Department Seminars

Speakers to be announced.

View the recording >>

Computer Science Seminar Series

December 19, 2024

Abstract: Emotion AI, increasingly used in mundane (e.g., entertainment) to high-stakes (e.g., education, health care, workplace) contexts, refers to technologies that claim to algorithmically recognize, detect, predict, and infer emotions, emotional states, moods, and even mental health status using a wide range of input data. While emotion AI is critiqued for associated scientific validity, bias, and surveillance concerns, it continues to be patented, developed, and used without public debate, resistance, or regulation. In this talk, Nazanin Andalibi highlights some of her research group’s work focusing on the workplace to discuss: 1) how emotion AI technologies are conceived of by their inventors and what values are embedded in their design, and 2) the perspectives of the humans who produce the data that make emotion AI possible, and whose experiences are shaped by these technologies: data subjects. Andalibi argues that emotion AI is not just technical, it is sociotechnical, political, and enacts/shifts power. She shows how emotion AI could harm the very conditions its advocates promise it will improve (e.g., worker well-being, work conditions), rendering it a problematic choice for addressing structural challenges workers face in the workplace. Andalibi concludes that even with technical reforms (e.g., reducing biases, improving accuracy) many emotion AI-inflicted harms (e.g., emotional labor, privacy harms) would persist.

Speaker Biography: Nazanin Andalibi is an assistant professor in the School of Information at the University of Michigan and is an affiliate faculty member at the university’s Digital Studies Institute; Center for Ethics, Society, and Computing; and Center for Social Media Responsibility. As a social computing and human-computer interaction (HCI) scholar, her research examines how marginality is experienced, enacted, facilitated, or disrupted in and as mediated through sociotechnical systems such as artificial intelligence and social media. For example, a central theme of her research examines the privacy, ethical, justice, and policy implications of emotion AI technologies in high-stakes contexts including the workplace, job interviews, social media, and health care. Andalibi has published 54 peer-reviewed full publications in highly competitive venues, including numerous award-winning publications. She has secured $1.16 million in funding from the National Science Foundation, including an NSF CAREER award. Her work’s policy and practical impact is evidenced by her publications’ citations in numerous policy documents, speaking at the Federal Trade Commission, and an invited keynote at Reddit’s ModSummit conference. Further, she is on the editorial board of the flagship HCI journal ACM Transactions on Computer-Human Interaction, regularly serves on conference organizing committees, and serves on the ACM Special Interest Group on Computer-Human Interaction’s CARES committee.

View the recording >>

Computer Science Seminar Series

December 17, 2024

Abstract: Bilevel optimization has emerged as a foundational framework in modern machine learning for developing principled computational tools across diverse domains, including meta-learning, automated ML, reinforcement learning, and robotics. This talk will explore recent advancements in bilevel optimization algorithms and their applications in machine learning. In the first part, Kaiyi Ji will introduce several efficient implicit gradient-based algorithms featuring flexible double- and single-loop structures. He will then present a novel adaptive tuning-free approach that significantly reduces hyperparameter tuning efforts while maintaining strong convergence guarantees. For the more challenging scenario involving non-unique lower-level solutions, Ji will discuss penalty-based methods that ensure provable convergence using only first-order gradient information. The second part of the talk will focus on two practical applications of bilevel optimization: coreset selection for machine learning and imperative learning for optimization and robotics. Finally, Ji will discuss promising directions for future research in theoretical and applied bilevel optimization and in other areas such as multi-objective learning and continual learning.

Speaker Biography: Kaiyi Ji is an assistant professor in the Department of Computer Science and Engineering (CSE) of the University at Buffalo and an affiliated faculty member with university’s Institute for Artificial Intelligence and Data Science. He was a postdoctoral research fellow in the Department of Electrical Engineering and Computer Science at the University of Michigan, Ann Arbor in 2022. Ji earned his PhD in electrical and computer engineering from the Ohio State University in 2021 and was a visiting student research collaborator in the Department of Electrical Engineering at Princeton University. He completed his BS degree at the University of Science and Technology of China in 2016. Ji’s research lies at the intersection of optimization, machine learning, and communication networks, spanning both theoretical foundations and practical applications. His current interests include bilevel optimization, multi-objective and multi-task learning, and continual learning, with applications in deep learning, communication, and recommendation systems. He has received several honors, including a CSE Junior Faculty Research Award, a UB Presidential Fellowship, and multiple NSF awards.

