Past events

The UMN Machine Learning Seminar Series brings together faculty, students, and local industrial partners who are interested in the theoretical, computational, and applied aspects of machine learning, to pose problems, exchange ideas, and foster collaborations. The talks are every Thursday from 12 p.m. - 1 p.m. during the Fall 2021 semester.

This week's speaker, Irfan Bulu (UnitedHealth Group), will be giving a talk titled "How to improve healthcare AI? Incorporating multimodal data and domain knowledge."

Abstract

Healthcare data is special. Its complex nature is a double-edged sword, possessing great potential but also presenting many difficulties to overcome. For example, administrative claims data —in contrast to the common data types (text, vision, audio) where AI has made eye-popping advances—is multi-modal (i.e., consisting of distinct data types including medical claims, pharmacy claims, and lab results), asynchronous (medication histories and diagnosis histories need not be aligned in time), and irregularly sampled (we only collect data when an individual interacts with the system). Along with such rich and complex data, there is a great deal of domain knowledge in various forms in the healthcare field. In this talk, I will present our work on deep learning architectures for incorporating multimodal data and domain knowledge into models.

Biography

I received a Ph.D. in Physics from Bilkent University in 2007. The focus of my Ph.D. work was novel structures such as photonic crystals, plasmonic devices, and metamaterials for controlling the flow of light. I joined Prof. Marko Loncar’s lab at Harvard University for postdoc after completing Ph.D. There, I tackled problems and challenges in communication security and communication bandwidth using diamond nano-photonic structures. In 2013, I took a career in industrial research at Schlumberger, the largest oil field services company, which started an exciting journey for me in taking innovations from lab to products at the hands of customers. For example, our team invented a new nuclear magnetic resonance logging tool, which improved logging speed by an order of magnitude, thereby addressing an important challenge for our customers in adopting nuclear magnetic resonance measurements. This work also led me to a career in machine learning as both the design of the instrument and interpretation of various measurements in oil field benefited from advances in deep learning. I joined United Health Group in 2018, where I research machine learning algorithms for healthcare applications.

Have all your questions about the Master of Science in Software Engineering (MSSE) program answered by attending this online information session.

RSVP now to reserve your spot.

Attendees will be sent a link prior to the event.
 

The computer science colloquium for Monday, October 25 has been canceled.

The colloquia series will resume Monday, November 1 at 11:15 a.m.

Alan Jeevanathan will be giving a talk titled "Cryptocurrency Craziness: Gamblers' paradise? Or decentralized currencies that governments and banks can't get their fingers on?"

Abstract

Cryptocurrencies have emerged as important financial software systems. Bitcoin and Ethereum have matured from being associated with extremists and techies to being considered by governments as a technology to implement digital money. El Salvador recently announced Bitcoin as a form of accepted currency and already have a month a 3rd of the population is actively spending this digital money. This is both nothing like how we conventionally spend our dollars. But with all the volatility and uncertainty as well as a lack of a central authority, where exactly are we headed? Through my own experiences of investing, buying/selling crypto, being involved in a startup as well setting up a bunch of servers mining different cryptocurrencies I will share what I know, what I've learned and where I think we're going. If you've totally unfamiliar with crypto, this will be a good seminar to help you understand if you want to dip a toe in.

Biography

I've been a techie ever since my dad stuck a Commodore Vic-20 in front of me on my 5th birthday. As a software engineer, I've been particularly passionate about the software development process and how it's evolved over the past 20 years.

I've taken this love of learning and been on a journey working in different industries from medical device to retail to legal to big data to trucking to insurance, looking for new and interesting problems to tackle and using what I've learned in some areas and applying them to others.

This week's speaker, Catherine Zhao, will be giving a talk titled "Attention in Vision-based AI Systems."

Abstract

Imagine that you are at a bus stop in a new city. You take a few glimpses around, parse and summarize the information you gather, and decide the next steps. Although intuitive, it implies a highly sophisticated and superior ability to select and parse information. Our research is along this line to develop and utilize machine attention for AI systems. In this talk, Professor Catherine Zhao will discuss the challenges and share the recent innovations in data, models, and applications from our research.

Zhao will first talk about attention prediction - the ability of machines to find the most relevant information. She will elaborate on our computational models and experimental methods for attention prediction and explain how they have advanced the state-of-the-art. She will then discuss new approaches that leverage attention in computer vision and language tasks, leading to better interpretability and task performance. She will also present preliminary data suggesting that this approach can help reveal and improve the black-box decision-making process of learning-based AI systems. Finally, Zhao will discuss the applications of our models and data in healthcare and will give two examples where our work leads to the discovery of neurobehavioral signature in autism patients, as well as cutting-edge brain-machine interface technology that restores the lost motor function in upper-limb amputee patients.

