skills that can handle the variety and uncertainty of the real world. Then I will present algorithms that train robots to effectively reuse skills learned from prior experiences for novel sequential tasks by learning to generate reachable subgoals. Finally, I will demonstrate how to enable robots to discover a repertoire of novel skills by adaptively generating tasks during training. The acquired skills can be used for solving a variety of complex tasks such as tool use and sequential manipulation based on raw sensory inputs. Bio:Kuan Fang is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences at UC Berkeley, working with Sergey Levine. He received his Ph.D. degree in Electrical Engineering from Stanford University, advised by Fei-Fei Li and Silvio Savarese. His research interests lie at the intersection of robotics, computer vision, and machine learning, with a focus on developing data-driven methods to enable intelligent robots to operate in unstructured environments. He is a recipient of the Stanford Graduate Fellowship and the Computing Innovation Fellowship. 32-G449 Kiva

March 22, 2023

EECS Special Seminar: Joachim Neu “Internet-Scale Consensus In The Blockchain Era”

Joachim Neu

Stanford University

11:00A

- 12:00P

Location

Grier A 34-401A

Add to Calendar 2023-03-22 11:00:00 2023-03-22 12:00:00 America/New_York EECS Special Seminar: Joachim Neu “Internet-Scale Consensus In The Blockchain Era” Abstract:Blockchains have ignited interest in Internet-scale consensus as a fundamental building block for decentralized applications and services, which in turn promise more egalitarian access and improved robustness to faults and abuse. While consensus has been studied in distributed systems for decades, Internet-scale consensus requires a fundamental rethinking of models, desiderata, and protocols. Participants are no longer a handful of computers in one company’s data centers, but numerous mistrustful entities distributed across the Internet. I will discuss two examples of challenges and solutions for this new setting: (1) Ethereum, the second largest blockchain, aims to strengthen consensus liveness under open participation where parties come and go at will, and to strengthen safety to enable accountability in case of any safety violation. However, we show that no traditional single-ledger protocol can satisfy both strengthened properties simultaneously. To resolve this dilemma, we develop the multi-ledger consensus paradigm that is now the security design-specification for Ethereum. A by-product of this work are attacks on Ethereum’s consensus protocol that prompted design changes. (2) Traditional network models do not capture rate constraints on communication and processing, leaving popular “provably secure” protocols vulnerable to attack. We show via a new queuing-based model how to schedule message handling securely. Our policies are simple enough to be forward-deployed at Internet service providers via a system that can also protect traffic of applications beyond blockchains.Bio:Joachim Neu is a PhD candidate at Stanford University with David Tse. His research focuses on the science and engineering of Internet-scale consensus as a fundamental building block for decentralized systems, using tools from distributed systems, applied probability, networking and communications, and applied cryptography. While a Masters student at Technical University of Munich, he worked in information and coding theory. Joachim has received the Protocol Labs PhD Fellowship and the Stanford Graduate Fellowship. Grier A 34-401A

Hot Topics in Computing: First Contact

Sebastien Bubeck

5:00P

- 6:00P

Location

32-123

Kirsch Auditorium

Add to Calendar 2023-03-22 17:00:00 2023-03-22 18:00:00 America/New_York Hot Topics in Computing: First Contact Title: First ContactAbstract: The new wave of AI systems, ChatGPT and its more powerful successors, exhibit extraordinary capabilities across a broad swath of domains. In light of this, we will discuss whether artificial INTELLIGENCE has arrived. Bio: Sebastien Bubeck is a Senior Principal Research Manager in the Machine Learning Foundations group at Microsoft Research (MSR). He joined the Theory Group at MSR in 2014, after three years as an assistant professor at Princeton University. His works on convex optimization, online algorithms and adversarial robustness in machine learning received several best paper awards (NeurIPS 2018 and 2021 best paper, ALT 2018 and 2023 best student paper in joint work with MSR interns, COLT 2016 best paper, and COLT 2009 best student paper). Recently he has been focused on exploring a physics-like theory of neural networks learning.http://sbubeck.com/ 32-123

March 23, 2023

Northrop Grumman Recruitment Session with Chief Engineer Scott Beasley (SB `88, Physics)

Scott Beasley (SB `88, Physics)

9:30A

- 12:30P

Location

32-G370 (above G-side stairs)

