The UMBC CSEE Seminar Series Presents

 Capturing Brain Activity at Rest: Recent Development of Resting-State Functional MRI and Its Potential in Clinical Applications

Dr. Yihong Yang
Neuroimaging Research Branch
National Institute on Drug Abuse, NIH

12noon-1pm Friday, 2 Oct. 2015, ITE 102

There has been growing interest in the intrinsic brain activity at “rest” that may be used to reveal circuit-level information of brain functions. Alterations of resting-state brain activity have been implicated in various neurological and psychiatric disorders. In this seminar, the recent development of resting-state fMRI techniques will be introduced and discussed. Applications of these new imaging techniques in clinical applications such as cocaine addiction and traumatic brain injury will be demonstrated.

Dr. Yihong Yang received his Ph.D. in Biophysics, 1995, at University of Illinois at Urbana-Champaign, under Paul C. Lauterbur who share 2003 Physiology or Medicine Nobel price with Peter Mansfield on the development of MRI. He is currently a senior investigator and the chief of MR Imaging and Spectroscopy Section at NIDA. Dr. Yang has made significant contributions to the development of MRI methodology and application of neuroimaging techniques to neurological and psychiatric disorders. He has published over 130 original research papers in leading journals and contributed several book chapters in the fields of functional MRI, diffusion tensor imaging and MR spectroscopy, as well as applications of MRI technology to the assessment of brain disorders, particularly in drug addiction. He has served on many NIH Study Sections and other research foundations including Medical Research Council (UK), Alzheimer’s Association, and National Science Foundation of China (NSFC). He is an editorial board member of the Brain Connectivity and Open Neuroimaging Journal. He has trained many post-doctoral and pre-postdoctoral fellows in neuroimaging.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

Appropriately Valuing Inter-Disciplinary Research
between Computer Science, Creativity and the Arts

Professor Celine Latulipe
Software and Information Systems
University of North Carolina at Charlotte

2:00pm Friday 2 October 2015, PAHB 132

Scientists and technologists conducting research in creativity and engaging with artists face political pressure to justify their work. A case study of the NSF-funded Dance.Draw project is used to illustrate the problematic aspects of pressure. I argue that a shift in dialogue is needed to appropriately value this type of inter-disciplinary research.

Dr. Celine Latulipe is an Associate Professor in the Department of Software and Information Systems in the College of Computing and Informatics at the University of North Carolina at Charlotte. Her research involves developing and evaluating novel interaction techniques, creativity and collaboration support tools and technologies to support the arts, and developing innovation computer science curriculum design patterns. Dr. Latulipe examines issues of how to support expression and exploration in complex interfaces and how interaction affordances impact satisficing behavior. She also conducts research into how to make computer science education a more social experience, both as a way of more deeply engaging students and as an approach to broadening participation in a field that lacks gender and racial diversity.

According to Virustotal, a prominent virus and malware tool, the Google Play Store has a few thousand apps from major malware families. Given such a revelation, access control systems for mobile data management, have reached a state of critical importance. We propose the development of a system which would help us detect the pathways using which user’s data is being stolen from their mobile devices. We use a multi layered approach which includes app meta data analysis, understanding code patterns and detecting and eventually controlling dynamic data flow when such an app is installed on a mobile device. In this presentation we focus on the first part of our work and discuss the merits and flaws of our unsupervised learning mechanism to detect possible malicious behavior from apps in the Google Play Store.

Multi-Microphone Speech Enhancement

Sharon Gannot
Bar-Ilan University, Israel

1:30pm Wednesday, 14 October 2015, ITE 325B, UMBC

Microphone array algorithms emerged in the early 1990s as viable solutions to speech processing problems. However, the adaptation of beamforming methods to speech processing is still an open issue. There are many difficulties which arise from the characteristics of the speech signal and the acoustic environment. The speech signal is a wide-band and non-stationary signal. Very long room impulse responses (RIRs), which are several thousands of taps long, may be attributed to multiple reflections of the sound source on objects in the enclosure. Moreover, due to the inevitable movements of both sources (speakers) and receivers (microphones), the room impulse responses become time-varying.

