DAX 2022

A one-day data science conference at UMBC

Saturday, 4 June 2022

The DAX 2022 Conference will focus on data science, analytics, and general data exploration. Engineers, data scientists, analytic developers, system architects, and business leaders are encouraged to share their experiences and present a topic that would be of interest to the local data community. Expected attendees include engineers, thought leaders, business leaders, and professionals from local government, government defense and intelligence agencies, start-up companies, large data analytic and data science companies, and local universities.

For more information and to register, see the DAX 2022 site. Special registration rate for students!

ArtIAMAS Seminar Series
Co-organized by UMBC, UMCP, and Army Research Lab

Iterative Preconditioning for
Accelerating Machine Learning Problems

Nikhil Chopra
Mechanical Engineering, UMCP

12-1 ET Wed. 27 April 2022, WebEx

We study a new approach to accelerating machine learning problems in this talk. The system comprises multiple agents, each with a set of local data points and an associated local cost function. The agents are connected to a server, and there is no inter-agent communication. The agents’ goal is to learn a parameter vector that optimizes the aggregate of their local costs without revealing their local data points. We propose an iterative preconditioning technique to mitigate the deleterious effects of the cost function’s conditioning on the convergence rate of distributed gradient-descent. Unlike the conventional preconditioning techniques, the pre-conditioner matrix in our proposed technique updates iteratively to facilitate implementation on the distributed network. In the particular case when the minimizer of the aggregate cost is unique, our algorithm converges superlinearly. We demonstrate our algorithm’s superior performance in machine learning, distributed estimation, and beamforming problems, thereby demonstrating the proposed algorithm’s efficiency for distributively solving nonconvex optimization problems.

Dr. Nikhil Chopra is a Professor in the Department of Mechanical Engineering at the University of Maryland, College Park. He received a Bachelor of Technology (Honors) degree in Mechanical Engineering from the Indian Institute of Technology, Kharagpur, India, in 2001, an M.S. degree in General Engineering in 2003, and a Ph.D. degree in Systems and Entrepreneurial Engineering in 2006 from the University of Illinois at Urbana-Champaign. His current research interests are in the areas of nonlinear control, robotics, and machine learning. He is the co-author of the book Passivity-Based Control and Estimation in Networked Robotics. He is currently an Associate Editor of Automatica and was previously an Associate Editor of IEEE Transactions on Control of Network Systems and IEEE Transactions on Automatic Control.

Prof. Cynthia Matuszek receives prestigious
NSF CAREER award for robotics research

CSEE professor Cynthia Matuszek received an NSFCAREER award to support her research on improving the ability of robots to interact with people in everyday environments. The five-year award will provide nearly $550,000 in funds to support research by Dr. Matuszek and her students in the Interactive Robotics and Language lab.

The CAREER award is part of the NSF Faculty Early Career Development Program and is considered one of NSF’s most prestigious grants.  It supports faculty members beginning their independent careers and who have “the potential to serve as academic role models in research and education and to lead advances in the mission of their department or organization.”  One of the program’s central goals is to help early-career faculty “build a firm foundation for a lifetime of leadership in integrating education and research.”

Dr. Matuszek joined UMBC in 2014 after receiving her Ph.D. at the University of Washington in Seattle, where she was co-advised by Dieter Fox and Luke Zettlemoyer.  Before beginning her graduate studies, she was a senior research lead at Cycorp.

Dr. Matuszek’s proposal, Robots, Speech, and Learning in Inclusive Human Spaces, addresses the problem of how robots can use spoken language and perception to learn how to help support people.  A description of her project is below.

“The goal of this project is to allow robots to learn to understand spoken instructions and information about the world directly from speech with end users. Modern robots are small and capable, but not adaptable enough to perform the variety of tasks people may require. Meanwhile, too many machine learning systems work poorly for people from under-represented groups. The research will use physical, real-world context to enable learning directly from speech, including constructing a data set that is large, realistic, and inclusive of speakers from diverse backgrounds.

