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 Tuesday from 11 a.m. - 12 p.m. during the Spring 2023 semester.

This week's speaker, Priya L. Donti (Climate Change AI), will be giving a talk titled "Tackling Climate Change with Machine Learning".

Abstract

Climate change is one of the greatest challenges that society faces today, requiring rapid action from all corners. In this talk, I will describe how machine learning can be a potentially powerful tool for addressing climate change, when applied in coordination with policy, engineering, and other areas of action. From energy to agriculture to disaster response, I will describe high-impact problems where machine learning can help through avenues such as distilling decision-relevant information, optimizing complex systems, and accelerating scientific experimentation. I will then dive into some of my own work in this area, which merges data-driven approaches with physical knowledge to facilitate the transition to low-carbon electric power grids. Specifically, I will present a framework called “optimization-in-the-loop machine learning,” and show how it can enable the design of machine learning models that explicitly capture relevant constraints and decision-making processes that are critical to enforce in power grids. I will end by presenting important considerations for developing and deploying work in this area, as well as routes to get involved.

Biography

Priya L. Donti is the Co-founder and Executive Director of Climate Change AI (CCAI), a global non-profit initiative to catalyze impactful work at the intersection of climate change and machine learning. Her research focuses on machine learning for forecasting, optimization, and control in high-renewables power grids. Specifically, my work explores methods to incorporate the physics and hard constraints associated with electric power systems into deep learning workflows.

This week's speaker, Qianwen Wang (Harvard University), will be giving a talk titled, "Interpreting and Steering AI Explanations with Interactive Visualizations".

Abstract

Artificial Intelligence (AI) has advanced at a rapid pace and is expected to revolutionize many biomedical applications. However, current AI methods are usually developed via a data-centric approach regardless of the usage context and the end users, posing challenges for domain users in interpreting AI, obtaining actionable insights, and collaborating with AI in decision-making and knowledge discovery. 

As a visualization researcher, I aim to address this challenge by combining interactive visualizations with interpretable AI. In this talk, I discuss and demonstrate the prospects for interactive visual explanations in the application of biomedical AI via real-world case studies. I present two methodologies for achieving this goal: 1) visualizations that explain AI models and predictions and 2) interaction mechanisms that integrate user feedback into AI models. Despite some challenges, I will conclude on an optimistic note: interactive visual explanations should be indispensable for Human-AI collaboration in biomedical applications. The methodology discussed can be applied generally to other applications where human-AI collaborations are involved, assisting domain experts in data exploration and insight generation with the help of AI.

Biography

Qianwen Wang is a Postdoctoral Fellow at Harvard University. Her research strives to facilitate the communication and collaboration between users and AI through interactive visual tools, with a special interest in their applications in solving biomedical challenges.

Her research has made contributions to visualization, human-computer interaction, and bioinformatics, as demonstrated by 18 publications in top-tier venues (IEEE VIS, TVCG, ACM CHI, Bioinformatics, ISMB). She has received two best abstract awards from BioVis ISMB, one honorable mention from IEEE VIS, and one Best paper award from IMLH@ICML. She is an awardee of the HDSI Postdoctoral Research Fund. Her Research has been covered by MIT News and Nature Technology Features. She serves as the abstract chair for the ISMB BioVis COSI, the Poster Chair for PacificVis, and an organizer for Visualization in Biomedical AI workshop@IEEE VIS.

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 Tuesday from 11 a.m. - 12 p.m. during the Spring 2023 semester.

This week's speaker, Zilin Li (Indiana University), will be giving a talk titled "STAARpipeline: an all-in-one rare-variant analysis tool for biobank-scale whole-genome sequencing data".

