Past events

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Larry Jackel (North C Technologies, Inc.), will be giving a talk titled "Taking the Long View: From Early Neural Nets to Self-Driving Cars".

Abstract

This talk presents an overview of the path from early research in Neural Networks at Bell Labs in Holmdel NJ that led to a current approach for enabling self-driving cars. We will discuss various DARPA programs that advanced autonomous driving, and an unconventional method that learned how to steer by observing human drivers. We will also discuss obstacles to achieve Level 5 autonomy, that is, driving as well as a human in all conditions. The talk will be at a high level, with numerous illustrative videos.

Biography

Larry Jackel is President of North C Technologies and has been consulting with NVIDIA since 2015. For most of his scientific career Jackel was a manager and researcher at Bell Labs and then AT&T Labs. He has created and managed research groups in Microscience and Microfabrication, in Machine Learning and Pattern Recognition, and in Carrier-Scale Telecom Services. Jackel’s Bell Labs department was instrumental in pioneering much of the machine learning technology in use today. From 2003-2007 Jackel  was a DARPA Program Manager where he conceived and managed programs in Autonomous Ground Robot Navigation and Locomotion. Jackel holds a PhD in Experimental Physics from Cornell University with a thesis in superconducting electronics. He is a Fellow of the American Physical Society and the IEEE.

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 2 - 3 p.m. during the fall 2022 semester.

This week's speaker, Maarten Sap (CMU), will be giving a talk titled "Towards Prosocial NLP: Reasoning about and Responding to Toxicity in Language".

Abstract

Data-driven AI systems, such as conversational AI agents, are increasingly capable and powerful, yet still suffer from severe toxic outputs. This harmful behavior hinders their safe deployment in the real world. In this talk, I will first examine how data-driven conversational AI systems acquire toxic behavior, by studying the conversation dynamics of contextually toxic language. In a dataset called ToxiChat, we collect annotations of the toxicity and stance of human and model responses to toxic inputs, finding that both humans and models are more likely to agree with toxic content than neutral content. Then, I will present Prosocial Dialogues, a new large-scale multi-turn dialogue dataset to teach conversational AI systems to respond to problematic content. By grounding responses in social norms or rules-of-thumb predicted by our safety model Canary, dialogue models can push back in the face of toxic or problematic inputs and generate socially acceptable responses. Finally, I will discuss the subjectivity challenges in conceptualizing toxicity detection as an NLP task, by examining perceptions of offensiveness of text depending on reader attitudes and identities. Through an online study, we find several correlates between over- or under-detecting text as toxic based on political leaning, attitudes about racism and free speech. I will conclude with future directions designing NLP systems with positive societal impact.

Biography

Maarten Sap is an assistant professor in Carnegie Mellon University's Language Technologies Department (CMU LTI). His research focuses on making NLP systems socially intelligent and understanding social inequality and bias in language. He has presented his work in top-tier NLP and AI conferences, receiving a best short paper nomination at ACL 2019 and a best paper award at the WeCNLP 2020 summit. His research has been covered in the New York Times, Forbes, Fortune, and Vox. Additionally, he and his team won the inaugural 2017 Amazon Alexa Prize, a social chatbot competition. Before joining CMU, he was a postdoc/young investigator at the Allen Institute for AI (AI2) on project MOSAIC. He received his PhD from the University of Washington's Paul G. Allen School of Computer Science & Engineering where he was advised by Yejin Choi and Noah Smith. In the past, he has interned at the Allen Institute for AI working on social commonsense reasoning, and at Microsoft Research working on deep learning models for understanding human cognition.

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 11 a.m. - 12 p.m. during the Fall 2022 semester.

This week's speaker, Aryan Mokhtari (UT Austin), will be giving a talk titled "The power of adaptivity in representation learning".

Abstract

From meta-learning to federated learning

A central problem in machine learning is as follows: How should we train models using data generated from a collection of clients/environments, if we know that these models will be deployed in a new and unseen environment?

In the setting of few-shot learning, two prominent approaches are: (a) develop a modeling framework that is “primed” to adapt, such as Model Adaptive Meta Learning (MAML), or (b) develop a common model using federated learning (such as FedAvg), and then fine tune the model for the deployment environment. We study both these approaches in the multi-task linear representation setting. We show that the reason behind generalizability of the models in new environments trained through either of these approaches is that the dynamics of training induces the models to evolve toward the common data representation among the clients’ tasks.

In both cases, the structure of the bi-level update at each iteration (an inner and outer update with MAML, and a local and global update with FedAvg) holds the key — the diversity among client data distributions are exploited via inner/local updates, and induces the outer/global updates to bring the representation closer to the ground-truth. In both these settings, these are the first results that formally show representation learning, and derive exponentially fast convergence to the ground-truth representation. Based on joint work with Liam Collins, Hamed Hassani, Sewoong Oh, and Sanjay Shakkottai.

