Past events

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Lucy Li (University of California, Berkeley), will be giving a talk titled "Modeling Language as Social and Cultural Data".

Abstract

Language models (LMs) are powerful because they embed social identities and beliefs. Their increasing capabilities have expanded the disciplinary overlap between AI and other fields, including those in the social sciences and humanities. My talk will illustrate how I've built reciprocal relationships between natural language processing (NLP) and two other fields: sociolinguistics and education. I'll discuss how a sociolinguistic lens can inform model development, by surfacing implicit social preferences of pretraining data curation practices. In return, LMs can answer sociolinguistic research questions, uncovering the social dynamics of language at billion-word scale. Within education, I will discuss how LMs can support content analyses of school curricula. Then, I'll show how I leverage educators' in-domain expertise to create challenging multimodal benchmarks. Altogether, my work emphasizing social aspects of language contributes to both human-centered model development and empirical studies of social and cultural media.

Biography

Lucy Li is a PhD candidate at the University of California, Berkeley, affiliated with Berkeley AI Research and the School of Information. Her research intersects natural language processing with computational social science and digital humanities (e.g. cultural analytics). She has worked with Microsoft Research’s Fairness, Accountability, Transparency, and Ethics (FATE) team and the Allen Institute for AI, and led collaborations with colleagues in education, psychology, and English literature. She has been recognized by EECS Rising Stars, Rising Stars in Data Science, an American Educational Research Association (AERA) Best Paper Award, and an NSF Graduate Research Fellowship.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Matthew Gombolay (Georgia Institute of Technology), will be giving a talk titled "Value Alignment and Safety via Interactive and Explainable Human-Robot Learning".

Abstract

Value alignment and safety in generative Artificial Intelligence (AI) are key problems receiving intense focus writ large due to the emergent success of large models in many intelligence benchmarks. The threat of misaligned models becomes even more stark as we look to embody agents physically for human-robot interaction. In this talk, I will present human-centered approaches to address these critical challenges in generative AI through interactive and explainable robot learning. First, I will define challenges for safety and alignment in AI/Robotics and show that addressing the problem is more difficult than it appears at first glance. Second, I will share our recent advances to align and verify the properties of learned robot behavior through interactive machine learning and explainable artificial intelligence. Finally, I will present a roadmap for future work to bring together critical stakeholders for transdisciplinary research for beneficent AI.

Biography

Dr. Matthew Gombolay is an Associate Professor of Interactive Computing at the Georgia Institute of Technology. He was named the Anne and Alan Taetle Early-career Assistant Professor in 2018. He received a B.S. in Mechanical Engineering from Johns Hopkins University in 2011, an S.M. in Aeronautics and Astronautics from MIT in 2013, and a Ph.D. in Autonomous Systems from MIT in 2017. Between defending his dissertation and joining the faculty at Georgia Tech, Dr. Gombolay served as technical staff at MIT Lincoln Laboratory, transitioning his research to the U.S. Navy and earning an R&D 100 Award. His publication record includes best paper awards and nominations from the American Institute for Aeronautics and Astronautics, the ACM/IEEE Conference on Human-Robot Interaction, the Conference on Robot Learning, and Robotics: Science and Systems. Dr. Gombolay was selected as a DARPA Riser and received the Early Career Award from the National Fire Control Symposium, a NASA Early Career Fellowship, and the NSF CAREER award.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Abhilasha Ravichander (University of Washington), will be giving a talk titled "Illuminating Generative AI: Mapping Knowledge in Large Language Models."

Abstract

Millions of everyday users are interacting with technologies built with generative AI, such as voice assistants, search engines, and chatbots. While these AI-based systems are being increasingly integrated into modern life, they can also magnify risks, inequities, and dissatisfaction when providers deploy unreliable systems. A primary obstacle to having reliable systems is the opacity of the underlying large language models— we lack a systematic understanding of how models work, where critical vulnerabilities may arise, why they are happening, and how models must be redesigned to address them. In this talk, I will first describe my work in investigating large language models to illuminate when models acquire knowledge and capabilities. Then, I will describe my work on building methods to enable data transparency for large language models, that allows practitioners to make sense of the information available to models. Finally, I will describe my work on understanding why large language models produce incorrect knowledge, and implications for building the next generation of responsible AI systems. 

Biography

Abhilasha Ravichander is a postdoctoral scholar at the University of Washington, advised by Professor Yejin Choi.  She received her PhD from Carnegie Mellon University in 2022. Her research spans natural language processing, machine learning, and artificial intelligence, with a focus on improving the robustness and interpretability of large-scale language models.  Abhilasha’s work has been presented at several top NLP conferences, receiving Best Resource Paper Award at ACL 2024, Best Theme Paper Award at ACL 2024, Best Paper Award at the Mid-Atlantic Student Colloquium 2024, Best Paper Award at the SoCalNLP 2022 symposium, and Area Chair Favorite Paper award at COLING 2018.  She has been recognized as a "Rising Star in Generative AI" (2024), "Rising Star in EECS" (2022), and "Rising Star in Data Science" (2021). 