Abstract

Large-scale whole-genome sequencing (WGS) studies have enabled the analysis of rare variant associations with complex human diseases and traits. Variant set analysis is a powerful approach to studying rare variant associations. However, existing methods have limited ability to define the variant set in the genome, especially for the noncoding genome. We propose a computationally efficient and robust rare variant association-detection framework, STAARpipeline, to automatically annotate a WGS study and perform flexible rare variant association analysis, including gene-centric analysis and fixed-window and dynamic-window-based non-gene-centric analysis by incorporating variant functional annotations. In gene-centric analysis, STAARpipeline groups coding and noncoding variants based on functional categories of genes and incorporate multiple functional annotations. In non-gene-centric analysis, in addition to fixed-size sliding window analysis, STAARpipeline provides a data-adaptive-size dynamic window analysis. All these variant sets could be automatically defined and selected in STAARpipeline. STAARpipeline also provides analytical follow-up of dissecting association signals independent of known variants via conditional analysis. We applied the STAARpipeline to analyze the total cholesterol in 30,138 samples from the NHLBI Trans-Omics for Precision Medicine (TOPMed) Program. All analyses scale well in computation time and memory. We discover several potentially new significant associations with lipids. In summary, STAARpipeline is a powerful and resource-efficient tool for association analysis of biobank-scale WGS studies.

Biography

Zilin Li is an Assistant Professor in the Department of Biostatistics and Health Data Science at Indiana University School of Medicine. Before being an assistant professor, he was a research scientist, research associate, and postdoctoral research fellow in Professor Xihong Lin’s lab in the Department of Biostatistics at Harvard T.H. Chan School of Public Health. He received my Ph.D. from Tsinghua University in 2016, supervised by Professor Xihong Lin. His research interests lie in statistical genetics and high-dimensional statistics with applications for analyzing massive health data, especially developing statistical methods for scalable analysis of large-scale genetics and genomics data.

This week's speaker, Christina Boucher (University of Florida), will be giving a talk titled, "Building Scalable Indexes That Can Be Efficiently Queried".

Abstract

Recently, Gagie et al. proposed a version of the FM-index, called the r-index, that can store thousands of human genomes on a commodity computer. We later showed how to build the r-index efficiently via a technique called prefix-free parsing (PFP) and demonstrated its effectiveness for exact pattern matching. Exact pattern matching can be leveraged to support approximate pattern matching but the r-index itself cannot support efficiently popular and important queries such as finding maximal exact matches (MEMs). To address this shortcoming, Bannai et al. introduced the concept of thresholds, and showed that storing them together with the r-index enables efficient MEM finding --- but they did not say how to find those thresholds. We present another novel algorithm that applies PFP to build the r-index and find the thresholds simultaneously and in linear time and space with respect to the size of the prefix-free parse. Our implementation can rapidly find MEMs between reads and large sequence collections of highly repetitive sequences. Compared to existing methods, ours used 2 to 11 times less memory and was 2 to 32 times faster for index construction. Moreover, our method was less than one thousandth the size of competing indexes for large collections of human chromosomes.

Biography

Dr. Boucher is an Associate Professor in the Department of Computer and Information Science and Engineering at the University of Florida. She has over 125 publications in bioinformatics, with over several dozens of them in succinct data structures and/or alignment. She has given keynote addresses at 2022 WABI Pangenomics workshop, HICOMB 2022. IGGSY 2022, SPIRE 2021, RECOMB-SEQ 2016 and the ECCB 2016 Workshop on Pan-Genomics.  She is a recipient of an ESA 2016 Best Paper Award. She oversees the development and maintenance of several software methods, including Moni, MEGARes and AMRPlusPlus, METAMarc, Kohdista, Vari, VariMerge — and most recently, Moni. In addition, she has built a team of collaborators in various biomedical sciences including microbiology, veterinarian medicine, epidemiology, public health, and clinical sciences.  Her lab
receives funding from NIH, NSF, and USDA. In addition, she actively works on increasing the diversity in bioinformatics education. Her efforts include being a member of the University of Florida’s Implicit Bias committee, being a panellist for the NSF-funded ACM BCB 2015 Women in Bioinformatics meeting, serving as a faculty advisor for an ACM-W chapter, and being an active member of the Diversity Committee for over three years. She also received a fellowship from The Institute for Learning and Teaching (TILT) for her course redevelopment and served on the advisory committee for an NSF Research Traineeships Program. She was the PC chair for several conferences, including WABI 2022, SPIRE 2020, RECOMB-SEQ 2019, and ACM-BCB 2018.  Most recently, she was nominated to serve on the NIH BDMA Study Section as a Standing Member, and a member of the Executive Board of ACM SIG BIO.