View the recording >>

Computer Science Seminar Series

December 12, 2024

Abstract: Our everyday choices—from what we eat to how we communicate—carry significant consequences for our health, relationships, and the environment. These actions impact critical societal challenges such as climate change and growing polarization. How could computer science help? While generative AI offers new opportunities to empower prosocial actions at scale, several key challenges limit our ability to leverage AI tools for good. In this talk, Kristina Gligorić will explore how AI can identify opportunities for intervention and deliver scalable solutions to address pressing societal issues. Gligorić will outline how she carries out this approach in two examples of societal issues fundamental to fostering a healthy and equitable society: civility and sustainability. First, she will discuss how generative AI can support constructive online conversations. She will present novel large language model methods for detecting harmful content (Proceedings of the National Academy of Sciences ’24) and describe a human-in-the-loop system that uses generative AI to assist users in creating constructive posts, developed in an ongoing collaboration with a social media company. In the second part, Gligorić will then shift focus to climate issues and promoting sustainable dietary habits in campus environments, highlighting findings on barriers to sustainability (PNAS Nexus ’24, Frontiers in Nutrition ’24). She will demonstrate how AI tools can assist dining hall chefs and food scientists in revising menus and designing experiments to promote more sustainable eating habits. These interventions have made a positive impact on many people; the work on sustainable dietary behaviors helped shape an on-campus food system that serves thousands of students and staff daily, while online interventions changed the behavior of thousands of users and moderators, with changes persisting over six months. Finally, Gligorić will outline efforts to develop AI interventions across diverse societal contexts. By combining LLM predictions with expert annotations (arXiv ’24), we can extend AI’s impact across domains like health care, policymaking, and law. Gligorić will conclude by discussing future directions for (1) improving LLMs to address social contexts better, (2) creating tools to assist social scientists, and (3) scaling AI assistance tools and human-in-the-loop interventions to empower experts in tackling complex problems. These directions will enable using AI to address societal challenges at scale.

Speaker Biography: Kristina Gligorić is a postdoctoral scholar in Stanford University’s Computer Science Department. Previously, she obtained her PhD in computer science at the École Polytechnique Fédérale de Lausanne, or EPFL. Her research focuses on computational approaches to address societal issues by developing and applying large language models, data science, and causal inference methods. Her work has been published in top computer science conferences focused on computational social science and natural language processing—such as the Association for Computational Linguistics, the ACM Conference on Computer-Supported Cooperative Work and Social Computing, and the International Association for the Advancement of Artificial Intelligence Conference on Web and Social Media (CWSC)—and broad audience journals including the Proceedings of the National Academy of Sciences, Nature Communications, and Nature Medicine. She is a Swiss National Science Foundation fellow, a Massachusetts Institute of Technology Electrical Engineering and Computer Science Rising Star, and a University of Chicago Rising Star in Data Science. She has received several awards for her work, including the EPFL Thesis Distinction, a CSCW 2021 Best Paper Honorable Mention Award, and the EPFL Best Teaching Assistant Award.

View the recording >>

Institute for Assured Autonomy & Computer Science Seminar Series

December 10, 2024

Abstract: Algorithmically guided decisions are becoming increasingly prevalent and, if left unchecked, can amplify pre-existing societal biases. In this talk, Allison Koenecke uses modern computational tools to examine the equity of decision-making in two complex systems: automated speech recognition and online advertising. In the former, Koenecke audits popular speech-to-text systems (developed by Amazon, Apple, Google, IBM, Microsoft, and OpenAI) and demonstrate disparities in transcription performance for African American English speakers and speakers with language impairments—patterns likely stemming from a lack of diversity in the data used to train the systems. These results point to hurdles faced by non-“standard” English speakers in using widespread tools driven by speech recognition technology. In the second part of her talk, Koenecke proposes a methodological framework for online advertisers to determine a demographically equitable allocation of individuals being shown ads for SNAP (food stamp) benefits. This framework measures what different populations believe is a “fair” allocation of ad budgets in a constrained setting, given cost trade-offs between English-speaking and Spanish-speaking SNAP applicants; she uncovers broad consensus across demographics for some degree of equity over pure efficiency. Both projects exemplify processes to reduce disparate impact in algorithmic decision-making.