Interested in learning more about the University of Minnesota's Master of Science in Software Engineering program?

Reserve a spot at an upcoming information session to get all your questions answered. Sessions are held both in person and virtually. 

Info sessions are recommended for those who have at least 1-2 years of software engineering experience.

Please join us at the Computer Science and Data Science Graduate Student Department Head Town Hall. Light refreshments and snacks will be available.

Prof. Heimdahl will be discussing department budgetary policies, and Prof. Mokbel will be speaking about department culture, upcoming policies to preliminary exams, and student-advisor relationships.

This is your chance to voice your opinion and offer critical feedback on teaching, student services, and any other items you think can be improved.  Your feedback and insights are important to help us improve your graduate experience.

Please use the link below to RSVP, submit questions to department representatives, and enter your dietary restrictions.

RSVP Link

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Charlie Carlson (University of California, Santa Barbara), will be giving a talk titled "Sampling Colorings with Markov Chains."

Abstract

Graph coloring is a fundamental problem at the intersection of computer science and mathematics. A graph is k-colorable if its vertices can be assigned k colors such that adjacent vertices receive different colors. The problems of finding or sampling such colorings play crucial roles in both theoretical and practical research; they serve as natural test beds for algorithm design techniques and arise in many engineering applications. While a simple greedy algorithm can efficiently find a (d+1)-coloring for a graph with maximum degree d, the problem of uniformly sampling such a coloring is significantly more challenging. A well-known conjecture states that it is possible to efficiently sample a (d+2)-coloring.

In this talk, we will explore the importance of sampling random colorings and the history of using Markov chains to address this conjecture. Our discussion will include a review of the recent breakthrough by Carlson and Vigoda, which demonstrates the rapid mixing of Flip dynamics and the existence of an efficient algorithm for sampling colorings when k > 1.809d. Along the way, we will introduce key concepts such as Glauber dynamics, path coupling, and Flip dynamics. We conclude with a discussion of open problems and future research directions for sampling colorings and analyzing similar Markov chains. 

Biography

Charlie Carlson is a postdoctoral researcher at SLMath and the University of California, Santa Barbara. She earned her PhD in Computer Science from the University of Colorado Boulder where she was advised by Alexandra Kolla. Her research interests lie mostly at the intersection of spectral graph theory, combinatorial optimization, and random algorithms. Recently, she has been investigating connections between different methods of analyzing Markov chains.  

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Santhoshini Velusamy (Toyota (TTIC)), will be giving a talk titled "Optimization in Modern Computational Settings: Algorithms and Impossibility Results."

Abstract

The amount of data being generated and stored has increased rapidly over the last few decades. While data is measured in massive units like petabytes and zetabytes, the capacity of a device's readily accessible memory, such as RAM, is only a few gigabytes. My research focuses on solving fundamental problems in modern models of computation that consider the space limitations of local memory. In this talk, I will discuss my work on Constraint Satisfaction Problems (CSPs)—a well-studied class of combinatorial optimization problems with wide-ranging applications in computer science—in the streaming model. In addition to fully characterizing the solvability of CSPs in this model through new algorithms and matching impossibility results, my work also reveals exciting new connections to other models. I will conclude the talk by discussing future directions.

Biography

Santhoshini Velusamy is currently a Research Assistant Professor at Toyota Technological Institute at Chicago. She completed her PhD in 2023 from Harvard University, where she was supervised by Prof. Madhu Sudan. She is the recipient of a Google PhD fellowship and an NSF CRII award. Santhoshini does research in theoretical computer science, and is specifically interested in the design and analysis of algorithms in modern computational settings like streaming. She is also interested in problems in algorithmic game theory.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Ali Vakilian (Toyota (TTIC)), will be giving a talk titled "Algorithms in the AI Age: Fair and Learning-Augmented."

Abstract

The widespread deployment of AI across diverse applications offers promising opportunities while raising significant societal concerns. Motivated by these opportunities and challenges, this talk will focus on two key research directions at the intersection of algorithms and machine learning:  “learning-augmented” algorithms and fairness in algorithms and machine learning.

The increasing reliance on automated decision-making in high-stakes domains such as hiring and criminal justice has led to substantial research on the societal and ethical implications of algorithms and machine learning. In particular, fair clustering has emerged as a critical area of interest in recent years. I will discuss my work on designing efficient clustering algorithms, a fundamental task in machine learning, under various fairness notions, including  “proportional representation” within clusters or centers and “equitable access” to facilities.

In the domain of learning-augmented algorithms, the goal is to leverage patterns in data to enhance the performance of classical algorithms. This approach offers a dual promise: when provided with accurate machine-learned predictions about the input, it outperforms classical algorithms, while still maintaining strong worst-case guarantees even if the predictions are adversarial. I will outline my contributions to this field, which initiated the study of learning-augmented algorithms in the streaming model and introduced such algorithms for fundamental problems, including frequency estimation and low-rank approximation.