This week's speaker, Jing Ma (University of Virginia), will be giving a talk titled, "When Causal Inference Meets Graph Machine Learning: Unleashing the Potential of Mutual Benefit".

Abstract

Recent years have witnessed rapid development in graph-based machine learning (ML) in various high-impact domains (e.g., healthcare, recommendation, and security), especially those powered by effective graph neural networks (GNNs). Currently, the mainstream graph ML methods are based on statistical learning, e.g., utilizing the statistical correlations between node features, graph structure, and labels for node classification. However, statistical learning has been widely criticized for only capturing the superficial relations between variables in the data system, and consequently, rendering the lack of trustworthiness in real-world applications. For example, ML models often make biased predictions toward underrepresented groups. Besides, these ML models often lack explanation for human. Therefore, it is crucial to understand the causality in the data system and the learning process. Causal inference is the discipline that investigates the causality inside a system, for example, to identify and estimate the causal effect of a certain treatment (e.g., wearing a face mask) on an important outcome (e.g., COVID-19 infection). Involving the concepts and philosophy of causal inference into ML methods is often considered as a significant component of human-level intelligence and can serve as the foundation of artificial intelligence (AI).  However, most traditional causal inference studies rely on strong assumptions, and focus on independent and identically distributed (i.i.d.) data, while causal inference on graphs is faced with many barriers in effectiveness. Fortunately, the interplay between causal inference and graph ML has the potential to bring mutual benefit to each other. In this talk, we will present the challenges and our contributions for bridging the gap between causal inference and graph ML, mainly including two directions: 1) leveraging graph ML methods to facilitate causal inference in effectiveness; and 2) leveraging causality to facilitate graph ML models in model trustworthiness (e.g., model fairness and explanation).

Biography

Jing Ma is a Ph.D. candidate in the Department of Computer Science at University of Virginia, under the supervision of Dr. Jundong Li and Dr. Aidong Zhang. She received her B.Eng. degree and M.Eng. degree at Shanghai Jiao Tong University with Outstanding Graduate Award. Her research interests broadly cover machine learning and data mining, especially include causal inference, graph mining, fairness, trustworthiness, and AI for social good. Her recent work focuses on bridging the gap between causality and machine learning. Her research papers have been published in top conferences and journals such as KDD, NeurIPS, IJCAI, WWW, AAAI, TKDE, WSDM, SIGIR, ECML-PKDD, AI Magazine, and IPSN. She has rich internship experience in companies and academic organizations such as Microsoft Research. She has won some important awards such as SIGKDD 2022 Best Paper Award and CAPWIC 2022 Best Poster Award.

BICB Colloquium Faculty Nomination Talks: Join us in person on the UMR campus in room 414, on the Twin Cities campus in MCB 2-122 or virtually at 5 p.m.

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 Tuesday from 11 a.m. - 12 p.m. during the Spring 2023 semester.

This week's speaker, Zhi Ding (ECE, UC Davis), will be giving a talk titled "Non-Blackbox Deep Learning for Massive MIMO Wireless Communication Systems".