Add to Calendar 2023-03-23 9:30:00 2023-03-23 12:30:00 America/New_York Northrop Grumman Recruitment Session with Chief Engineer Scott Beasley (SB `88, Physics) MIT alumnus and Northrop Grumman IBCS Chief Engineer Scott Beasley (SB `88, Physics) will be in Room G-370 at CSAIL (above the G entrance staircase) on Thursday, March 23rd from 9:30am-12:30pm to meet with prospective graduate students and postdocs who are interested in opportunities at Northrop and specifically their Future Technical Leaders (FTL) Program. FTL is a professional development opportunity aimed at identifying and investing in Northrop Grumman’s next generation of technologists, engineers, architects, and leaders. Northrop’s FTL Program is seeking talented MS and PhD Dec 2023- June 2024 expected graduates interested in learning more about our exclusive three year rotational program working side-by-side with senior technical management. To be eligible for the FTL Program, you must be at the top of your class in a Master’s or Doctoral program in STEM-focused majors. Due to the classified nature of the work, you must be a U.S. citizen and able to obtain a security clearance. Our majors of interest are wide and varied. Info sessions will be available during the places and times below. No appointment or registration is required so please stop by for a few minutes! When/Where: Room 32-G370 (above G stairs) on Thursday, March 23rd 9:30am-12:30pmFTL information: https://www.northropgrumman.com/future-technical-leaders-at-northrop-grumman/Scott’s contact email: SCOTT BEASLEY | WISŁA IBCS ART Chief Engineer | FTL RecruiterNorthrop Grumman Corporation | Defense SystemsO: 256-428-5584 | C: 757-593-7767 | scott.beasley@ngc.com & beez@mit.edu

EECS Special Seminar: Thierry Tambe “Architecting High Performance Silicon Systems for Accurate and Efficient On-Chip Deep Learning”

Thierry Tambe

Harvard University

10:00A

- 12:00P

Location

Grier B 34-401B

Add to Calendar 2023-03-23 10:00:00 2023-03-23 12:00:00 America/New_York EECS Special Seminar: Thierry Tambe “Architecting High Performance Silicon Systems for Accurate and Efficient On-Chip Deep Learning” Abstract:The unabated pursuit for omniscient and omnipotent AI is levying hefty latency, memory, and energy taxes at all computing scales. At the same time, the end of Dennard scaling has sunsetted traditional performance gains commonly attained with reduction in transistor feature size. Faced with these challenges, my research is building a heterogeneity of solutions co-optimized across the algorithm, memory subsystem, hardware architecture, and silicon stack to generate breakthrough advances in arithmetic performance, compute density and flexibility, and energy efficiency for on-chip machine learning, and natural language processing (NLP) in particular. I will start, in the algorithm front, by discussing award-winning work on developing a novel floating-point based data type, AdaptivFloat, which enables resilient quantized AI computations; and is particularly suitable for NLP networks with very large parameter distribution. Then, I will describe a 16nm chip prototype that adopts AdaptivFloat in the acceleration of noise-robust AI speech and machine translation tasks – and whose fidelity to the front-end application is verified via a formal hardware/software compiler interface. Towards the goal of lowering the prohibitive energy cost of inferencing large language models on TinyML devices, I will describe a principled algorithm-hardware co-design solution, validated in a 12nm chip tapeout, that accelerates Transformer workloads by tailoring the accelerator’s latency and energy expenditures according to the complexity of the input query it processes. Finally, I will conclude with some of my current and future research efforts on further pushing the on-chip energy-efficiency frontiers by leveraging specialized non-conventional dynamic memory structures for on-device training — and recently prototyped in a 16nm tapeout.Bio:Thierry Tambe is a final year Electrical Engineering PhD candidate at Harvard University advised by Prof. Gu-Yeon Wei and Prof. David Brooks. His current research interests focus on designing energy-efficient and high-performance algorithms, hardware accelerators and systems for machine learning and natural language processing in particular. He also bears a keen interest in agile SoC design methodologies. Prior to debuting his doctoral studies, Thierry was an engineer at Intel in Hillsboro, Oregon, USA designing various mixed-signal architectures for high-bandwidth memory and peripheral interfaces on Xeon and Xeon-Phi HPC SoCs. He received a B.S. (2010) and M.Eng. (2012) in Electrical Engineering from Texas A&M University. Thierry Tambe is a recipient of the Best Paper Award at the 2020 ACM/IEEE Design Automation Conference, a 2021 NVIDIA Graduate PhD Fellowship, and a 2022 IEEE SSCS Predoctoral Achievement Award. Grier B 34-401B

