2/20 -“Planning and Design for Minimalist Robots” - Jason O’Kane, Ph.D.
When - Thurs, Feb 20th 11:30 am
Where - Woodward 154
Jason M. O'Kane
Abstract
The usefulness of mobile robots operating in the physical world depends on how effectively they can sense and move through their environments. Unfortunately, sensors provide only limited(and sometimes incorrect) information and actuators provide constrained(and sometimes unreliable) movement capabilities. This reality motivates a careful study of the information requirements of the problems our robots intend to solve. In this talk, I will present two related lines of research in this direction. First, I will present a series of results that use careful planning over the robot's information space to find plans that succeed in spite of strong sensor limitations, for a variety of tasks. Second, I will present some recent research that seeks to automate some parts of this process, by representing models for a robot's interaction with the world as formal, algorithmically-manipulable objects, and posing various kinds of questions on those data structures. The results include both both bad news(i.e., hardness results) and good news(practical algorithms).
Bio
Jason M. O'Kane is Professor of Computer Science and Engineering, Associate Chair for Academics in the Department of Computer Science and Engineering, and Director of the Center for Computational Robotics at the University of South Carolina. He holds the Ph.D. and M.S. degrees from the University of Illinois at Urbana-Champaign and the B.S. degree from Taylor University, all in Computer Science. He has won the CAREER Award from NSF, the Breakthrough Star Award and the Distinguished Research Service Award from Office of Research at the University of South Carolina, the Outstanding Senior Researcher Award and the Most Valuable Professor Award from his home department, the Professor for Student Affordability Award from the University of South Carolina Student Government, and the Outstanding Graduate in Computer Science Award from Taylor University. He serves as Associate Editor for the IEEE Transactions on Robotics and the IEEE Robotics and Automation Letters. His research spans algorithmic robotics, planning under uncertainty, and computational geometry.
Fairness in AI systems is receiving increasing attention. Fairness-aware machine learning studies the problem of building machine learning models that are subject to fairness requirements. In this talk, I will present my dissertation research on developing a causality-based framework for measuring discrimination and achieving fairness in classification. In our research, we formulate discrimination based on the causal inference framework where the causal effect is measured from a causal graph and observed data. We propose a unified definition that covers most of previous causality-based fairness notions, namely the path-specific counterfactual fairness(PC fairness). We target an inherent challenge in causal inference, unidentification, which means some causal effects cannot be uniquely computed from observed data. To overcome this challenge, we propose novel estimation methods to bound the unidentifiable fairness quantities. Then we develop an efficient post-processing method to achieve fairness in unidentifiable counterfactual cases. At the end of my talk, I will briefly introduce my other works dealing with discrimination issues in various machine learning tasks and applications. I will also discuss future research directions on fairness-aware machine learning and FATE(Fairness, Accountability, Transparency, and Explainability) in AI.
Bio
Yongkai Wu is a Ph.D. candidate in the Department of Computer Science and Computer Engineering at the University of Arkansas. He received his B.Eng. degree in Electronic Engineering from Tsinghua University, China in 2014. His research interests focus on machine learning, data mining, and artificial intelligence, particularly fairness-aware machine learning and causal inference. His publications have appeared in prestigious conferences including IJCAI, KDD, NeurIPS, WWW, and a premier journal TKDE. He has served as a PC member for several international conferences including AAAI, IJCAI, KDD, PAKDD.
2/24 -“An NLP Perspective on Offensive Content in Social Media” - Marcos Zampieri, Ph.D
When - Monday, Feb 21st 11-12pm
Where - CHHS 281
Marcos Zampieri
Abstract
Offensive language is pervasive in social media. Individuals frequently take advantage of the perceived anonymity of computer-mediated communication, using this to engage in behavior that many of them would not consider in real life. Online communities, social media platforms, and technology companies have been investing heavily in ways to cope with offensive content to prevent abusive behavior in social media. One of the most effective strategies for tackling this problem is to use computational methods to identify offense, aggression, and hate speech in user-generated content(e.g. posts, comments, microblogs, etc.). In this talk, I discuss some of the challenges of using NLP to recognize offensive content online. I present a new taxonomy that my colleagues and I have created to annotate offensive language datasets. The challenges of collecting and annotated multilingual datasets for offensive language identification are also discussed. Finally, I present the set-up and the results of the two editions of the OffensEval competition hosted at SemEval-2019 and SemEval-2020(https://sites.google.com/site/offensevalsharedtask/home).