Speaker Biography: Allison Koenecke is an assistant professor of information science at Cornell University. Her research on algorithmic fairness applies computational methods, such as machine learning and causal inference, to study societal inequities in domains from online services to public health. Koenecke is regularly quoted as an expert on disparities in automated speech-to-text systems. She previously held a postdoctoral researcher role at Microsoft Research and received her PhD from Stanford University’s Institute for Computational and Mathematical Engineering. Named on Forbes’ 2023 “30 Under 30 in Science” list, Koenecke is the recipient of several NSF grants and the Cornell Bowers Computing and Information Science Diversity, Equity, Inclusion, and Belonging Faculty of the Year Award.

View the recording >>

Computer Science & Center for Language and Speech Processing Seminar Series

December 9, 2024

Abstract: To achieve truly autonomous AI agents that can handle complex tasks on our behalf, we must first trust them with our sensitive data while ensuring they can learn from and use this information responsibly. Yet current language models frequently expose private information and reproduce copyrighted content in unexpected ways, highlighting why we need to move beyond simplistic “good” or “bad” blanket rules toward context-aware systems. In this talk, we will examine data exposure through membership inference attacks, develop controlled generation methods to protect information, and design privacy frameworks grounded in contextual integrity theory. Looking ahead, we’ll explore emerging directions in formalizing semantic privacy, developing dynamic data controls, and creating evaluation frameworks that bridge technical capabilities with human needs.

Speaker Biography: Niloofar Mireshghallah is a postdoctoral scholar at the Paul G. Allen Center for Computer Science & Engineering at the University of Washington. She received her PhD in 2023 from the Department of Computer Science and Engineering at the University of California, San Diego. Mireshghallah’s research interests include privacy in machine learning, natural language processing, and generative AI and law. She is a recipient of the 2020 National Center for Women & IT Aspirations in Computing Collegiate Award, a finalist for the 2021 Qualcomm Innovation Fellowship, and a 2022 Rising Star in both Adversarial Machine Learning and Electrical Engineering and Computer Science.

View the recording >>

Computer Science & Center for Language and Speech Processing Seminar Series

December 6, 2024

Abstract: Recent advances in large language models have achieved remarkable results across a wide range of natural language processing applications, including text classification, summarization, machine translation, and dialogue systems. As LLMs grow increasingly capable, the need to control their generation process becomes more pressing, particularly for high-stakes applications that demand reliable outputs adhering to specific guidelines or creative outputs within defined boundaries. However, the dominant auto-regressive paradigm—training models to predict the next word based on prior context—poses significant challenges for enforcing structural or content-specific constraints. In this talk, Nanyun “Violet” Peng will present her recent work on controllable natural language generation that moves beyond the conventional auto-regressive framework to enhance both the reliability and creativity of generative models. She will introduce controllable decoding-time algorithms that guide auto-regressive models to better align with user-specified constraints. Additionally, she will discuss a novel insertion-based generation paradigm that breaks away from the limitations of auto-regressive methods. These approaches enable more reliable and creative outputs, with applications spanning creative writing, lexical-controlled generation, and commonsense-compliant text generation.

Speaker Biography: Nanyun “Violet” Peng is an associate professor of computer science at the University of California, Los Angeles. Her research focuses on controllable and creative language generation, multilingual and multimodal models, and the development of automatic evaluation metrics, with a strong commitment to advancing robust and trustworthy AI. Her work has been recognized with honors such as an Outstanding Paper Award at the Annual Conference of the North American Chapter of the Association for Computational Linguistics in 2022, three Outstanding Paper Awards at Empirical Methods in Natural Language Processing in 2024, oral paper selections at the 2022 Conference and Workshop on Neural Information Processing Systems and the 2023 International Conference on Machine Learning, and several Best Paper Awards at workshops affiliated with premier AI and NLP conferences; she was also featured in the 2022 International Joint Conferences on Artificial Intelligence Early Career Spotlight. Peng’s research has received support from prestigious funding sources, including an NSF CAREER Award, a National Institutes of Health R01 grant, grants from DARPA and the Intelligence Advanced Research Projects Activity, and multiple industrial research awards. She received her PhD from the Center for Language and Speech Processing at the Johns Hopkins University.