EECS Special Seminar: Saikat Dutta, "Randomness-Aware Testing of Machine Learning-based Systems"

Saikat Dutta

University of Illinois Urbana-Champaign

11:00A

- 12:00P

Location

34-401A

Add to Calendar 2023-03-23 11:00:00 2023-03-23 12:00:00 America/New_York EECS Special Seminar: Saikat Dutta, "Randomness-Aware Testing of Machine Learning-based Systems" Abstract: The goal of my research is to develop novel testing techniques and tools to make Machine Learning-based systems more reliable. Machine Learning is rapidly revolutionizing the way modern-day systems are developed. However, testing Machine Learning-based systems is challenging due to 1) the presence of non-determinism, both internal (e.g., stochastic algorithms) and external (e.g., execution environment), and 2) the absence of well-defined accuracy specifications. Most traditional software testing techniques widely used today cannot tackle these challenges because they often assume determinism and require a precise test oracle.In this talk, I will present my work on automated testing of Machine Learning-based systems and on improving developer-written tests in such systems. To achieve these goals, I develop principled techniques that build on solid mathematical foundations from probability theory and statistics to reason about the underlying non-determinism and accuracy. I implement my techniques in practical and scalable tools that help developers to detect more bugs and to efficiently navigate trade-offs between test quality and efficiency. To date, my research has exposed more than 50 bugs and improved the quality of more than 200 tests in over 60 popular open-source ML libraries, many of which are widely used at companies like DeepMind, Google, Meta, Microsoft, and Uber as well as in many academic and scientific communities.Biography: Saikat Dutta is a PhD candidate in Computer Science at the University of Illinois Urbana-Champaign, advised by Prof. Sasa Misailovic. Saikat’s research interests lie at the intersection of Software Engineering and Machine Learning. Saikat’s current research focuses on improving the reliability of Machine-learning based systems by developing novel testing techniques and tools. Saikat has received the Facebook PhD Fellowship, the 3M Foundation Fellowship, and the Mavis Future Faculty Fellowship for his contributions. 34-401A

Top-down Network Management at Scale

Ying Zhang (Meta)

4:00P

- 5:00P

Location

32-G882

Add to Calendar 2023-03-23 16:00:00 2023-03-23 17:00:00 America/New_York Top-down Network Management at Scale Abstract: Network management facilitates a healthy and sustainable network. However, its practice is not well understood outside the network engineering community. In this talk, I will present Meta’s network management suit, ranging from network planning, to risk-driven network management, and ultimately to network operation. I will present our intent-driven top-down network management automation that facilitates the network reliability, performance, and long-term evolution. By sharing our operational experiences of running a large-scale network for years, as well as our experiences in supporting a diverse set of networks from backbone, to PoP, to data center, we hope to inspire new research in the area of reliable and verifiable network management.Bio: Ying Zhang is Senior Engineering Manager in Meta. She works on large scale network management problems. Her team owns software for network modeling, design, configuration and verification for Metaverse Datacenter, backbone and edge networks. Her research interests are in network management, Network for AI, Software-Defined Networks, Network Function Virtualization, Internet routing, and network security.Zoom link: https://mit.zoom.us/j/91470317185 32-G882

From Robustness to Privacy and Back

Lydia Zakynthinou

Northeastern University

4:00P

- 5:00P

Location

32-D507

32-D507

Add to Calendar 2023-03-23 16:00:00 2023-03-23 17:00:00 America/New_York From Robustness to Privacy and Back Abstract: We study the relationship between two desiderata of algorithms in statistical inference and machine learning—differential privacy and robustness to adversarial data corruptions. Their conceptual similarity was first observed by Dwork and Lei (STOC 2009), who observed that private algorithms satisfy robustness, and gave a general method for converting robust algorithms to private ones. However, all general methods for transforming robust algorithms into private ones lead to suboptimal error rates. Our work gives the first black-box transformation that converts any adversarially robust algorithm into one that satisfies pure differential privacy. Moreover, we show that for any low-dimensional estimation task, applying our transformation to an optimal robust estimator results in an optimal private estimator. Thus, we conclude that for any low-dimensional task, the optimal error rate for ε-differentially private estimators is essentially the same as the optimal error rate for estimators that are robust to adversarially corrupting 1/ε training samples. We apply our transformation to obtain new optimal private estimators for several high-dimensional tasks, including Gaussian (sparse) linear regression and PCA. Finally, we present an extension of our transformation that leads to approximate differentially private algorithms whose error does not depend on the range of the output space, which is impossible under pure differential privacy.Joint work with Hilal Asi and Jonathan Ullman. 32-D507

