Track 1 – Autonomy: AI for Autonomous Systems with RACECAR
Time: 1000-1600
 |
Description: Autonomous systems couple artificial intelligence (AI) with vehicles and robots to accomplish goals in complex physical environments. This tutorial will provide a hands-on experience with autonomous systems using the Rapid Autonomous Complex-Environment Competing Ackermann-Steering Robot (RACECAR) platform. RACECAR is a 1/10 scale high-speed vehicle equipped with sensing and processing similar to full-size self-driving cars. Participants will learn about autonomous vehicle perception and planning AI algorithms, including both traditionally engineered and data-driven approaches. Working in small teams, participants will deploy, assess, and tune autonomy algorithms on a RACECAR with the objective of navigating a lap around a racetrack. Experience with software development in a Linux environment is helpful, but not required.. |
|
Lead Instructors Ms. Tate DeWeese |
|
 |
Tate DeWeese is currently an autonomy engineer and researcher for the Control and Autonomous Systems Engineering Group at MIT Lincoln Laboratory. She graduated from MIT with her BS in Mechanical Engineering (2017) and Stanford with her MS in Mechanical Engineering (2019). At Stanford, she focused on mechatronics and autonomy. Currently, she is working on a database management system and a sensor monitoring system for autonomous vehicles, and is also teaching an MIT Junior Seminar on the RACECAR platform. Tuesday, 12 November, 2019 Track 1 – Autonomy: AI for Autonomous Systems with RACECAR |
Track 2 – Computer Vision: Introduction to Deep-Learning Computer Vision Applications in GEOINT
Time: 1000-1400
 |
Description: Over the past decade, the confluence of three major factors has brought about a surge of progress in the field of artificial intelligence:
Data-driven deep-learning approaches to classic computer vision problems such as image classification, object detection, and image segmentation are dramatically outpacing algorithmically determined solutions. With each year, deep-learning techniques further percolate through the space of imagery analysis and provide new footing for tackling challenging tasks, such as fine-tuning vision models to operate on radar, lidar, and other imagery modalities. In addition, hardware and algorithmic advances are enabling the deployment of these deep-learning systems onto smaller platforms, potentially providing new autonomous capabilities. In light of these profound and rapid advancements, it is important for professionals to ground themselves in a firm understanding of the fundamental concepts and technologies that underpin deep learning. Attendees will be shown, in a concise way, the essential mathematical underpinnings of supervised deep learning, along with lessons that help to build an intuition for the data-driven needs and limitations of learned models. They will be introduced to the popular open-source technologies that are most widely used among deep-learning practitioners. Attendees will also be provided an overview of the current state of deep learning as it pertains to computer vision applications and guidance for assessing deep-learning approaches for solving relevant problems. Last, this training course will present research into approaches that mitigate some of the challenges associated with deploying these AI systems onto small UAVs. These challenges include the limited onboard computational resources as well as the data requirements for developing robust autonomous systems. |
|
Lead Instructors Dr. Ryan Soklaski |
|
 |
Dr. Ryan Soklaski is a technical staff member of Lincoln Laboratory’s AI Software Architectures and Algorithms Group. There, he researches machine learning techniques that are performed under data-restricted circumstances, and works as a core developer for a lab-internal machine learning library. Prior to joining the laboratory, Dr. Soklaski earned his PhD in theoretical condensed matter physics at Washington University in St. Louis. His doctoral thesis involved conducting physics simulations on high-performance computing clusters to study the physical mechanisms that drive the glass formation process in metallic liquids. His background in education includes working as a lead-instructor for an undergraduate physics course at Washington University, and as a graduate-level teaching assistant. His interests include methods of numerical analysis software development in Python, and quantum mechanics. |
|
Mr. Greg Angelides |
|
 |
Greg Angelides is a technical staff member of MIT Lincoln Laboratory's AI Software Architecture and Algorithms Group. His research focuses on computer vision algorithms employed in both data and computer resource-constrained environments. This includes investigation of deep neural network architecture characteristics as well as active learning approaches. Prior to his work on machine learning algorithms, Mr. Angelides was a systems analyst on the USAF Red Team, leading analyses on electronic attack system capabilities and overseeing development of high-fidelity air defense simulation software. He received BS and BE degrees from Tufts University and an MS degree in applied mathematics from Northeastern University. |