Abstract

The proliferation of advanced wireless services, such as virtual reality, autonomous driving, and internet of things, has generated increasingly intense pressure to develop intelligent wireless communication systems to meet networking needs posed by extremely high data rates, massive numbers of connected devices, and ultra-low latency. Deep learning (DL) has been recently emerged as an exciting design tool to advance the development of wireless communication system with some demonstrated successes. In this tutorial, we review the principles of applying DL for improving wireless network performance by integrating the underlying characteristics of channels in practical massive MIMO deployment. We introduce important insights derived from the physical RF channel properties and present a comprehensive overview on the application of DL for accurately estimating channel state information (CSI) of forward channels with low feedback overhead. We provide examples of successful DL application in CSI estimation for massive MIMO wireless systems and highlight several promising directions for future research.

Biography

Dr. Zhi Ding (S'88-M'90-SM'95-F'03, IEEE) holds the position of Distinguished Professor of Electrical and Computer Engineering at the University of California, Davis. He received his Ph.D. degree in Electrical Engineering from Cornell University in 1990. From 1990 to 2000, he was a faculty member of Auburn University and later, the University of Iowa. He has coauthored over 400 technical papers and two books. Dr. Ding is a coauthor of the text: Modern Digital and Analog Communication Systems, 4^th edition and 5^th edition, Oxford University Press.

Dr. Ding is a Fellow of IEEE and has been an active member of IEEE, serving on technical programs of several workshops and conferences. He served both as a Member and also the Chair of the IEEE Transactions on Wireless Communications Steering Committee from 2007-2001. Dr. Ding was the Technical Program Chair of the 2006 IEEE Globecom and the General Chair of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). He served as an IEEE Distinguished Lecturer (Circuits and Systems Society, 2004-06, Communications Society, 2008-09). He received the 2012 Wireless Communications Recognition Award and the 2020 Education Award from the IEEE Communications Society.

This week's speaker, Yogatheesan Varatharajah (University of Illinois), will be giving a talk titled, "Trustworthy Machine Learning for Health via Domain-guided Modeling: The Case for Neurological Diseases".

Abstract

Recent advances in wearables, brain implants, and sensing technology have enabled us to design systems that continuously monitor patients' brain health and ascertain individualized treatments for neurological diseases. However, there is a lack of efficient methods that translate continuous physiological data streams into meaningful biological models of underlying diseases, relate them to existing clinical knowledge and biomarkers, and provide actionable treatment parameters. Machine learning (ML) holds great promise in tackling these challenges; however, the mainstream black-box-ML approaches have proven to be untrustworthy because of label inconsistencies, spurious correlations, and the lack of deployment robustness. My goal is to ensure trustworthiness in ML for healthcare, particularly neurology, via a novel framework known as “Domain-guided Machine Learning” or “DGML” that merges machine learning with clinical domain expertise. In this talk, I will discuss the need for trustworthy ML in healthcare, how to leverage clinical domain knowledge to engineer trustworthy ML models, and several real-world applications of DGML in neurological care and decision making.

Biography

Dr. Yoga Varatharajah is currently a Research Assistant Professor in the Department of Bioengineering at the University of Illinois at Urbana-Champaign and a Visiting Scientist at the Mayo Clinic, Rochester. He obtained his Ph.D. in Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign under the supervision of Prof. Ravishankar Iyer. Over the past seven years, he has been working closely with domain experts at Mayo Clinic and Cleveland Clinic to develop, evaluate, and deploy domain-guided ML models to inform clinical decisions related to neurological diseases. His research has been published at reputed engineering conferences (e.g., Neurips, ML4H, BIBM, ISBI, EMBC, NER) and medical journals (e.g., Scientific Reports, Journal of Neural Engineering, Brain Communications, Epilepsia, Neuroimage), has contributed to an ongoing clinical trial in neuromodulation for epilepsy, and has resulted in a joint patent between Mayo and Illinois. He also received several honors, including a CSL Ph.D. Thesis Award, a Mayo-Clinic-Illinois Alliance Fellowship, an American Epilepsy Society Young Investigator Award, an NSF CRII Research Initiation Award, and several best paper awards and nominations.