Fireside Chat and "Ask Me Anything"

Sanjit Biswas

CEO, Samsara

4:00P

- 5:00P

Location

32-123

Add to Calendar 2023-03-23 16:00:00 2023-03-23 17:00:00 America/New_York Fireside Chat and "Ask Me Anything" ABSTRACT: Join us for a fireside chat and AMA with Sanjit Biswas, MIT alum and successful entrepreneur who co-founded Meraki in 2006 based on the MIT Roofnet research project. Meraki was acquired by Cisco in 2012. Then, in 2015, he co-founded Samsara, one of the most successful companies in the Internet of Things (IoT) area, which had a successful IPO in December 2021 and is seeing vibrant growth. This AMA promises to teach us about converting research to commercial products and lessons on entrepreneurship, leadership, and bringing advanced technologies to market.BIO: Sanjit Biswas is the CEO and co-founder of Samsara, which allows businesses that depend on physical operations to harness IoT (Internet of Things) data to develop actionable business insights and improve their operations. He co-founded the company with the mission to increase the safety, efficiency, and sustainability of the operations that power the global economy. Prior to Samsara, Sanjit was the CEO and co-founder of Meraki, one of the most successful networking companies of the past decade. Sanjit has been recognized as an MIT Technology Review “Innovator Under 35” honoree, a Technology Pioneer by the World Economic Forum, Glassdoor’s 2018 Top CEO, one of Goldman Sachs 100 Most Intriguing Entrepreneurs, and one of The Software Report’s top 50 SaaS CEOs. Sanjit holds a B.S. in Computer Systems Engineering from Stanford and an S.M. in Electrical Engineering and Computer Science from MIT. 32-123

Specialization and the End of Moore's Law

Neil Thompson

MIT

7:00P

- 8:00P

Location

zoom

Add to Calendar 2023-03-23 19:00:00 2023-03-23 20:00:00 America/New_York Specialization and the End of Moore's Law IEEE Computer Society and GBC/ACMonline 7:00 PM, Thursday, 23 March 2023Specialization and the End of Moore's LawNeil Thompson, MITRegister in advance for this webinar at https://acm-org.zoom.us/webinar/register/5016750322508/WN_2k-jR3BqSvihFD8EnJnIzQAfter registering, you will receive a confirmation email containing information about joining the webinar.Abstract:For decades, Moore’s Law made the economics of specialized chips unattractive because the upfront costs couldn’t be justified when the alternative was fast-improving CPUs. As Moore’s Law fades, however, this is changing. Not only is specialization becoming more economically attractive, but it is now one of the best ways to get performance improvements for many applications. In this talk, I will discuss (1) how the economics of specialization have changed, (2) how specialization is fracturing computing in ways commonly seen in other technologies, and (3) how long we can expect the gains from specialization to make up for the slowdown in Moore’s Law.Bio:Neil Thompson is the Director of the FutureTech research project at MIT’s Computer Science and Artificial Intelligence Lab, where his group studies the economic and technical foundations of progress in computing. He is also a Principal Investigator at MIT’s Initiative on the Digital Economy.Previously, Neil was an Assistant Professor of Innovation and Strategy at the MIT Sloan School of Management, where he co-directed the Experimental Innovation Lab (X-Lab), and a Visiting Professor at the Laboratory for Innovation Science at Harvard. He has advised businesses and government on the future of Moore’s Law, has been on National Academies panels on transformational technologies and scientific reliability, and is part of the Council on Competitiveness’ National Commission on Innovation & Competitiveness Frontiers.He has a PhD in Business and Public Policy from Berkeley, where he also did Masters degrees in Computer Science and Statistics. He also has a masters in Economics from the London School of Economics, and undergraduate degrees in Physics and International Development. Prior to academia, he worked at organizations such as Lawrence Livermore National Laboratory, Bain and Company, the United Nations, the World Bank, and the Canadian Parliament.This joint meeting of the Boston Chapter of the IEEE Computer Society and GBC/ACM will be online only due to the COVID-19 lockdown.Up-to-date information about this and other talks is available online at https://ewh.ieee.org/r1/boston/computer/. You can sign up to receive updated status information about this talk and informational emails about future talks at https://mailman.mit.edu/mailman/listinfo/ieee-cs, our self-administered mailing list.

