Learning from High-Dimensional Data via Transformations

Dr. Hossein Mobahi, MIT

12:00pm Monday 29, February 2016, ITE325b

High-dimensional data is ubiquitous in the modern world, arising in images, movies, biomedical measurements, documents, and many other contexts. The “curse of dimensionality” tells us that learning in such regimes is generally intractable. However, practical problems often exhibit special simplifying structures which, when identified and exploited, can render learning in high dimensions tractable. This is a great prospect however, but how to get there in not trivial. In this talk, I will address two challenges associated with high-dimensional learning and discuss my proposed solution.

First, parsimony (sparsity, low rank, etc.) is one of the most prevalent structures in high-dimensional learning applications. However, its presence is often implicit and it reveals itself only after a transformation of the data. Studying the space of such transformations and the associated algorithms for their inference constitute an important class of problems in high-dimensional learning. I will present some of my work in this direction related to image segmentation. I will show how low-rank structures become abundant in images when certain spatial and geometric transformations are considered. This work resulted in a state of the art algorithm for segmentation of natural images.

Second, important scenarios such as deep learning involve high-dimensional nonconvex optimization. Such optimization is generally intractable. However, I show how some properties in the optimization landscape, such as smoothness and stability, can be exploited to transform the objective function to simpler subproblems and allow obtaining reasonable solutions efficiently. The theory is derived by combining the notion of convex envelopes with differential equations. This results in algorithms involving high-dimensional convolution with the Gaussian kernel, which I show has a closed form in many practical scenarios. I will present applications of this work in image alignment, image matching, and deep learning. Furthermore, I will discuss how this theory justifies heuristics currently used in deep learning and suggests new training algorithms that offer a significant speedup.

Hossein Mobahi is a postdoctoral researcher in the Computer Science and Artificial Intelligence Lab at the Massachusetts Institute of Technology. His research interests include machine learning, computer vision, optimization, and especially the intersection of the three. He obtained his PhD from the University of Illinois at Urbana-Champaign in Dec 2012. He is the recipient of Computational Science & Engineering Fellowship, Cognitive Science & AI Award, and Mavis Memorial Scholarship. His recent work on machine learning and optimization have been covered by the MIT news.

The UMBC Cyber Defense Lab presents

Mini-MAC: Raising the Bar for Vehicular Security with a
Lightweight Message Authentication Protocol

Jackson Schmandt, CSEE, UMBC
11:15am-12:30pm Friday, 26 February 2016, ITE 237

We propose Mini-MAC, a new message authentication protocol that works in existing automotive computer networks without delaying any message or increasing network traffic. Deployed in many vehicles, the CAN bus is a low-speed network connecting electronic control units, including those that control critical functionality such as braking and acceleration. The CAN bus is extremely vulnerable to malicious actors with bus access, including wireless access. Traditionally, Message Authentication Codes (MACs) help authenticate the sender of a message, and variants prevent message replay attacks; however, standard MACs are unsuitable for use on the CAN bus because of small payload sizes. Restrictions of the CAN bus, including the need not to delay messages or increase bus traffic, severely limit how well this network can be protected.

Mini-MAC is based on a counter-seeded keyed-Hash MAC (HMAC), augmented with message history and truncated to fit available message space. It does not increase bus traffic and incurs a very small performance penalty relative to the provably secure HMAC. It is the first proposal to combine these two tenets for vehicle networks. The message history feature protects against all transient attackers, even if they know the keys. Though the CAN bus cannot be properly secured against a dedicated attacker, Mini-MAC meaningfully raises the bar of vehicular security, enhancing the safety of drivers and others.

Jackson Schmandt is a MS student in Computer Engineering in the Mobile Pervasive Sensor System Lab. Joint work with Alan Sherman and Nilanjan Banerjee.

Host: Alan T. Sherman, Sorry, you need javascript to view this email address.