As robots become more capable and ubiquitous, they are increasingly moving into traditionally human-centric environments such as health care, education, and eldercare. As robots engage in tasks as diverse as helping with household work, deploying medication, and tutoring students, it becomes increasingly critical for them to interact naturally with the people around them. Key to this progress is the development of robots that acquire an understanding of goals and objects from natural communications with a diverse set of end-users. One way to address this is using language to build systems that learn from people they are interacting with. Algorithms and systems developed in this project will allow robots to learn about the world around them from linguistic interactions. This research will focus on understanding spoken language about the physical world from diverse groups of people, resulting in systems that are more able to robustly handle a wide variety of real-world interactions. Ultimately, the project will increase the usability and fairness of robots deployed in human spaces.

This CAREER project will study how robots can learn about noisy, unpredictable human environments from spoken language combined with perception, using context derived from sensors to constrain the learning problem. Grounded language refers to language that occurs in and refers to the physical world in which robots operate. Human interactions are fundamentally contextual: when learning about the world, we focus on learning by considering not only direct communication but also the context of that interaction. This work will focus on learning semantics directly from perceptual inputs combined with speech from diverse sources. The goal is to develop learning infrastructure, algorithms, and approaches to enable robots to learn to understand task instructions and object descriptions from spoken communication with end users. The project will develop new methods of efficiently learning from multi-modal data inputs, with the ultimate goal of enabling robots to efficiently and naturally learn about their world and the tasks they should perform.”

ArtIAMAS Seminar Series
Co-organized by UMBC, UMCP, and Army Research Lab

Risk-Aware Coordination between Aerial and Ground Robots

Pratap Tokekar
Computer Science, University of Maryland, College Park

12-1 PM ET, Wed. 2 March 2022 via Webex

As autonomous systems are fielded in unknown, dynamic, potentially contested conditions, they will need to operate with partial, uncertain information. Successful long-term deployments will need agents to reason about their energy logistics and require careful coordination between robots with vastly different energetics (e.g., air and ground platforms), which is especially challenging in the face of uncertainty. To make matters complicated, communication between the agents may not always be available. In this talk, I will present our ongoing ArtIAMAS work on risk-aware route planning and coordination algorithms that can reason about uncertainty in a provable fashion to enable long-term autonomous deployments.

Dr. Pratap Tokekar is an Assistant Professor in the Department of Computer Science and UMIACS at the University of Maryland. Between 2015 and 2019, he was an Assistant Professor at the Department of Electrical and Computer Engineering at Virginia Tech. Previously, he was a Postdoctoral Researcher at the GRASP lab of the University of Pennsylvania. He obtained his Ph.D. in Computer Science from the University of Minnesota in 2014 and Bachelor of Technology degree in Electronics and Telecommunication from the College of Engineering Pune, India in 2008. He is a recipient of the NSF CAREER award (2020) and CISE Research Initiation Initiative award (2016). He serves as an Associate Editor for the IEEE Transactions on Robotics, IEEE Transactions of Automation Science & Engineering, and the ICRA and IROS Conference Editorial Board.

ArtIAMAS Seminar Series 
Co-organized by UMBC, UMCP, and Army Research Lab

Interactive Natural Language Processing 

Prof. Hal Daume
Department of Computer Science
University of Maryland at College Park

12-1 ET, Wed., February 23
https://umbc.webex.com/meet/nroy

To achieve the goal of building natural language processing systems that help real-world users with tasks, such systems need to be able to communicate bi-directionally with their users. This includes systems describing and explaining their solutions and their difficulties. This also involves systems that can learn from a wide variety of feedback — including language — that a user may provide. I’ll describe our past and ongoing work in this space, and how it ties into our current ArtIAMAS project on systems to help triage documents efficiently and effectively.