March 24, 2023

CSL Seminar - Sam Gershman - Reverse engineering human exploration

Sam Gershman

Harvard University

3:00P

- 4:00P

Location

G449 (Patil/Kiva)

G449 (Patil/Kiva)

Add to Calendar 2023-03-24 15:00:00 2023-03-24 16:00:00 America/New_York CSL Seminar - Sam Gershman - Reverse engineering human exploration Abstract: This talk explores how humans solve the exploration-exploitation dilemma in reinforcement learning. Mirroring the multiplicity of algorithmic solutions studied in machine learning, the human brain also appears to employ several different algorithms. In particular, evidence indicates that humans use a combination of uncertainty-guided directed and random exploration, as well as more sophisticated algorithms that rely on structured world knowledge. These findings point towards a convergence of natural and artificial intelligence.Bio: Sam Gershman received his B.A. in Neuroscience and Behavior from Columbia University in 2007 and his Ph.D. in Psychology and Neuroscience from Princeton University in 2013. From 2013-2015, he was a postdoctoral fellow in the Department of Brain and Cognitive Sciences at MIT. He is currently a Professor in the Department of Psychology and Center for Brain Science at Harvard. His research focuses on computational cognitive neuroscience approaches to learning, memory and decision making. G449 (Patil/Kiva)

March 30, 2023

EECS Special Seminar: Orson(Xuhai) Xu, "Toward The Next-Generation Health and Well-being: The Synergy of Behavior Modeling and Intervention"

Orson(Xuhai) Xu

University of Washington

11:00A

- 12:00P

Location

34-401A

Add to Calendar 2023-03-30 11:00:00 2023-03-30 12:00:00 America/New_York EECS Special Seminar: Orson(Xuhai) Xu, "Toward The Next-Generation Health and Well-being: The Synergy of Behavior Modeling and Intervention" AbstractWhile the intelligence of everyday smart devices (e.g., smartphones and wearables) continues to evolve, they can already capture basic health behaviors such as physical activities and heart rates. However, we still face significant challenges in leveraging these devices for modeling and influencing long-term behavior, such as mental well-being. Existing AI techniques for longitudinal behavior are still far from being deployable.In this talk, I will describe innovations in machine learning approaches for building longitudinal behavior models. These novel approaches will address key issues in AI deployability, such as cross-dataset generalizability and interpretability, in the context of mental health. Based on these models, I will further present novel intervention techniques that influence behavior and promote mental well-being.BioOrson Xuhai Xu is a 5th-year PhD candidate in the Information School at the University of Washington, advised by Anind K. Dey and Jennifer Mankoff. His research straddles multiple disciplines, including human-computer interaction, ubiquitous computing, machine learning, and health. By leveraging sensing data from everyday devices, Xu develops novel, deployable machine learning algorithms to model long-term human behavior related to well-being. Based on these behavior models, he designs new behavior change intervention methods and novel interaction techniques for promoting well-being. Xu has earned several awards, including 5 Best Paper, Best Paper Honorable Mention, and Best Artifact awards. His research has been covered by media outlets such as The Washington Post, ACM News, and UW News. He was recently recognized as the Gaetano Borriello Outstanding Student Award Winner at UbiComp 2022. 34-401A