|
Track 2 – Computer Vision: Introduction to Deep-Learning Computer Vision Applications in GEOINT |
|
Track 3 – Natural Language Processing: AI for Text Analytics
Time: 1000-1500
 |
Description: Natural language processing (NLP) is one of the key technologies of the information age. Understanding complex language expressions is also a crucial part of artificial intelligence. Applications of NLP are everywhere because people communicate almost everything in language. For example, the large volume of unstructured text data pouring into government agencies presents significant challenges for agency operations, rulemaking, and policy analysis. NLP can provide the tools needed to identify patterns and glean insights from all of this data, allowing government agencies to improve operations, identify potential risks, solve crimes, etc. There are a large variety of underlying tasks and machine learning models powering NLP applications. This workshop will introduce the major techniques for mining and analyzing text data to discover interesting patterns, extract useful knowledge, and support decision making, with an emphasis on statistical approaches that can be generally applied to arbitrary text data in any natural language with minimal to no human effort. You will learn about the basic concepts, principles, and major algorithms in text mining as well as their Intelligence Community applications. |
|
Lead Instructors |
|
|
Dr. Olga Simek |
|
 |
Olga Simek is a member of technical staff in the Artificial Intelligence Technology and Systems Group at Lincoln Laboratory. For the past eight years, she has been leading projects, conducting applied research and publishing in the area of text analytics, crowdsourcing, and social networks analysis. Her expertise is in algorithm and workflow design for unstructured open-source data exploration, crowdsourced data analysis and truth datasets creation, recommender systems, threat network detection and classification, credibility estimation, and detection of early indications and warnings. She has applied her skills across multiple projects and missions, including intelligence, surveillance, and reconnaissance, and humanitarian assistance and disaster relief. Prior to joining Lincoln Laboratory, Olga worked in industry, designing machine learning algorithms for fraud detection and financial stability estimation, as well as at Lawrence Livermore and Los Alamos National Laboratories. Olga received PhD in mathematics from the University of Arizona, MS in mathematics from Western Washington University, and BA in computer science and mathematics from Eastern Washington University. |
|
Dr. Charlie Dagli |
|
 |
Dr. Charlie K. Dagli has been a member of the research staff in the Artificial Intelligence Technology and Systems Group (formerly the Human Language Technology Group) at MIT Lincoln Laboratory since January 2010. His primary research interests are in the areas of multimedia understanding, machine learning, and network analysis. Prior to joining Lincoln Laboratory, he held positions at Hewlett-Packard Laboratories, Ricoh Innovations, and State Farm Corporate Research. He was the recipient of the Best Student Paper award at the 2006 ACM International Conference on Image and Video Retrieval and holds three patents for technologies in computer vision and multimedia analysis. Dr. Dagli received a BS degree from Boston University in 2001, and MS and PhD degrees from the University of Illinois, Urbana-Champaign, in 2003 and 2009, all in electrical and computer engineering. |
|
Dr. Lin Li |
|
 |
Dr. Lin Li is a member of the technical staff in the Artificial Intelligence Technology and Systems group at MIT Lincoln Laboratory. Her research is focused on social network analysis, advanced graph analytics, and anomaly detection in networks. She is currently investigating approaches on entity disambiguation in networks, community detection across different social networks, and dark web analytics. Before joining Lincoln Laboratory in 2015, Dr. Li worked at the U.S. Army Research Laboratory, Adelphi, Maryland, as a postdoctoral researcher. Her research was on learning the sparse representations of high-dimensional data, developing a modeling framework for learning the dynamic patterns in the collective behaviors of social agents, and numerically analyzing interactions between connected agents in social networks. Dr. Li is a member of the IEEE. She received the 2014 Young Author Best Paper Award from the IEEE Signal Processing Society for her work on distributed principal subspace estimation in wireless sensor networks. Dr. Li received a PhD degree in 2013 and a BS degree in 2008, both in electrical and computer engineering from the University of California, Davis. Track 3 – Natural Language Processing: AI for Text Analytics
|