Dr. Hal Daumé III is a Perotto Professor in Computer Science and Language Science at the University of Maryland, College Park; he has a joint appointment as a Senior Principal Researcher at Microsoft Research, New York City. His primary research interest is in developing new learning algorithms for prototypical problems that arise in the context of natural language processing and artificial intelligence, with a focus on interactive learning and understanding and minimizing social harms that can be caused or exacerbated by computational systems.

ArtIAMAS Seminar Series, co-organized by UMBC, UMCP & Army Research Lab

Top-K Ranking Deep Contextual Bandits for Information Selection Systems

Dr. Jade Freeman, Army Research Lab

12-1pm ET Wed. 8 Dec. 2021, Online via Webex

In today’s technology environment, information is abundant, dynamic, and heterogeneous in nature. Automated filtering and prioritization of information is based on the distinction between whether the information adds substantial value toward one’s goal or not. Contextual multi-armed bandit has been widely used for learning to filter contents and prioritize according to user interest or relevance. Learn-to-Rank technique optimizes the relevance ranking on items, allowing the contents to be selected accordingly. We propose a novel approach to top-K rankings under the contextual multi-armed bandit framework. We model the stochastic reward function with a neural network to allow non-linear approximation to learn the relationship between rewards and contexts. We demonstrate the approach and evaluate the performance of learning from the experiments using real-world data sets in simulated scenarios. Empirical results show that this approach performs well under the complexity of a reward structure and high dimensional contextual features.

Dr. Jade Freeman is the Chief of the Battlefield Information Systems Branch, DEVCOM U.S. Army Research Laboratory (ARL), overseeing military information systems and analysis research projects. Prior to joining ARL, Dr. Freeman served as the Senior Statistician for the Chief of Staff at the Department of Homeland Security, Office of Cybersecurity and Communications, currently known as The Cybersecurity and Infrastructure Security Agency (CISA), Dr. Freeman obtained her Ph. D. in Statistics from George Washington University

Human-in-the-Loop Entity Mining from Noisy Web Data

Professor Eduard Dragut, Temple University

1-2 pm, Tuesday, 6 April 2021
online via WebEx

Recognizing entities that follow or closely resemble a regular expression (regex) pattern is an important task in information extraction. Due to a vast diversity of web documents and ways in which they are generated, even seemingly straightforward tasks such as identifying mentions of date in a document becomes very challenging. It is reasonable to claim that it is impossible to create a regex that is capable of identifying such entities from web documents with perfect precision and recall. Rather than abandoning regex as a go-to approach for entity detection, we present methods to combine the expressive power of regexes, the ability of deep learning to learn from large data, and the human-in-the-loop approach into a new integrated framework for entity identification from web data. The framework starts by creating or collecting the existing regexes for a particular type of entity. Those regexes are then used over a large document corpus to collect weak labels for the entity mentions and a neural network is trained to predict those regex-generated weak labels. Finally, a human expert is asked to label a set of documents and the neural network is fine-tuned on those documents.

While human effort is critical to build an entity recognition model, surprisingly little is known about how to best invest that effort given a limited time budget. Should a human’s effort be spent on writing a regex recognizing an entity or on manually label entity mentions in a document corpus? When a user is allowed to choose between regex construction and manual labeling, we discover that (1) if the time budget is low, spending all time for regex construction is often advantageous, (2) if the time budget is high, spending all time for manual labeling seems to be superior, and (3) between those two extremes, writing regexes followed by manual labeling is typically the best approach. I will also give an overview of the ongoing and future projects.

Eduard Dragut is an Associate Professor in the Computer and Information Sciences Department at Temple University. He received his Ph.D. degree in Computer Science from the University of Illinois at Chicago. He previously was a Postdoctoral Research Associate at Purdue University, Discovery Park, Cyber Center. His main area of research is Web data management, e.g., retrieval, extraction, representation, cleaning, analysis, and integration. He is actively pursuing projects in  Data Cleaning, Social  Media Mining (e.g., user behavior and fake news), the Future of Work, and Cyber-Infrastructure for Scientific Research. He is co-author of a book on Deep Web data integration, Deep Web Query Interface Understanding, and Integration.