Biography

Aryan Mokhtari is an Assistant Professor in the Electrical and Computer Engineering Department of the University of Texas at Austin (UT Austin) where he holds the Fellow of Texas Instruments/Kilby. Before joining UT Austin, he was a Postdoctoral Associate in the Laboratory for Information and Decision Systems (LIDS) at MIT.  Prior to that, he was a Research Fellow at the Simons Institute for the program on “Bridging Continuous and Discrete Optimization”. He received his Ph.D. in electrical and systems engineering from the University of Pennsylvania (Penn). He is the recipient of the Army Research Office (ARO) Early Career Program Award, the Simons-Berkeley Research Fellowship, and Penn’s Joseph and Rosaline Wolf Award for Best Doctoral Dissertation.

RSVP today!.

During each session, the graduate staff will review:

• Requirements (general)
• Applying
• Prerequisite requirements
• What makes a strong applicant
• Funding
• Resources
• Common questions
• Questions from attendees

Students considering the following programs should attend:

• Computer Science M.S.
• Computer Science M.C.S.
• Computer Science Ph.D.
• Data Science M.S.
• Data Science Post-Baccalaureate Certificate
  
   

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m.

This week's talk is a part of the Cray Distinguished Speaker Series. This series was established in 1981 by an endowment from Cray Research and brings distinguished visitors to the Department of Computer Science & Engineering every year.

This week's speaker, Juliana Freire (New York University), will be giving a talk titled "Towards Usability, Transparency, and Trust for Data-Intensive Computations".

Abstract

The abundance of data, coupled with cheap and widely-available computing and storage, has revolutionized science, industry and government. Now, to a large extent, the bottleneck to obtaining actionable insights lies with people. To extract knowledge from data, complex computations that are often out of reach for domain experts who do not have training in computing need to be carried out.  Additionally, there is much room for error in the path from data to decisions, from problems with the data and computations to human mistakes. I will present a set of techniques and systems we have developed to guide users and support the interactivity required for exploratory analyses. I will also reflect on the importance of provenance in this context, not only for transparency and reproducibility purposes, but to enable experts to debug and build trust in the insights they derive.

Biography

Juliana Freire is a Professor of Computer Science and Data Science at New York University.  She was the elected chair of the ACM Special Interest Group on Management of Data (SIGMOD), served as a council member of the Computing Research Association’s Computing Community Consortium (CCC), and was the NYU lead investigator for the Moore-Sloan Data Science Environment. She develops methods and systems that enable a wide range of users to obtain trustworthy insights from data. These span topics in large-scale data analysis and integration, visualization, machine learning, provenance management, web information discovery, and different application areas, including urban analytics, predictive modeling, and computational reproducibility. Freire has co-authored over 200 technical papers (including 11 award-winning publications), several open-source systems, and is an inventor of 12 U.S. patents. According to Google Scholar, her h-index is 64 and her work has received over 17,000 citations. She is an ACM Fellow, a AAAS Fellow, and recipient of the ACM SIGMOD Contributions Award, an NSF CAREER award, two IBM Faculty awards, a Google Faculty Research award. Her research has been funded by the National Science Foundation, DARPA, Department of Energy, National Institutes of Health, Sloan Foundation, Gordon and Betty Moore Foundation, W. M. Keck Foundation, Google, Amazon, AT&T Research, Microsoft Research, Yahoo! and IBM. She received a B.S. degree in computer science from the Federal University of Ceara (Brazil), and M.Sc. and Ph.D. degrees in computer science from the State University of New York at Stony Brook.

Abstract

We develop an assisted learning framework for assisting organization-level learners in improving their learning performance with limited and imbalanced data. In particular, learners at the organizational level usually have sufficient computation resource, but are subject to stringent collaboration policy and information privacy. Their limited imbalanced data often cause biased inference and sub-optimal decision-making. In our assisted learning framework, an organizational learner purchases assistance service from a service provider and aims to enhance its model performance within a few assistance rounds. We develop effective stochastic training algorithms for assisted deep learning and assisted reinforcement learning. Different from existing distributed algorithms that need to frequently transmit gradients or models, our framework allows the learner to only occasionally share information with the service provider, and still achieve a near-oracle model as if all the data were centralized.
 

Biography

Yi Zhou is an assistant professor affiliated with the Department of ECE at the University of Utah. Before, he worked as a post-doctorate research associate in the Department of ECE at Duke University. He obtained a Ph.D. degree in ECE from The Ohio State University in 2018. His research interests include deep learning, reinforcement learning, statistical machine learning, nonconvex and distributed optimization, and statistical signal processing.

You may attend the talk either in person in Walter 402 or register via Zoom. Registration is required to access the Zoom webinar.

Title: Does the Data Induce Capacity Control in Deep Learning?