Bio
Marcos Zampieri is an assistant professor at the Rochester Institute of Technology in Rochester, NY where he leads the Language Technology Lab. He obtained his PhD from Saarland University in Germany with a thesis on computational approaches to language variation. He has previously held research and teaching positions in Germany and the UK. Marcos published papers on many topics in Computational Linguistics and Natural Language Processing such as language acquisition and variation, offensive language identification, and machine translation. Since 2014, he is the main organizer of the workshop series on NLP for Similar Languages, Varieties and Dialects(VarDial) co-located yearly with international top-tier NLP conferences such as COLING, EACL, and NAACL. He has co-edited a volume on the same topic to appear at the series Studies in Natural Language Processing by Cambridge University Press.
2/25 -“Machine learning for images in space and time: towards real-word applications” - Yingying Zhu, Ph.D
When - Tuesday, Feb 25th 11:30-12:30pm
Where - CHHS 128
Yingying Zhu
Abstract
With the rapid growth in the number of imaging procedures, digitization of images, internet explosion and healthcare’s inexorable digital migration, we are facing the challenges of developing machine learning algorithms to analyze these images efficiently and improve current healthcare system workflow. To address these challenges, I have been working on interdisciplinary research of machine learning, medical imaging analysis and computer vision. I would like to present my work in these areas over the years. Here are the topics my presentation will cover: 1. Medical imaging analysis: 1a) Progressive neurodegenerative disease early prediction using bayesian inference model on longitudinal images. 1b) Modelling dynamic brain connectome from noisy functional imaging data with application on brain disorder diagnosis. 1c) Combining multi-modal imaging data(labelled and unlabelled) in a semi-supervised graph model to reduce the imaging labelling expense and improve diagnosis performance. 1d) Cross-domain medical imaging segmentation on non-contrast CT and contrast enhanced CT by shared latent union of subspaces imaging translation. 1e) Interpreting chest x-ray by decomposing the abnormal chest x-ray images into a normal image and abnormal regions using adversary imaging synthesis framework. 2. Computer Vision. Jointly reconstruction and segmentation of human activities in 3D from 2D videos using convolutional sparse coding and low rank subspace clustering with potential application for patients surveillance. I will conclude my talk with the connection/difference between computer vision and medical imaging analysis, and future research plan.
Bio
Dr. Yingying Zhu is currently a staff scientist leading a team of postdoc fellows, intern students and post-bachelor fellows to collaborate with radiologists on developing machine learning models for real-world clinical problems in clinical center, NIH. Her research lies on the intersection of machine learning, computer vision and medical imaging analysis. She graduated from University of Queensland, Australia with a PhD in Computer Science under the supervision of Profess Simon Lucey(currently research associate professor in CMU, Pittsburgh). Her PhD thesis focusing on computer vision and machine learning. After PhD study, she started to work as a postdoc fellow in UNC chapel hill and Cornell University on medical imaging data analysis. She has published many papers in top computer vision and medical imaging conferences including CVPR, MICCAI, IPMI and also papers on top journals including TPAMI, TMI, MIA, Neuroimage. She served as managing guest editors for learning journals including Pattern Recognition Letters, Sensors and Multimedia Tools and Applications. She also co-authored a book: Connectomics: Applications to Neuroimage, published by Elsevier.
2/26 -“Fingerprinting Human Expression for Computational Health, Wellness, and Creativity” - Mary Pietrowicz, Ph.D
Summary
2/20 - “Planning and Design for Minimalist Robots” - Jason O’Kane, Ph.D.
2/21 - “Achieving Causal Fairness in Machine Learning” - Yongkai Wu, Ph.D Candidate
2/24 - “An NLP Perspective on Offensive Content in Social Media” - Marcos Zampieri, Ph.D
2/25 - “Machine learning for images in space and time: towards real-word applications” - Yingying Zhu, Ph.D
2/26 - “Fingerprinting Human Expression for Computational Health, Wellness, and Creativity” - Mary Pietrowicz, Ph.D