In this talk, we will focus on spatial processors, a.k.a, beamformers, based on the linearly constrained minimum variance (LCMV) criterion, and its special case, the minimum variance distortionless (MVDR) beamformer. We show that classical beamformers that merely take into account angular information (as reflected by the so-called beam-pattern), are too simplistic to fully address the intricate propagation regime of the sound source in reverberant environment. We will therefore reformulate the LCMV beamformer in the shorttime Fourier transform (STFT) domain and substitute the free-field steering vector by the entire acoustic transfer function (ATF). The corresponding relative transfer function (RTF) will be then introduced, and its applicability to the design of beamformers in reverberant environments will be discussed. We will then elaborate on several blind RTF estimation techniques, e.g. based on subspace analysis, that enable the implementation of all necessary beamformer’s blocks. Several applications of the powerful LCMV beamformer, e.g. speech enhancement, extraction of desired speakers in multiple competing speaker environment, and binaural processing, will then be presented.

We will conclude the talk with an overview of the emerging field of distributed algorithms for ad hoc microphone arrays, and discuss the advantages and challenges they raise. The presentation will be accompanied by audio clips demonstrating the capabilities of the introduced schemes.

Sharon Gannot received his B.Sc. degree (summa cum laude) from the Technion-Israel Institute of Technology, Haifa, Israel in 1986 and the M.Sc. (cum laude) and Ph.D. degrees from Tel-Aviv University, Israel in 1995 and 2000 respectively, all in Electrical Engineering. In 2001 he held a post-doctoral position at the department of Electrical Engineering (ESAT-SISTA) at K.U.Leuven, Belgium. In 2002-2003 he held a research and teaching position at the Faculty of Electrical Engineering, Technion-Israel Institute of Technology, Haifa, Israel. Currently, he is a Full Professor at the Faculty of Engineering, Bar-Ilan University, Israel, where he is heading the Speech and Signal Processing laboratory and the Signal Processing Track. Prof. Gannot is the recipient of Bar-Ilan University outstanding lecturer award for 2010 and 2014. Prof. Gannot has served as an Associate Editor of the EURASIP Journal of Advances in Signal Processing in 2003-2012, and as an Editor of several special issues on Multi-microphone Speech Processing of the same journal. He has also served as a Guest Editor of ELSEVIER Speech Communication and Signal Processing journals. Prof. Gannot has served as an Associate Editor of IEEE Transactions on Speech, Audio and Language Processing in 2009-2013. Currently, he is a Senior Area Chair of the same journal. He also serves as a reviewer of many IEEE journals and conferences. Prof. Gannot is a member of the Audio and Acoustic Signal Processing (AASP) technical committee of the IEEE since Jan., 2010. He is also a member of the Technical and Steering committee of the International Workshop on Acoustic Signal Enhancement (IWAENC) since 2005. He was the general co-chair of IWAENC held at Tel-Aviv, Israel in August 2010. Prof. Gannot has served as the general co-chair of the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), New-Paltz, NY, USA in October 2013. Prof. Gannot was selected (with colleagues) to present a tutorial sessions in ICASSP 2012, EUSIPCO 2012, ICASSP 2013 and EUSIPCO 2013. His research interests include multi-microphone speech processing and specifically distributed algorithms for ad hoc microphone arrays for noise reduction and speaker separation; machine learning methods in speech processing; dereverberation; single microphone speech enhancement and speaker localization and tracking.

Ph.D. Dissertation Proposal

Verifiable Randomness and its Applications

Christopher Vatcher

10:30am Thursday, 24 September 2015, ITE 325b

We propose to create a public verifiable randomness beacon, to integrate with the Random-Sample Voting system, constructed to be secure against adversaries who have even almost complete control over the system’s source of public randomness including the entropy source.

By verifiable randomness, we do not mean we can prove a sequence of bits to be random. Instead, verifiability means it is possible to prove: (a) a consumer used uniform bits originating from a specific entropy source and therefore cannot lie about the bits used; and (b) the bits used were unpredictable prior to their generation and, with overwhelming probability, were free of adversarial influence. This is in contrast to ordinary public randomness where parties must agree to trust some randomness provider, who becomes a target of corruption. Verifiable randomness is an enhancement of public randomness used to perform random selection in voting, conduct random audits, preserve privacy, generate random challenges for secure multi-party computation, and public lottery draws. Random-Sample Voting specifically requires verifiable randomness for random voter selection and random audits.