Biography

Catherine Qi Zhao is an associate professor in the Department of Computer Science & Engineering at the University of Minnesota. Dr. Zhao’s research interests are in computer vision and machine learning, and their applications in healthcare. Her current research on machine attention is supported by NSF and NIH. Dr. Zhao has published more than 100 papers in peer reviewed conferences and journals. She is an associate editor at the IEEE Transactions on Neural Networks and Learning Systems and the IEEE Transactions on Multimedia, a program chair at WACV '2022, and an area chair at CVPR and other computer vision and AI venues.

The UMN Machine Learning Seminar Series brings together faculty, students, and local industrial partners who are interested in the theoretical, computational, and applied aspects of machine learning, to pose problems, exchange ideas, and foster collaborations. The talks are every Thursday from 12 p.m. - 1 p.m. during the Fall 2021 semester.

This week's speaker, Song Mei (University of California, Berkeley), will be giving a talk titled "The efficiency of kernel methods on structured datasets."

Abstract

Inspired by the proposal of tangent kernels of neural networks (NNs), a recent research line aims to design kernels with a better generalization performance on standard datasets. Indeed, a few recent works showed that certain kernel machines perform as well as NNs on certain datasets, despite their separations in specific cases implied by theoretical results. Furthermore, it was shown that the induced kernels of convolutional neural networks perform much better than any former handcrafted kernels. These empirical results pose a theoretical challenge to understanding the performance gaps in kernel machines and NNs in different scenarios.

In this talk, we show that data structures play an essential role in inducing these performance gaps. We consider a few natural data structures, and study their effects on the performance of these learning methods. Based on a fine-grained high dimensional asymptotics framework of analyzing random features models and kernel machines, we show the following: 1) If the feature vectors are nearly isotropic, kernel methods suffer from the curse of dimensionality, while NNs can overcome it by learning the best low-dimensional representation; 2) If the feature vectors display the same low-dimensional structure as the target function (the spiked covariates model), this curse of dimensionality becomes milder, and the performance gap between kernel methods and NNs become smaller; 3) On datasets that display some invariance structure (e.g., image dataset), there is a quantitative performance gain of using invariant kernels (e.g., convolutional kernels) over inner product kernels. Beyond explaining the performance gaps, these theoretical results can further provide some intuitions towards designing kernel methods with better performance.

Biography

Song Mei is an Assistant Professor in statistics at UC Berkeley. His research is motivated by data science and lies at the intersection of statistics, machine learning, information theory, and computer science. His work often builds on insights that originate within statistical physics literature. His recent research interests include theory of deep learning, high dimensional geometry, approximate Bayesian inferences, and applied random matrix theory.

The John V. Carlis Memorial Lecture is dedicated to the advancement of education and inclusion in the field of computing.

This year's speaker is Juan Gilbert from University of Florida, giving a talk titled "A Model for Building Diversity, Equity and Community in Computing".

Abstract

The Computer & Information Science & Engineering (CISE) Department at the University of Florida is one of, if not the most, diverse computer science departments in the nation. CISE has the nation's largest group of African-American faculty and PhD students. CISE is also in the top with respect to women tenure-track faculty. In this John V. Carlis Memorial Lecture, Dr. Juan Gilbert will discuss how this unprecedented diversity was accomplished and how it can be replicated. He will also share data on the state of African-Americans in computing.

Biography

Juan E. Gilbert received his MS and PhD degrees in computer science from the University of Cincinnati in 1995 and 2000, respectively. He also received his BS in Systems Analysis from Miami University in Ohio in 1991. Dr. Gilbert is currently the Andrew Banks Family Preeminence Endowed Professor and Chair of the Computer & Information Science & Engineering Department at the University of Florida where he leads the Human Experience Research Lab. He has research projects in election security/usability/accessibility, advanced learning technologies, usability and accessibility, Human-Centered AI/machine learning and Ethnocomputing (Culturally Relevant Computing). He is an ACM Fellow, a Fellow of the American Association of the Advancement of Science and a Fellow of the National Academy of Inventors. In 2012, Dr. Gilbert received the Presidential Award for Excellence in Science, Mathematics, and Engineering Mentoring from President Barack Obama. He also received the American Association for the Advancement of Science (AAAS) 2014 Mentor Award. He received the 2021 ACM SIGCHI Social Impact Award. Dr. Gilbert received the 2018 Computer Research Association's A. Nico Habermann Award. Dr. Gilbert has served on 3 National Academies consensus committees, "The Role of Authentic STEM Learning Experiences in Developing Interest and Competencies for Technology and Computing", "The Science of Effective Mentoring in Science, Technology, Engineering, Medicine, and Mathematics (STEMM)" and "The Future of Voting: Accessible, Reliable, Verifiable Technology"

The application deadline for the computer science integrated program (Bachelor's/Master's) is October 15.