View the recording >>

Computer Science Seminar Series

December 5, 2024

Abstract: Modern AI and machine learning algorithms can deliver significant performance improvements for decision-making under uncertainty, where traditional, worst-case algorithms are often too conservative. These improvements can be potentially transformative for energy and sustainability applications, where rapid advances are needed to facilitate the energy transition and reduce carbon emissions. However, AI and ML lack worst-case guarantees, hindering their deployment to real-world problems where safety and reliability are critical. In this talk, Nico Christianson will discuss his recent work developing algorithms that bridge the gap between the good average-case performance of AI/ML and the worst-case guarantees of traditional algorithms. In particular, he will focus on the question of how to robustly leverage the recommendations of black-box AI “advice” for general online optimization problems, describing both algorithmic upper bounds and fundamental limits on the tradeoff between exploiting AI and maintaining worst-case performance. He will also highlight some recent steps toward leveraging uncertainty quantification for risk-aware decision-making in these settings, as well as experimental results on energy resource management in high-renewables power grids.

Speaker Biography: Nico Christianson is a final-year PhD candidate in computing and mathematical sciences at the California Institute of Technology, advised by Adam Wierman and Steven Low. Christianson’s research broadly focuses on decision-making under uncertainty, with a specific emphasis on developing new algorithms to enable the reliable and safe deployment of modern AI/ML tools to real-world sustainability challenges such as energy resource operation and carbon-aware computing. His work is supported by an NSF Graduate Research Fellowship and a PIMCO Data Science Fellowship. Christianson has interned at Microsoft Research (Redmond) and collaborated with industry partners including Beyond Limits and Amazon. Previously, he received an AB in applied mathematics from Harvard University.

View the recording >>

Computer Science & Center for Language and Speech Processing Seminar Series

December 2, 2024

Abstract: Over the past decades, machine learning has primarily relied on labeled data, with success often depending on the availability of vast, high-quality annotations and the assumption that test conditions mirror training conditions. In contrast, humans learn efficiently from conceptual explanations, instructions, rules, and contextual understanding. With advancements in large language models, AI systems can now understand descriptions and follow instructions, paving the way for a paradigm shift. This talk explores how teaching machines through language and rules can enable AI systems to gain human trust and enhance their inclusivity, robustness, and ability to learn new concepts. Kai-Wei Cheng will highlight his journey in developing vision-language models capable of detecting unseen objects through rich natural language descriptions. Additionally, he will discuss techniques for guiding the behavior of language models and text-to-image models using language. He will also describe his efforts to incorporate constraints to control language models effectively. Finally, he will conclude his talk by discussing future directions and challenges in building trustworthy language agents.

Speaker Biography: Kai-Wei Chang is an associate professor in the Department of Computer Science at the University of California, Los Angeles and an Amazon Scholar at Amazon AGI. His research focuses on trustworthy AI and multimodal language models. He has published extensively in natural language processing and machine learning, with his work widely recognized through multiple paper awards at top conferences, including Empirical Methods in Natural Language Processing (EMNLP), the Annual Meeting of the Association for Computational Linguistics (ACL), the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, and the Conference on Computer Vision and Pattern Recognition. In 2021, Chang was honored as a Sloan Fellow for his contributions to trustworthy AI as a junior faculty member. Recently, Chang was elected as an officer of the ACL Special Interest Group on Linguistic Data & Corpus-based Approaches to Natural Language Processing, the organizing body behind EMNLP, and will serve as its vice president in 2025 and president in 2026. He is an associate editor for leading journals such as the Journal of Artificial Intelligence Research, the Journal of Machine Learning Research, Transactions of the Association for Computational Linguistics, and the ACL Rolling Review. He also served as an associate program chair at the Thirty-Seventh AAAI Conference on Artificial Intelligence and as senior area chair for most NLP, machine learning, and AI conferences. Since 2021, Kai-Wei has organized five editions of the Trustworthy NLP Workshop at ACL, a platform that fosters research on fairness, robustness, and inclusivity in NLP. Additionally, he has delivered multiple tutorials on topics such as Fairness, Robustness, and Multimodal NLP at EMNLP (2019, 2021) and ACL (2023). Chang received his PhD from the University of Illinois at Urbana-Champaign in 2015 and subsequently worked as a postdoctoral researcher at Microsoft Research in 2016. For more details, visit kwchang.net.