Biography

Ali Vakilian is a Research Assistant Professor at TTIC. His research interests include fairness of algorithms and machine learning, learning-augmented algorithms, and algorithms for massive data. Ali received his Ph.D. from MIT EECS, where he was advised by Erik Demaine and Piotr Indyk. He completed his MS studies at UIUC where he was a recipient of the Siebel Scholar award. He is a recipient of the Outstanding Student Paper Highlight Award at AISTATS 2024. For more information, visit his website at http://www.mit.edu/~vakilian/.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Zhenlin An (University of Pittsburgh), will be giving a talk titled "Bring intelligence to next-generation wireless systems."

Abstract

AI is transforming wireless systems from rule-based, manually optimized architectures into adaptive, self-learning networks. Leveraging machine learning algorithms, these networks can sense and interpret their environments through wireless signals, dynamically adjust to varying conditions, and optimize the network performance autonomously. However, traditional wireless AI models often face challenges with interoperability, limited learning efficiency, and generalizability in dynamic scenarios. In this talk, I will introduce my research on creating a physically interpretable AI framework designed to address these challenges in wireless communication, sensing, and localization. This approach fuses physics-based ray-object interaction models with neural networks to capture the intrinsic radio properties of objects. By embedding these models within a differentiable ray tracing framework, we can efficiently learn the complex interactions of wireless signals with objects, achieving precise channel predictions even with far less training data. Beyond improving channel prediction, this framework can generate high-fidelity synthetic datasets that significantly improve AI training for downstream applications like localization, sensing, and network optimization. This innovation paves the way for constructing highly accurate digital twins of radio environments, playing a crucial role in unlocking the full capabilities of NextG wireless networks.

Biography

Zhenlin An is currently a Postdoctoral Associate in Computer Science at the University of Pittsburgh and Princeton University. He earned his Ph.D. from The Hong Kong Polytechnic University. His research interests span wireless systems and networking, sensing and localization, and wireless security and privacy. Dr. An has been recognized with several prestigious awards, including the ACM China SIGBED Doctoral Dissertation Award, ACM MobiCom Best Paper Runner-Up in 2023, three Best Demo Runner-Up awards at ACM MobiCom, and two IEEE SECON Best Paper Awards. His work has been published in top-tier conferences and journals such as MobiCom, MobiSys, SIGCOMM, NSDI, INFOCOM, S&P, TNeT, and TMC. For more information about his work, visit his webpage at https://anplus.github.io.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Yanchao Zhang (Arizona State University), will be giving a talk titled "NextG Wireless Security: A Convergence of Sensing and AI."

Abstract

NextG (6G and beyond) wireless networks will seamlessly integrate sensing and AI, revolutionizing network security through real-time situational awareness, adaptive intelligence, and proactive defense mechanisms. This convergence enables context-aware security by integrating AI-driven data analytics with wireless sensing to detect, analyze, and mitigate cyber threats in real time. In this talk, I will first present a vision for the key security applications enabled by AI-sensing integration in NextG wireless systems. I will then introduce two AI-driven sensing-based security mechanisms. The first, GNN-SML, is an innovative RF-sensing-based framework that accurately and simultaneously localizes multiple wireless spectrum misusers. It employs location-centric, RF sensor-agnostic Graph Neural Networks (GNNs) with inductive learning capabilities to enhance spectrum monitoring and misuser detection. The second, WaveKey, is a cross-modal deep learning-based approach for secure and efficient key establishment in mobile ad hoc environments. WaveKey leverages a random user gesture to induce correlated IMU motion sensor data and RFID signals, extracts the complex cross-modal correlation using deep learning, and implements an Oblivious Transfer-based key agreement protocol for secure in-situ access. I will conclude by discussing open challenges and future directions in AI-sensing-driven NextG wireless security, highlighting key research opportunities at the intersection of wireless sensing, AI, and cybersecurity.

Biography

Yanchao Zhang received his Ph.D. in Electrical and Computer Engineering from the University of Florida in 2006. He is currently a Full Professor in the School of Electrical, Computer and Energy Engineering at Arizona State University. His research focuses on fundamental and experimental investigations of security and privacy in networked systems, with an emphasis on future-generation wireless and sensing systems, trustworthy wireless AI, dynamic spectrum access, autonomous and unmanned systems, immersive technologies, and their applications in critical domains. He received the NSF CAREER Award in 2009 and was elevated to IEEE Fellow in 2019 for his contributions to wireless and mobile security. He has held key leadership roles in the research community, including serving as TPC Co-Chair for IEEE INFOCOM 2023 and organizing multiple workshops for the National Science Foundation and the Army Research Office. He currently serves as Principal Investigator and Director of the DoD Center of Excellence in Future Generation Wireless Technology, a $10 million research initiative hosted at ASU that advances cutting-edge trustworthy wireless solutions for defense and critical infrastructure applications.