April 04, 2023

Advanced Machine Learning Methods to Accelerate Materials Discovery

Santiago Miret

10:00A

- 11:00A

Location

Kiva/Patel 32-G449

Add to Calendar 2023-04-04 10:00:00 2023-04-04 11:00:00 America/New_York Advanced Machine Learning Methods to Accelerate Materials Discovery The ability to apply advanced machine learning (ML) to process large amounts of heterogenous data can greatly aid in the understanding, discovery, and design of new materials for advanced application like clean energy, sustainable semiconductor manufacturing and drug discovery. The heterogenous nature of data encompasses materials structural data, material property data and qualitative descriptions across various types of materials, including solid-state (e.g., semiconductors, batteries) and molecules. This multitude of multi-modal data types necessitates the application of a diverse of advanced ML techniques across different fields. Additionally, computational materials design is only the first step in the materials creation process, which involves a large range of experimental synthesis and analysis techniques creating additional technical challenges and uncertainties. Overall, applying ML to materials discovery provides a notable platform for interdisciplinary ML research with real-world impactful applications. In this talk, I will present an overview of Intel Labs’ research efforts focused on applying ML to materials discovery through a closed-loop discovery paradigm spanning automated materials design, automated material synthesis and automated materials characterization. Based on the closed-loop discovery framework, I will illustrate some example research efforts detailing materials property modeling using geometric deep learning, materials discovery using new generative algorithms and materials understanding through natural language processing.Bio:Santiago Miret is an AI Researcher at Intel Labs where he focuses on applying AI for scientific problems with an emphasis on materials discovery and materials understanding. Through this effort, Santiago manages a wide range of academic collaborations focused on applying AI for scientific application. Among these collaborations, there have been notable engagements with the Matter Lab led by Alán Aspuru-Guzik at the University of Toronto and various AI laboratories at MILA in Montreal that have led to cross-institutional publications at various machine learning venues. Santiago was a primary organizer of the 1st AI for Accelerated Materials Discovery (AI4Mat) workshop at NeurIPS 2022, which brought together domain experts from various fields of materials science and AI to exchange research work and ideas in an interdisciplinary forum. Prior to working at Intel Labs, Santiago obtained his PhD in Materials Science and Engineering from the University of California, Berkeley. Kiva/Patel 32-G449

EECS Special Seminar:Milijana Surbatovich, "Designing Formally Correct Intermittent Systems"

Milijana Surbatovich

Carnegie Mellon

11:00A

- 12:00P

Location

34-401A

Add to Calendar 2023-04-04 11:00:00 2023-04-04 12:00:00 America/New_York EECS Special Seminar:Milijana Surbatovich, "Designing Formally Correct Intermittent Systems" Abstract: "Extreme edge computing" is an emerging computing paradigm targeting application domains like medical wearables, disaster-monitoring tiny satellites, or smart infrastructure. This paradigm brings sophisticated sensing and data processing into an embedded device's deployment environment, enabling computing in environments that are too harsh, inaccessible, or dense to support frequent communication with a central server. Batteryless, energy harvesting devices (EHDs) are key to enabling extreme edge computing; instead of using batteries, which may be too costly or even impossible to replace, they can operate solely off energy collected from their environment. However, harvested energy is typically too weak to power a device continuously, causing frequent, arbitrary power failures that break traditional software and make correct programming difficult. Given the high assurance requirements of the envisioned application domains, EHDs must execute software without bugs that could render the device inoperable or leak sensitive information. While researchers have developed intermittent systems to support programming EHDs, they rely on informal, undefined correctness notions that preclude proving such necessary correctness and security properties.My research lays the foundation for designing formally correct intermittent systems that provide correctness guarantees. In this talk, I show how existing correctness notions are insufficient, leading to unaddressed bugs. I then present the first formal model of intermittent execution, along with correctness definitions for important memory consistency and timing properties. I use these definitions to design and implement both the language abstractions that programmers can use to specify their desired properties and the enforcement mechanisms that uphold them. Finally, I discuss my future research directions in intermittent system security and leveraging formal methods for full-stack correctness reasoning. Bio: Milijana Surbatovich is a PhD Candidate in the Electrical and Computer Engineering Department at Carnegie Mellon University, co-advised by Professors Brandon Lucia and Limin Jia. Her research interests are in applied formal methods, programming languages, and systems for intermittent computing and non-traditional computing platforms broadly. She is excited by research problems that require reasoning about correctness and security across the architecture, system, and language stack. She was awarded CMU's CyLab Presidential Fellowship in 2021 and was selected as a 2022 Rising Star in EECS. Previously, she received an MS in ECE from CMU in 2020 and a BS in Computer Science from the University of Rochester in 2017. 34-401A

April 05, 2023

ELSA: Secure Aggregation for Federated Learning with Malicious Actors

Mayank Rathee

UC Berkeley

4:00P

- 5:00P

Location

32-G449 (Kiva)