View the recording >>

Computer Science Seminar Series

November 21, 2024

Abstract: In this talk, Eli N. Weinstein studies how to efficiently manufacture samples from a generative model in the real world. He shows how computational methods for approximate sampling can be adapted into experimental design methods for efficiently making samples in the laboratory. He also develops tools to rigorously evaluate the quality of manufactured samples, proposing nonparametric two-sample tests with strong theoretical guarantees and scalable algorithms. He applies these methods to DNA synthesis, since the cost of DNA synthesis is considered a fundamental technological driver of progress of biology and biomedicine. He demonstrates manufacturing ~10^17 samples from a generative model of human antibodies at a sample quality comparable to that of state-of-the-art protein language models. These samples cost roughly a thousand dollars to make (~$10^3), while using previous methods they would cost roughly a quadrillion dollars (~$10^15).  

Speaker Biography: Eli N. Weinstein is a postdoctoral research scientist at Columbia University advised by David Blei. He also serves as the director of machine learning research at Jura Bio, a biotechnology startup focused on genetic medicine. Weinstein’s research is in probabilistic machine learning, with an emphasis on causal inference and experimental design. His applied work focuses on biology, especially therapeutics. He completed his PhD in biophysics at Harvard University in 2022, advised by Debora Marks and Jeffrey Miller and supported by a Hertz Foundation Fellowship. Previously, he received an AB in chemistry and physics with highest honors, also from Harvard, advised by Adam Cohen. His work has been published at venues including the Conference and Workshop on Neural Information Processing Systems, the International Conference on Machine Learning, the International Conference on Artificial Intelligence and Statistics, and the Journal of Machine Learning Research, and has received Best Paper Awards from the New England Statistical Society in 2021 and 2023 and the Molecular Machine Learning Conference in 2024.

View the recording >>

Computer Science Seminar Series

November 21, 2024

Abstract: As AI increasingly permeates our daily lives, it brings immense benefits but also introduces critical privacy challenges, such as data misuse, surveillance, and loss of control. In this talk, Yaxing Yao will share his vision of an ecosystem where AI and privacy are not adversaries, but allies, and mutually reinforce and enrich one another. Rather than treating privacy as an obstacle, this ecosystem positions it as a foundational element of AI’s development, fostering a balanced environment that benefits both technology and society. Yao will introduce three fundamental relationships that drive this symbiotic ecosystem: 1) mutual benefit, wherein AI empowers users to understand and manage their data usage; 2) co-adaptation, wherein AI dynamically adapts to diverse privacy needs across contexts; and 3) ecosystem balance, wherein AI is properly anchored within regulatory frameworks and public policies to ensure users’ privacy. These three relationships redefine AI as a respectful partner in our digital lives—one that supports human-centered values and upholds our autonomy and privacy. Yao will discuss how his research contributes to the advancement of this ecosystem and how he expands its impact to privacy literacy development among families and children via community-based research efforts. Finally, he will discuss challenges and open questions in achieving this vision.

Speaker Biography: Yaxing Yao is an assistant professor in the Department of Computer Science at Virginia Tech. His research lies at the intersection of human-computer interaction, privacy, and accessibility, focusing on exploring privacy issues in user interactions with computing systems and developing solutions to empower users to be aware and control their privacy. He has published in top human-computer interaction venues (e.g., the ACM Conference on Human Factors in Computing Systems, the ACM Conference on Computer-Supported Cooperative Work and Social Computing) and privacy/security venues (e.g., the USENIX Security Symposium, the Symposium on Usable Privacy and Security) and has received multiple paper awards, a Google PSS Faculty Research Award, and two Meta Research Awards. Yao’s work has influenced public policy, including the opt-out icon in the California Consumer Privacy Act. He also founded Kids’ Tech University, a program that engages K-12 students in privacy research through weekly design workshops and summer camps. Yao’s research is generously supported by the NSF, Google, and Meta.