Establishment of Trust and Integrity in Modern Supply Chains

Ujjwal Guin, University of Connecticut

11:30 Thursday, 25 February 2016, ITE325b

With the advent of globalization and resulting horizontal integration, modern supply chain becomes extremely complex and requires immediate solutions for eliminating counterfeit integrated circuits (ICs), which pose a serious threat to the safety and security of our day-to-day lives. The reliability of such ICs could be questionable as they may have many defects and might not go through as much of a rigorous test process as their authentic counterparts. An adversary can also create a backdoor to bypass the security modules in these ICs. In this research, I have systematically addressed the aforementioned issues by risk analysis and assessment of test methods, and by proposing different Design-for-Anti-Counterfeit (DfAC) measures. As a part of risk analysis, I have developed taxonomies for counterfeit IC types, counterfeit defects, and test methods. Based on these taxonomies, I have introduced novel test metrics and developed a comprehensive framework for assessing a set of test methods to maximize test coverage. In the DfAC domain, I have proposed a suite of solutions to detect counterfeit ICs without performing conventional tests. A set of lightweight negative-bias temperature instability (NBTI)-aware ring oscillators have been developed for combating die and IC recycling. In addition, I have developed a comprehensive solution for preventing intellectual property piracy and IC overproduction by assuring forward trust between all entities involved in the system-on-chip design and fabrication process.

Ujjwal Guin is a PhD candidate at the Electrical and Computer Engineering department of University of Connecticut, where he has been working with Dr. Mark M. Tehranipoor. His current research interests include Hardware Security and Trust, Supply Chain Security, Cybersecurity, and VLSI Design and Test. He has developed several on-chip structures and techniques to improve the security, trustworthiness, and reliability of integrated circuits. He has co-authored a book entitled “Counterfeit Integrated Circuits – Detection and Avoidance”. He has published several journal articles and refereed conference papers. He received Best Student Paper Award from the IEEE North Atlantic Test Workshop (NATW’2013). He is an active participant in the SAE International’s G-19A Test Laboratory Standards Development Committee. Mr. Guin received his B.E. degree from the Department of Electronics and Telecommunication Engineering of Bengal Engineering and Science University, Howrah, India in 2004 and the M.Sc. degree from the Department of Electrical and Computer Engineering of Temple University, Philadelphia, PA, USA in 2010.

Host: Chintan Patel

CSEE professor Marie desJardins has received an Undergraduate Research Faculty Mentoring Award from the Computing Research Association (CRA). The award recognizes faculty who have provided exceptional mentorship, undergraduate research experiences, and, in parallel, guidance on admission and matriculation of undergraduate students to research-focused graduate programs in computing. The award will be presented to Dr. desJardins at the CRA Conference at Snowbird in July.

Dr. desJardins is a professor in UMBC’s Computer Science and Electrical Engineering department and Associate Dean of the College of Engineering and Information Technology. Her research is in the area of artificial intelligence, machine learning, and computer science education. Currently, her research group includes six undergraduates and five graduate students. Since 2005, she has mentored over 70 undergraduate students. At least 29 of them have enrolled in graduate programs in computing, with eight of the 29 having pursued a Ph.D. in computer science.

In announcing the award, the CRA noted that

“Many of the undergraduates Marie has mentored never imagined that they would be involved in research and consider graduate education. Marie’s strategy for working with undergraduate majors involves engaging with students in their first two years and building teams in which her more senior research students (graduate and undergraduate) help train and lead the junior students. She is known for her unconditional support, encouragement, and dedication. She encourages students from underrepresented minorities to get involved with national organizations and programs, including CRA-W, NCWIT, Grace Hopper, IJCAI and AAAI. She stays in touch with her mentees beyond their graduation, and she has helped some to become effective mentors themselves.”

This is the first year for the CRA Undergraduate Research Faculty Mentoring  Award, which was also given to Pieter Abbeel of the University of California, Berkeley and Judy Goldsmith from the University of Kentucky.

Founded in 1972, CRA‘s membership includes more than 200 North American organizations active in computing research: academic departments of computer science and computer engineering, laboratories and centers (industry, government, and academia), and affiliated professional societies (AAAI, ACM, CACS/AIC, IEEE Computer Society, SIAM, USENIX). Its mission is to enhance innovation by joining with industry, government and academia to strengthen research and advanced education in computing. CRA executes this mission by leading the computing research community, informing policymakers and the public, and facilitating the development of strong, diverse talent in the field.

Larry Stevens, Ph.D.

Department of Psychological Sciences

Northern Arizona University

P.O. 15106

Flagstaff, AZ 86011