Abstract: Accepted statistical wisdom suggests that larger the model class, the more likely it is to overfit the training data. And yet, deep networks generalize extremely well. The larger the deep network, the better its accuracy on new data. This talk seeks to shed light upon this apparent paradox.

We will argue that deep networks are successful because of a characteristic structure in the space of learning tasks. The input correlation matrix for typical tasks has a peculiar (“sloppy”) eigenspectrum where, in addition to a few large eigenvalues (salient features), there are a large number of small eigenvalues that are distributed uniformly over exponentially large ranges. This structure in the input data is strongly mirrored in the representation learned by the network. A number of quantities such as the Hessian, the Fisher Information Matrix, as well as others activation correlations and Jacobians, are also sloppy. Even if the model class for deep networks is very large, there is an exponentially small subset of models (in the number of data) that fit such sloppy tasks. This talk will demonstrate the first analytical non-vacuous generalization bound for deep networks that does not use compression. We will also discuss an application of these concepts that develops new algorithms for semi-supervised learning.

Bio:
Pratik Chaudhari is an Assistant Professor in Electrical and Systems Engineering and Computer and Information Science at the University of Pennsylvania. He is a member of the GRASP Laboratory. From 2018-19, he was a Senior Applied Scientist at Amazon Web Services and a Postdoctoral Scholar in Computing and Mathematical Sciences at Caltech. Pratik received his PhD (2018) in Computer Science from UCLA, his Master's (2012) and Engineer's (2014) degrees in Aeronautics and Astronautics from MIT. He was a part of NuTonomy Inc. (now Hyundai- Aptiv Motional) from 2014—16. He received the NSF CAREER award and the Intel Rising Star Faculty Award in 2022.

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. 

This week's speaker, Kangjie Lu (CS&E), will be giving a talk titled "Principled and Practical Approaches to Secure Modern Open-Source Systems".

Abstract

Open-source software is everywhere and has become the backbone of today’s cyber world. In particular, systems software such as operating-system kernels and browsers is arguably the most important one. Modern systems have become extremely large and complex, often containing millions of lines of code written in unsafe programming languages. As a result, they are unfortunately buggy, and a single security bug (or vulnerability) may compromise the whole system. In this talk, I will discuss how to (and why we should) secure modern system software using an overarching, three-pronged approach: program understanding and reasoning, secure-by-design principles and defense, and continuous security assurance. For each part of the approach, I will specifically share our recent accomplishments. At last, I will conclude by discussing some challenging but exciting research opportunities for future work.

Biography

Dr. Kangjie Lu is an assistant professor in the Computer Science & Engineering Department of the University of Minnesota-Twin Cities. His research interests include security and privacy, program analysis, and operating systems. He is particularly interested in developing foundational techniques that enable security mechanismsand analyses, automatically finding and eliminating classes of vulnerabilities introduced by both developers and compilers, and hardening systems while preserving their reliability and efficiency. His research results are regularly published at top-tier venues and have led to many important security updates in the Linux kernel, the Android OS, the FreeBSD kernel, Apple’s iOS, OpenSSL, PHP, etc.  He is a recipient of the NSF CAREER award 2021 and won the best paper award at ACM CCS 2019.  He received his Ph.D. in Computer Science from the Georgia Institute of Technology. More details can be found at https://www-users.cs.umn.edu/~kjlu

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 Wednesday from 11 a.m. - 12 p.m. during the Fall 2022 semester.

This week's speaker, Hong Pang (UMN Department of Pharmaceutics), will be giving a talk titled "Peptide Targeting for Novel Therapies and Cell Biology."

Biography

Dr. Pang is an assistant professor at the Department of Pharmaceutics, University of Minnesota. He graduated from University of Utah as a PhD of biochemistry, and did postdoc training at Sanford-Burnham-Prebys institute at La Jolla, CA.

His lab is specialized in developing peptide-based targeted therapeutics, and eliciting the cellular transport processes. Using phage display, his lab is screening for peptides targeting any given disease or cell type, and deciphering the underlying biomarkers. Through chemical modification and nanotechnology, his lab applies targeting peptides to improve the delivery efficiency and specificity of therapeutic and diagnostic payloads. Moreover, he is interested in the endocytosis, exocytosis and intercellular communication triggered by peptides and their payloads. Using simulation and experimental approaches, his group is investigating the biophysical and molecular basis of these transport processes.

The Inventing Tomorrow Tour Series is your golden ticket to a behind-the-scenes peek into the University of Minnesota College of Science and Engineering's most prestigious labs. Join us to experience how CSE’s unique intersection of science and engineering is creating the possibilities of tomorrow every day here on campus.

This immersive tour of the Gemini-Huntley Robotics Research Laboratory, a state-of-the-art research lab at the forefront of robotics discoveries, will showcase cutting-edge projects hosted by the faculty and students who are designing them. Experience what’s possible when science and engineering combine with innovation and imagination to design and create our future.