View the recording >>

Computer Science Seminar Series

November 14, 2024

Abstract: Causal knowledge is central to solving complex decision-making problems across engineering, medicine, and cyber-physical systems. Causal inference has been identified as a key capability to improve machine learning systems’ explainability, trustworthiness, and generalization. After a brief introduction to causal modeling, this talk explores two key problems in causal ML. In the first part of the talk, we will focus on the problem of root-cause analysis (RCA), which aims to identify the source of failure in large, modern computer systems. We will show that by leveraging ideas from causal discovery, it is possible to automate and efficiently solve the RCA problem by systematically using invariance tests on normal and anomalous data. In the second part of the talk, we consider causal inference problems in the presence of high dimensional variables, e.g., image data. We show how deep generative models, such as generative adversarial networks and diffusion models, can be used to obtain a representation of the causal system and help solve complex, high-dimensional causal inference problems. This approach enables both causal invariant prediction and evaluation of black box conditional generative models.

Speaker Biography: Murat Kocaoglu received his BS degree in electrical and electronics engineering with a minor in physics from the Middle East Technical University in 2010, his MS from the Koç University in Turkey in 2012, and his PhD from the University of Texas at Austin in 2018. Kocaoglu was a research staff member at the MIT-IBM Watson AI Lab at IBM Research in Cambridge, Massachusetts from 2018 to 2020. He is currently an assistant professor in the Elmore Family School of Electrical and Computer Engineering, the Department of Computer Science (by courtesy), and the Department of Statistics (by courtesy) at Purdue University, where he leads the CausalML Lab. Kocaoglu received an Adobe Data Science Research Award in 2022, an NSF CAREER Award in 2023, and an Amazon Research Award in 2024. His current research interests include causal inference, deep generative models, and information theory.

View the recording >>

Computer Science Seminar Series

November 12, 2024

Abstract: In the age of big data and AI, we are witnessing an erasure of voices from underrepresented communities, a phenomenon that can be described as an “ideocide”—the systematic annihilation of the ethical frameworks and data of marginalized groups. Drawing inspiration from anthropologist Arjun Appadurai’s concept of the “Fear of Small Numbers,” Ishtiaque Ahmed argues that modern AI systems—which overwhelmingly prioritize large datasets—inadvertently silence smaller, non-dominant populations. These systems impose an ethical monoculture shaped by Western neoliberal ideologies, further marginalizing communities who are already underrepresented in data-driven systems. Based on his twelve years of ethnographic and design work with communities in Bangladesh, India, Pakistan, the U.S., Canada, and beyond, Ahmed will explore how the exclusion of these “small data” sets undermines the diversity of ideas and ethics, leading to biased and unjust AI systems. This talk will outline how this silence represents not just a technical gap but a profound ethical failure in AI, one that needs urgent addressing through pluriversal, community-based approaches to AI development. Ahmed will further demonstrate how collaborative, co-designed technologies with marginalized communities can resist ideocide, allowing for the inclusion of multiple ethical and cultural perspectives to create more just, inclusive, and ethical AI systems.