This is exclusively available to students officially admitted to the College of Science & Engineering Bachelor’s of Science in Computer Science, Bachelor’s of Computer Engineering, the College of Liberal Arts Bachelor’s of Arts in Computer Science, and the College of Liberal Arts Second Major in Computer Science. The program allows students with strong academic performance records to take additional credits (up to 16 credits) at undergraduate tuition rates during their last few semesters which can be applied towards the Computer Science M.S. program.

Applicants must have at least 75 credits completed at the time of their application. Read more about the program eligibility requirements.

Applications must be submitted online. Before applying, students should review the application procedures.

Students will be notified of the outcome of their application via email by December 1 for a spring start. In some cases, an admission decision will be put on hold until semester grades are finalized. Students will be notified if their application is on hold.

The Minnesota Natural Language Processing (NLP) Seminar is a venue for faculty, postdocs, students, and anyone else interested in theoretical, computational, and human-centric aspects of natural language processing to exchange ideas and foster collaboration. The talks are every other Friday from 12 p.m. - 1 p.m. during the Fall 2021 semester.

This week's speaker, Philippe Laban (Salesforce Research), will be giving a talk titled "Pushing the Boundary of Unsupervised Text Generation."

Abstract

Recent progress in automated text generation relies predominantly on the use of large datasets, sometimes requiring millions of examples for each application setting. In the first part of the talk, we'll develop novel text generation methods that balance the goals of fluency, consistency, and relevancy without requiring any training data. We focus on text summarization and simplification by directly defining a multi-component reward, and training text generators to optimize this objective. The novel approaches that we introduce perform better than all existing unsupervised approaches and in many cases outperform those that rely on large datasets, showing that high-performing NLP models are possible when little data is available.

Biography

Philippe Laban is a Research Scientist at Salesforce Research, where he works on text generation projects, including summarization and interactive question answering. Previously, he obtained his Ph.D. from UC Berkeley, where he was advised by Marti Hearst and John Canny. His work in Berkeley focused on designing unsupervised methods for text generation and on building and adapting NLP techniques to a very large, noisy and evolving news dataset. He did his undergraduate education at Georgia Tech, doing research in signal processing and discrete mathematics.

The UMN Machine Learning Seminar Series brings together faculty, students, and local industrial partners who are interested in the theoretical, computational, and applied aspects of machine learning, to pose problems, exchange ideas, and foster collaborations. The talks are every Thursday from 12 p.m. - 1 p.m. during the Fall 2021 semester.

This week's speaker, Uday V. Shanbhag (Penn State), will be giving a talk titled "Probability Maximization via Minkowski Functionals: Convex Representations and Tractable Resolution."

Abstract

In this paper, we consider the maximization of a probability P{ζ∣ζ∈K(x)} over a closed and convex set X, a special case of the chance-constrained optimization problem. We define K(x) as K(x)≜{ζ∈K∣c(x,ζ)≥0} where ζ is uniformly distributed on a convex and compact set K and c(x,ζ) is defined as either {c(x,ζ)≜1−|ζTx|m, m≥0} (Setting A) or c(x,ζ)≜Tx−ζ (Setting B). We show that in either setting, P{ζ∣ζ∈K(x)} can be expressed as the expectation of a suitably defined function F(x,ξ) with respect to an appropriately defined Gaussian density (or its variant), i.e. Ep~[F(x,ξ)]. We then develop a convex representation of the original problem requiring the minimization of g(E[F(x,ξ)]) over X where g is an appropriately defined smooth convex function. Traditional stochastic approximation schemes cannot contend with the minimization of g(E[F(⋅,ξ)]) over X, since conditionally unbiased sampled gradients are unavailable. We then develop a regularized variance-reduced stochastic approximation ({\textbf{r-VRSA}}) scheme that obviates the need for such unbiasedness by combining iterative {regularization} with variance-reduction. Notably, ({\textbf{r-VRSA}}) is characterized by both almost-sure convergence guarantees, a convergence rate of O(1/k1/2−a) in expected sub-optimality where a>0, and a sample complexity of O(1/ϵ6+δ) where δ>0.

Biography

Uday V. Shanbhag has held the Gary and Sheila Bello Chaired professorship in Ind. & Manuf. Engr. at Penn State University (PSU) since Nov. 2017 and has been at PSU since Fall 2012, prior to which he was at the University of Illinois at Urbana-Champaign (between 2006–2012, both as an assistant and a tenured associate professor). His interests lie in the analysis and solution of optimization problems, variational inequality problems, and noncooperative games complicated by nonsmoothness and uncertainty. He holds undergraduate and Master’s degrees from IIT,Mumbai (1993) and MIT, Cambridge (1998) respectively and a Ph.D. in management science and engineering (Operations Research) from Stanford University (2006).