Add to Calendar 2023-04-05 16:00:00 2023-04-05 17:00:00 America/New_York ELSA: Secure Aggregation for Federated Learning with Malicious Actors Federated learning (FL) is an increasingly popular approach for machine learning (ML) when the training dataset is highly distributed. Clients perform local training on their datasets and the updates are then aggregated into the global model. Existing protocols for aggregation are either inefficient or don’t consider the case of malicious actors in the system. This is a major barrier to making FL an ideal solution for privacy-sensitive ML applications.In this talk, I will present ELSA, a secure aggregation protocol for FL that breaks this barrier - it is efficient and addresses the existence of malicious actors (clients + servers) at the core of its design. Similar to prior work Prio and Prio+, ELSA provides a novel secure aggregation protocol built out of distributed trust across two servers that keeps individual client updates private as long as one server is honest, defends against malicious clients, and is efficient end-to-end. Compared to prior works, the distinguishing theme in ELSA is that instead of the servers generating cryptographic correlations interactively, the clients act as untrusted dealers of these correlations without compromising the protocol’s security. This leads to a much faster protocol while also achieving stronger security at that efficiency compared to prior work. We introduce new techniques that retain privacy even when a server is malicious at a small added cost of 7-25% in runtime with a negligible increase in communication over the case of a semi-honest server.ELSA improves end-to-end runtime over prior work with similar security guarantees by big margins - single-aggregator RoFL by up to 305x (for the models we consider), and distributed-trust Prio by up to 8x (with up to 16x faster server-side protocol). Additionally, ELSA can be run in a bandwidth-saver mode for clients who are geographically bandwidth-constrained - an important property that is missing from prior works. Based on joint work with Conghao Shen, Sameer Wagh, and Raluca Ada Popa. 32-G449 (Kiva)

April 06, 2023

EECS Special Seminar: Michael Kim, "Foundations of Responsible Machine Learning"

Michael Kim

UC Berkeley

11:00A

- 12:00P

Location

34-401A

Add to Calendar 2023-04-06 11:00:00 2023-04-06 12:00:00 America/New_York EECS Special Seminar: Michael Kim, "Foundations of Responsible Machine Learning" Abstract: Algorithms make predictions about people constantly. The spread of such prediction systems has raised concerns that machine learning algorithms may exhibit problematic behavior, especially against individuals from marginalized groups. This talk will provide an overview of my research building a theory of "responsible" machine learning. I will highlight a notion of fairness in prediction, called Multicalibration (ICML'18), which requires predictions to be well-calibrated, not simply overall, but on every group that can be meaningfully identified from data. This "multi-group" approach strengthens the guarantees of group fairness definitions, without incurring the costs (statistical and computational) associated with individual-level protections. Additionally, I will present a new paradigm for learning, Outcome Indistinguishability (STOC'21), which provides a broad framework for learning predictors satisfying formal guarantees of responsibility. Finally, I will discuss the threat of Undetectable Backdoors (FOCS'22), which represent a serious challenge for building trust in machine learning models.Bio: Michael P. Kim is a Postdoctoral Research Fellow at the Miller Institute for Basic Research in Science at UC Berkeley, hosted by Shafi Goldwasser. Before this, Kim completed his PhD in Computer Science at Stanford University, advised by Omer Reingold. Kim's research addresses basic questions about the appropriate use of machine learning algorithms that make predictions about people. More generally, Kim is interested in how the computational lens (i.e., algorithms and complexity theory) can provide insights into emerging societal and scientific challenges. 34-401A

Deep Generative Physical Models for MRI Reconstruction

Jon Tamir

University of Texas at Austin

1:00P

- 2:00P

Location

32-370

Add to Calendar 2023-04-06 13:00:00 2023-04-06 14:00:00 America/New_York Deep Generative Physical Models for MRI Reconstruction Recently, deep learning techniques have been used as powerfuldata-driven reconstruction methods for inverse problems, and inparticular have led to reduced scan times in magnetic resonanceimaging (MRI). Typically, these methods are implemented usingend-to-end supervised learning based on idealized imaging conditions.While promising, reconstruction quality is known to degrade whenapplied to natural measurement and anatomy perturbations. In this talkwe present an alternative approach to deep learning reconstructionbased on distribution learning, in which we train a deep generativemodel to learn image priors without reference to the measurementprocess. We show that decoupling the measurement and statisticalmodels provides a powerful framework for MRI reconstruction. Weleverage recent advances in score-based generative modeling to learnthe prior distribution and we pose the image reconstruction task asposterior sampling. We show that this approach is competitive withend-to-end methods when applied to in-distribution data, and wedemonstrate theoretical and empirical robustness to variousout-of-distribution shifts. In cases where the distribution shift islarge, we empirically show a small amount of training data issufficient to recover the performance. Finally, we demonstrate theutility of our framework for image reconstruction in the presence ofsubject motion. 32-370