Speaker Biography: Syed Ishtiaque Ahmed is an associate professor of computer science at the University of Toronto and the founding director of the Third Space research group. His research interest is in the intersection between human-computer interaction and artificial intelligence. Ahmed received a PhD and master’s degree from Cornell University, and bachelor’s and master’s degrees from the Bangladesh University of Engineering and Technology. In the last fifteen years, he’s studied and developed successful computing technologies with various marginalized communities in Bangladesh, India, Canada, the U.S., Pakistan, Iraq, Turkey, and Ecuador. Ahmed has published over 100 peer-reviewed research articles and has received multiple Best Paper Awards in top computer science venues including the ACM Conference on Human Factors in Computing Systems, the ACM Conference on Computer-Supported Cooperative Work and Social Computing, the International Conference on Information & Communication Technologies and Development, and the ACM Conference on Fairness, Accountability, and Transparency. He has received numerous honors and accolades, including the International Fulbright Science and Technology Award, the Intel PhD Fellowship, the Institute of International Education Centennial Fellowship, a Schwartz Reisman fellowship, the Walter Massey Fellowship, the Connaught International Scholarship for Doctoral Students, the a Microsoft Research AI & Society fellowship, a Google’s Award for Inclusion Research, and a Meta Research Award. His research has also received generous funding support from all three branches of the Canadian Tri-Council (the Natural Sciences and Engineering Research Council of Canada, the Canadian Institutes of Health Research, and the Social Sciences and Humanities Research Council of Canada), the U.S. NSF and National Institutes of Health, and the Bangladesh Information and Communication Technology Division. Ahmed was named a Future Leader by the Computing Research Association in 2024.

View the recording >>

Computer Science Seminar Series

November 8, 2024

Abstract: The broad agenda of Fei Miao’s work is to develop the foundations for the science of embodied AI—that is, to assure safety, efficiency, robustness, and security of AI systems by integrating learning, optimization, and control. Miao’s research interests span several technical fields, including multi-agent reinforcement learning, robust optimization, uncertainty quantification, control theory, and game theory. Application areas include connected and autonomous vehicles (CAVs), intelligent transportation systems and transportation decarbonization, smart cities, and power networks. Miao’s research experience and current ongoing projects include robust reinforcement learning and control, uncertainty quantification for collaborative perception, game theoretical analysis for the benefit of information sharing for CAVs, data-driven robust optimization for efficient mobile cyber-physical systems (CPS), conflict resolution of smart cities, and resilient control of CPS under attacks. In addition to system modeling, theoretical analysis, and algorithmic design, Miao’s work involves experimental validation in real urban transportation data, simulators, and small-scale autonomous vehicles.

Speaker Biography: Fei Miao is a Pratt & Whitney Associate Professor in the School of Computing and courtesy faculty in the Department of Electrical and Computer Engineering at the University of Connecticut. She is also affiliated with the Pratt & Whitney Institute for Advanced Systems Engineering. Before joining UConn, Miao was a postdoctoral researcher in the General Robotics, Automation, Sensing, & Perception Lab and the Penn Research In Embedded Computing and Integrated Systems Engineering Center with George J. Pappas and Daniel D. Lee in the Department of Electrical and Systems Engineering at the University of Pennsylvania. Miao earned her PhD—as well as the Charles Hallac and Sarah Keil Wolf Award for the best doctoral dissertation—in electrical and systems engineering in 2016, along with a dual Master’s degree in statistics from the Wharton School at the University of Pennsylvania. She received her bachelor’s degree of science from Shanghai Jiao Tong University in 2010 with a major in automation and a minor in finance.

View the recording >>

Computer Science Seminar Series

October 17, 2024

Abstract: Prohibitive pretraining costs makes pretraining research a rare sight—however, this is not the case for analyzing, using, and fine-tuning those models. This talk focuses on one option to improve models in a scientific way, in small measurable steps; specifically, it introduces the concept of merging multiple fine-tuned/parameter-efficient fine-tuning models into one and discusses works tackling what we understand about it, how it works, more up-to-date methods, and how iteratively merging models may allow collaborative continual pretraining.

Speaker Biography: Leshem Choshen is a postdoctoral researcher at the Massachusetts Institute of Technology and IBM who aims to study model development openly and collaboratively, allow feasible pretraining research, and evaluate efficiently. To do so, they co-created model merging, TIES merging, the BabyLM Challenge. They were chosen for postdoctoral Rothschild and Fulbright fellowships and received a Best PhD Thesis Award from the Israeli Association for Artificial Intelligence, as well as a Blavatnik Prize for Computer Science. With broad natural language processing and machine learning interests, Choshen has also worked on reinforcement learning, understanding how neural networks learn, and Project Debater, the first machine system capable of holding a formal debate (as of 2019), which was featured on the cover of Nature.

