Petar Djuric - US grants

This proposal deals with the analysis of systems and signals by models. The PI plans to use predictive densities in a Bayesian formalism in order to select the appropriate model from a class of potential models.

Model selection, parameter estimation, and spectral analysis are three important areas in statistical signal processing. This research explores some difficult and unresolved problems in these disciplines by exploiting Bayesian theory. Topics of interest include the derivation of model selection rules based on asymptotic assumptions and their applications to problems in array processing, rank determination in time series analysis, and segmentation of vector fields; analysis of transient signals and parameter estimation of highly nonlinear models such as threshold signal models and bilinear models; and, Bayesian spectral analysis of nonstationary signals. This effort primarily consists of three equally important components. The first is a theoretical investigation into these problems that leads to an improved understanding of various signal models and concepts. The second is the practical application of the solutions, which includes automatic segmentation of medical images and the processing of single channel patch clamp currents. The third is purely educational. Students are given a practical exposition into the versatility of Bayesian inference and its applicability for solving a wide range of signal processing problems.

This Research Equipment Grant will make it possible for the Department of Electrical Engineering at the State university of New York at Stony Brook to strengthen its research program on Parallel and Neural Computing. The department will purchase a 128-processor CNAPS computer and a Sun Ultra Sparc workstation which will be dedicated to support research in parallel and neural computing. The equipment will be used for four research project s including: 1) neural control; 2) image processing; 3) parallel system simulation; and 4) Bayesian neural network for modeling and control. These projects all require data intensive simulations. The equipments selected is especially designed with these needs in mind. The 128-processor CNAPS computer is a SIMD (Single Instruction Multiple Data) parallel computer ideal for neural network and image processing applications. The ultra-fast Sun workstation will allow fast, concurrent access to the proposed 128-processor CNAPS system and the existing 64-processor CNAPS system. It can also provide an infrastructure for parallel system simulation.

Abstract

Filtering, prediction, and smoothing are basic signal processing operations. They are all extremely important and are implemented in every radio and TV set, CD player, modem, telephone receiver, and practically any device that extracts information by processing data. Most of the research on these operations in the past has been related to "easy" problems, which are tractable and amenable to nice mathematical analyses. In reality, however, there are many situations where the standard methods of signal processing do not work well and require an altogether different, more general approach. This project is developing methods and techniques for a very wide range of difficult problems that arise in such situations. The areas that will benefit from the results of this research include communications, radar, sonar, seismology, robotics, financial engineering, and control.

This research involves work with dynamic nonlinear and non-Gaussian models. The applied methodology is Bayesian, and the main role is played by the concept of sample filters. Their role is to propagate filtering densities, which are represented by samples drawn from these densities The method is implemented by generating samples from predictive densities, followed by drawing samples from posteriors. Sampling from the posteriors is challenging and will be implemented by perfect samplers. These are based on Markov chain Monte Carlo sampling and the concept of coupling from the past. Perfect sampling is a new concept, and not many results about it are available. It is even more so in the context of sampling from continuous spaces. Therefore, a significant component of the research will be on constructing efficient samplers that will fit the needs of the addressed application. The scope of work also includes performance analyses of the developed sample filters, work on case dynamic models, and development of general guidelines for the new filters.

Petar M. Djuric
Project Summary
Signal processing is an area that already plays a significant role in the current GSM and IS-95 systems.
This role will only increase in the projected third generation wireless communications, and in fact, the
success in implementing these systems will strongly depend on the ability of the signal processing methods
to resolve the new technical problems that will emerge with it. The underlying technology of the third
generation systems will be based on the wideband CDMA (WCDMA) modulation scheme. In the core
of this technology will be sequential signal processing algorithms with abilities to capture fast-changing
characteristics of transmission channels very quickly and to exploit known system information optimally.
The significant advances of latest CDMA signal processing methods notwithstanding, it is clear that the
requirements for much lower bit error rates than in current systems will markedly increase the demands on
signal processing capabilities of sequential algorithms for WCDMA signals. Additional challenge arises due
to complexities that are a result from the need to handle high data rates and users with high mobility. Since
the communication channels will be rapidly time varying, the signals will undergo quick attenuations and the
new algorithms on channel estimation and tracking, channel equalization, interference rejection, and RAKE
receiver adaptations must have extremely fast convergence rates. The objective of the proposed research is
to develop algorithms that will meet the challenges of this new technology.
The basic methodology for the proposed processing of WCDMA signals will be based on particle filters, which recently have gained much attention for their potential in handling nonlinear and non-Gaussian
models. The underlying principle used in the design of such filters is the representation of the posterior
distribution of state variables (the unknowns of the system) by a set of particles (samples). Each particle
is given an importance weight so that the set of particles and their weights represent a random measure
that approximates the desired posterior distribution. The particles may also represent means of density
functions, usually Gaussians, in which case the particles have additional variables, the covariances of the
Gaussians. As new information becomes available, these particles propagate recursively through the state
space and their weights are modified using the principles of Bayesian theory. There are several ways of
applying particle filters including sampling-importance-resampling, mixture Kalman filtering, and Monte
Carlo and Metropolis-Hastings importance resampling. These approaches have their advantages and disad-
vantages in performance, and impose different demands for real-time implementation. In the proposal, new
schemes will be studied that naturally combine the best features of the existing schemes, and tailor them
for processing of WCDMA signals. Not only will the new schemes be able to replicate or surpass the best
possible performance of the known methods, but they will also be general enough to provide foundations for
development of new task specific schemes.
Four important topics will be investigated. The first is the examination of fundamental schemes for
propagation of state particles. This issue is critical for two important reasons: (a) it aspects the performance
of the algorithm and (b) it subsumes the implementation, which although parallelizible, is in some cases too
computationally demanding and therefore not too practical. The second topic is task specific and is related
to multiuser detection and channel estimation as well as exploitation of the physical characteristics of the
channel and the base station/mobile asymmetry for development of improved algorithms. The third one
is examination of the flexibility of the proposed methodology and the interaction of the various algorithms
and tasks in order to improve their performances and robustness. Finally, the fourth topic will be related to
investigation of computational requirements, and structures that would allow for real-time use.