Collaborative Robotics: From Dexterity to Teammate Prediction

Monroe Kennedy III

Computer Science Department, Stanford University

4:00P

- 5:00P

Location

32-D463(Star)

Add to Calendar 2023-04-06 16:00:00 2023-04-06 17:00:00 America/New_York Collaborative Robotics: From Dexterity to Teammate Prediction Abstract: There have been amazing advances in the field of robotic autonomy over the past few decades, we’ve seen robots move from factory floors to working examples of autonomous vehicles on public roads. However, there is still a notable barrier to having robots in everyday environments performing functional tasks. These barriers are usually due to current robots having limited dexterity when it comes to manipulating objects that humans can handle with minimal difficulty, to barriers that exist when a robot is supposed to work alongside a human to complete a task or enable a human to perform a task better. In this talk, we will address the catalyst problem of robotic autonomy in human environments by addressing robotic dexterity and the necessity of improving the sense of robotic touch. We will then look at how robots can become effective teammates by modeling the behavior of their collaborators in the context of the task or environment, which enables them to predict future behavior and take collaborative actions.Biography: Monroe Kennedy is an assistant professor in Mechanical Engineering and by courtesy, Computer Science at Stanford University. Monroe is the recipient of the NSF Faculty Early Career Award. He directs the Assistive Robotics and Manipulation Laboratory (ARMLab), where the focus is on developing collaborative, autonomous robots capable of performing dexterous, complex tasks with human and robotic teammates. Monroe received his Ph.D. in Mechanical Engineering and Applied Mechanics and master’s in Robotics from the University of Pennsylvania. 32-D463(Star)

April 19, 2023

Thesis Defense

Sameer Khurana

MIT CSAIL

3:00P

- 5:00P

Location

Seminar Room G449 (Patil/Kiva)

Add to Calendar 2023-04-19 15:00:00 2023-04-19 17:00:00 America/New_York Thesis Defense Seminar Room G449 (Patil/Kiva)

May 02, 2023

Thesis Defense

Sameer Khurana

MIT CSAIL

3:00P

- 6:00P

Location

Seminar Room G882 (Hewlett Room)

Add to Calendar 2023-05-02 15:00:00 2023-05-02 18:00:00 America/New_York Thesis Defense Seminar Room G882 (Hewlett Room)

Q. Vera Liao - Human-Centered Explainable AI: From Algorithms to User Experiences

Q. Vera Liao

Microsoft Research

4:00P

- 5:00P

Location

32-D463 (Star)

Add to Calendar 2023-05-02 16:00:00 2023-05-02 17:00:00 America/New_York Q. Vera Liao - Human-Centered Explainable AI: From Algorithms to User Experiences Abstract:Artificial Intelligence technologies are increasingly used to aid human decisions and perform autonomous tasks in critical domains. The need to understand AI in order to improve, contest, develop appropriate trust, and better interact with AI systems has spurred great academic and public interest in Explainable AI (XAI). The technical field of XAI has produced a vast collection of algorithms in recent years. However, explainability is an inherently human-centric property and the field is starting to embrace human-centered approaches. Human-computer interaction (HCI) research and user experience (UX) design in this area are increasingly important especially as practitioners begin to leverage XAI algorithms to build XAI applications. In this talk, I will draw on my own research and broad HCI works to highlight the central role that human-centered approaches should play in shaping XAI technologies, including driving technical choices by understanding users’ explainability needs, uncovering pitfalls of existing XAI methods, and providing conceptual frameworks for human-compatible XAI.Bio:Q. Vera Liao is a Principal Researcher at Microsoft Research Montréal, where she is part of the FATE (Fairness, Accountability, Transparency, and Ethics in AI) group. Her current research interests are in human-AI interaction, explainable AI, and responsible AI. Prior to joining MSR, she worked at IBM Research and studied at the University of Illinois at Urbana-Champaign and Tsinghua University. Her research received multiple paper awards at ACM and AAAI venues. She currently serves as the Co-Editor-in-Chief for Springer HCI Book Series, on the Editorial Board of ACM Transactions on Interactive Intelligent Systems (TiiS), an Editor for CSCW, and an Area Chair for FAccT 2023.The talk will also be streamed over Zoom: https://mit.zoom.us/j/96445121768. 32-D463 (Star)