View the recording >>

Computer Science Seminar Series

October 15, 2024

Abstract: Massive efforts are under way to develop and adapt generative AI to solve any and all inferential and design tasks across engineering and science disciplines. Framing or reframing problems in terms of distributional modeling can bring a number of benefits, but also comes with substantial technical and statistical challenges. Tommi S. Jaakkola’s work has focused on advancing machine learning methods for controlled generation of complex objects, ranging from molecular interactions (e.g., docking) and 3D structures to new materials tailored to exhibit desirable characteristics such as carbon capture. In this talk, Jaakkola will cover a few research vignettes along with their specific challenges, focusing on diffusion and flow models that surpass traditional or alternative approaches to docking, protein design, or conformational ensembles. Time permitting, he will highlight general challenges and opportunities in this area.

Speaker Biography: Tommi S. Jaakkola is the Thomas Siebel Professor of Electrical Engineering and Computer Science in the Massachusetts Institute of Technology’s Department of Electrical Engineering and Computer Science and the MIT Institute for Data, Systems, and Society; he is also an investigator at the MIT Computer Science and Artificial Intelligence Laboratory. He is a fellow of the Association for the Advancement of Artificial Intelligence with many awards for his publications. His research covers how machines can learn, generate, or control and do so at scale in an efficient, principled, and interpretable manner, from foundational theory to modern design challenges. Over the past several years, Jaakkola’s applied work has been focused on molecular modeling and design.

View the recording >>

Computer Science Seminar Series

October 10, 2024

Abstract: Large-scale pretraining has become the standard solution to automated reasoning over text and/or visual perception. But how far does this approach get us to systems that generalize to language use in realistic multi-agent situated interactions? First, Alane Suhr will talk about existing work in evaluating the spatial and compositional reasoning capabilities of current multimodal language models. Then Suhr will talk about how these benchmarks miss a key aspect of real-world situated interactions: joint embodiment. Suhr will discuss how joint embodiment in a shared world supports perspective-taking, an underlooked aspect of situated reasoning, and introduce a new environment and benchmark for studying the influence of perspective-taking on language use in interaction.

Speaker Biography: Alane Suhr is an assistant professor in the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley. Also affiliated with the Berkeley Artificial Intelligence Research Lab, Suhr researches language use and learning in situated, collaborative interactions. This includes developing datasets and environments that support such interactions; designing and evaluating models that participate in collaborative interactions with human users by perceiving, acting, and using language; and developing learning algorithms for training such models from signals acquired in these interactions. Suhr received a BS in computer science and engineering from the Ohio State University in 2016 and a PhD in computer science from Cornell University in 2022.

View the recording >>

Institute for Assured Autonomy & Computer Science Seminar Series

September 17, 2024

Abstract: Despite our tremendous progress in AI, current AI systems—including large language models—still cannot adequately understand humans and flexibly interact with humans in real-world settings. One of the key missing ingredients is Theory of Mind, which is the ability to understand humans’ mental states from their behaviors. In this talk, Tianmin Shu will discuss how we can engineer human-level machine Theory of Mind. He will first show how we can leverage insights from cognitive science studies to develop model-based approaches for physically grounded, multimodal Theory of Mind. He will then discuss how we can improve multimodal embodied AI assistance based on Theory of Mind reasoning. Finally, he will briefly talk about exciting future work toward building open-ended Theory of Mind models for real-world AI assistants.

Speaker Biography: Tianmin Shu is an assistant professor of computer science at the Johns Hopkins University, with a secondary appointment in the university’s Department of Cognitive Science. His research goal is to advance human-centered AI by engineering human-level machine social intelligence to build socially intelligent systems that can understand, reason about, and interact with humans in real-world settings. Shu’s work has received multiple awards, including an Outstanding Paper Award at the 2024 Annual Meeting of the Association for Computational Linguistics and the 2017 Cognitive Science Society Computational Modeling Prize in Perception/Action. His research has also been covered by multiple media outlets, such as New Scientist, Science News, and VentureBeat. Shu received his PhD from the University of California, Los Angeles in 2019. Before joining Johns Hopkins, he was a research scientist at the Massachusetts Institute of Technology.