Our work extends the work of Eastlake and Clark and Hengartner by considering (a) adversaries who have fine control over the entropy source and (b) physical entropy sources, which we can make verifiable.

Our specific aims include (a) creating adversary models for three entropy source abstractions based on trusted providers, sensor networks, and distributed proof-of-work systems; (b) create a verifiable random beacon that integrates each model; (c) integrate our work with the Random-Sample Voting system; and (d) integrate with NIST’s beacon and propose a verifiable randomness standard based on our work.

Our method is to weaken the trust assumption on the entropy source by introducing verifiable entropy sources, which have mechanisms for limiting adversarial influence and accumulating evidence that their outputs obey a known distribution. Combined with an appropriate randomness extractor, we can generate verifiable random bits. Using sources like weather, we will construct a verifiable randomness beacon: a public randomness provider unencumbered by generous and often unfounded trust assumptions. Such a beacon can serve as a singular gateway for accessing and aggregating multiple entropy sources without compromising the randomness provided to consumers.

Committee: Drs. Alan T. Sherman (Chair), Konstantinos Kalpakis, Weining Kang (Math/Stat), David Chaum (Random-Sample Voting), Aggelos Kiayias (University of Athens)

Ebiquity Lab Meeting

Topic Modeling for RDF Graphs

Jennifer Sleeman

10:30 Monday, 21 September 2015, ITE 346

Topic models are widely used to thematically describe a collection of text documents and have become an important technique for systems that measure document similarity for classification, clustering, segmentation, entity linking and more. While they have been applied to some non-text domains, their use for semi-structured graph data, such as RDF, has been less explored. We present a framework for applying topic modeling to RDF graph data and describe how it can be used in a number of linked data tasks. Since topic modeling builds abstract topics using the co-occurrence of document terms, sparse documents can be problematic, presenting challenges for RDF data. We outline techniques to overcome this problem and the results of experiments in using them. Finally, we show preliminary results of using Latent Dirichlet Allocation generative topic modeling for several linked data use cases.

See: Jennifer Sleeman, Tim Finin and Anupam Joshi, Topic Modeling for RDF Graphs, 3rd Int. Workshop on Linked Data for Information Extraction, 14th Int. Semantic Web Conf., Oct. 2015.

Jennifer Sleeman is a Ph.D. student in Computer Science at the University of Maryland, Baltimore County. Her research interests include the Semantic Web, Machine Learning and ontology matching.

· More talks and directions ·

The UMBC CSEE Seminar Series Presents

Technical challenges and opportunities
in studying the brain’s network activity

Dr. Hanbing Lu
National Institute of Drug Abuse, NIH

12:00-1:00pm, Friday 25 September 2015, ITE 325b

Brain structures do not work in isolation; they work in concert to produce sensory perception, motivation and behavior. Recent advances in fMRI technology offer the opportunity to investigate brain’s network activity. Data are accumulated suggesting that dysregulations within and between network activity are implicated in a number of neurodegenerative and neuropsychiatric disorders, including Alzheimer’s disease and drug addiction. Despite wide application of this approach in systems neuroscience, the fundamentals of brain network activity remain poorly understood. Animal models permit invasive manipulations and are uniquely advantageous in this regard. In this talk, Dr. Lu will discuss technical challenges and opportunities in studying brain networks by integrating multiple modalities, including MRI, electrophysiological recording, optical and electromagnetic neural modulation.

Dr. Hanbing Lu received his doctorate training in Biophysics at the Medical College of Wisconsin, during which he developed hardware and imaging sequence for functional magnetic resonance imaging (fMRI) in rodents. He is currently a staff scientist in the Neuroimaging Research Branch, National Institute on Drug Abuse, NIH. Dr. Lu pioneered animal models to investigate brain’s large scale networks. Current efforts include integrating multiple modalities to better understand the neurobiology of brain’s network activity

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

· More talks and directions ·

PhD Dissertation Proposal

Secured Embedded Many-Core Accelerator for Big Data Processing

Amey Kulkarni

2:00-4:00pm Friday, 18 September 2015, ITE 325b

I/O bandwidth and stringent delay constraints on processing time, limits the use of streaming Big Data for a large variety of real world problems. On the other hand, examining Big Data in applications such as intelligence, surveillance and reconnaissance unveils sensitive information in terms of hidden patterns or unknown correlations, thus demanding secured processing environment. In this PhD research, we propose a scalable and secured framework for a many-core accelerator architecture for efficient big data parallel processing. We propose to merge a compressive sensing-based framework to reduce IO Bandwidth and a machine learning-based framework to secure many-core communications. Four different reduced complexity architectures and two different modifications to Orthogonal Matching Pursuit (OMP) compressive sensing reconstruction algorithm are proposed. We implement the proposed OMP architectures on FPGA, ASIC, CPU/GPU and Many-Core to investigate hardware overhead cost. To secure communications within many-core, we propose two different machine learning-based Trojan detection framework which have minimal hardware overhead. To conclude this work, we aim to implement and evaluate the proposed scalable and secured many-core accelerator hardware for image and multi-channel biomedical signal processing on quad-core and sixteen-core architectures.

Committee: Drs. Tinoosh Mohsenin, (Chair), Mohamed Younis, Seung-Jun Kim, Farinaz Koushanfar (Rice University) and Houman Homayoun (George Mason University)

The UMBC CSEE Seminar Series Presents

Optical Measurements and Devices for Biotechnology and Biomedicine

Dr. Yordan Kostov

Assistant Director, Center for Advanced Sensor Technology, UMBC

12-1pm Friday, 18 September 2015
ITE 102 (Lecture Hall VIII)

A variety of approaches for measurement of bioprocess and biomedical variables are presented. Classical optical measurements (fluorescence, absorption, decay time, etc.) are employed together with miniaturized versions of benchtop spectroscopy equipment to measure a number of bioprocess variables (pH, DO, protein concentration, etc.).  Similar approach allows for measurement of biomedical parameters (transcutaneous O2 and CO2, glucose). The sensing is made possible by the developed miniaturized versions of lab equipment, use of microfluidics and actuation, as well as the use of proper data processing coupled with customizable user interface. A number of examples will be given.

Dr. Yordan Kostov holds an M.Sc. in Electrical engineering from Odessa Polytechnical Institute (Ukraine) and a combined Ph.D. Degree in ChemE./EE from Bulgarian Academy of Sciences. He has industry experience as electronic technology engineer. In his doctoral studies, he focused on optical sensing of biomedical parameters, and pursues this line of research ever since. Currently, he is Assistant Director of the Center for Advanced Sensor Technology at UMBC and Adjunct Professor at CSEE. His main interests are in the area of Biomedical measurements and devices.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

Ph.D. Proposal

Detecting Objects with High Accuracy and in Real-time: A Vision-based
Scene-specific Object Detector in Mobile Systems with Human-in-the-loop Training

Zheng Li

10:00am 15 September 2015, ITE 325b

In computer vision, researchers pursue to train machines to detect objects as well as humans — with high accuracy and in real-time. Though the goal of highly intelligent machine vision has been the target of research for years, machines still perform inferior to humans. Present research continues to specifically investigate new robust features types that lead to improvement of effective detection accuracy. While use of carefully hand-engineered features usually helps, it requires decades of expertise effort to design a good feature representation. Moreover, the machine-end real-time performance often suffers due to the complicated feature extraction and matching. In application where low latency is as critical as high accuracy, such as with unmanned aerial vehicles (UAVs), or assistive guidance and navigation systems for people with visual impairments, approaches to achieve lower execution times are required.

In this proposal, a vision-based Scene-Specific object Detector (SSD) is proposed which transforms the general vision problem into scene-specific sub-problems in order to incorporate scene-specific a priori knowledge to achieve higher detection accuracy and real-time performance. This SSD deeply involves human-in-the-loop training to acquire possible a priori knowledge. With the combination of human-acquired a priori information and sensed real-time information from multi-sensors, a hierarchical coarse-grain to fine-grain search scheme can be used to detect objects efficiently and robustly in a real-time hardware platform. Such a solution can achieve performance exceeding traditional state-of-the-art approaches.

Committee: Drs. Ryan Robucci (chair), Nilanjan Banerjee, Chein-I Chang, Ting Zhu