PROPOSAL NO.: 0230487
PRINCIPAL INVESTIGATOR: Berndt, Christopher
INSTITUTION NAME: SUNY at Stony Brook
TITLE: Undergraduate Research and Development in the Engineering Curriculum

Abstract

The goal of this proposal is to develop a plan for curriculum revision that integrates the benefits
of research experience into the more traditional engineering science and engineering degree programs at SUNY at Stony Brook. All our students (and not just the more advanced students) will develop the necessary skills for engineering research, which may include experimental design, searching the literature, performing research using modern scientific instruments and techniques, analyzing and interpreting data, and preparing a report for publication or presentation at an institutional, regional, or national scientific meeting. Then the curriculum will be revised so that the benefits of research experience are integrated into the more traditional engineering science and engineering practice components of our educational programs.

ABSTRACT
0220011
Djuric, Petar
SUNY @ Stony Brook

Optimization of Reconfigurable Architectures for Efficient Implementation of
Particle Filters

In recent years particle filters have attracted great attention in several research communities. These filters are used in problems where time-varying signals must be processed in real time and the objective is to estimate various unknowns of the signals and to detect events described by the signals. The standard solutions of such problems in many applications are based on the
Kalman or extended Kalman filters. In situations when the problems are highly nonlinear or the noise that distorts the signals is non-Gaussian, the Kalman filters provide solutions that may be far from optimal.
A major drawback of the particle filters is that their implementation is computationally very intensive. They are, however, inherently parallelizable, and special hardware can be built for their implementation that can meet the stringent requirements of real-time processing. In this research,
reconfigurable and physically feasible VLSI architectures for particle filters are developed. In the development of these architectures, many important problems are researched. The most critical of them is the balancing of hardware and software, which itself is tightly related to other important issues. They include reductions of computational complexities by transformations and
approximations, investigation of the degree of parallelism implemented in the filter, investigation of various interconnection mechanisms, random communication schemes, hardware optimization, and design of low power VLSI processors. This effort also includes building of reconfigurable hardware so that it is suitable for different types of particle filters.

In the past decade, particle filtering has generated astounding interest among engineers and scientists with its capacity to process data that are modeled by dynamic systems. These methods belong to the family of procedures for sequential signal processing where the objectives are to filter, predict, or smooth unknown and time-varying signals from available observations. The general area of work in this research effort is the building of a new class of particle filters, the development of their theory, and their application to a number of important tasks.

The known particle filtering methods require a mathematical representation of the system dynamics and assumptions about the state transition probability distribution function, and the likelihood of the states. These probabilistic assumptions are often inaccurate and made out of convenience, and in many cases lead to formidable degradations in performance of the particle filters. We develop a more general class of particle filters which do not use probabilistic model assumptions. Instead, the new filters are based on discrete measures defined by particle streams and associated costs that are sequentially updated. With the developed theory, we are able to build particle filters that are simpler, more accurate, more robust, and more flexible than the conventional ones. The standard particle filters, however, are particular instances of the new filters. We investigate in great detail various important issues including the foundations of the new filters, their convergence, connections of the new theory with existing theories, and its extensions to batch type signal processing. The filters are tested on various challenging problems.

This EAGER award addresses the exploratory development of a new technology that can potentially be used for the detection of ultra high energy cosmic rays (UHECR) and that can be used for studying the source of their origin. Owing to their very low event rates, one would need detectors that extend over very large areas, which with conventional technology is very costly. Therefore, the development of new and less expensive methods for UHECR detection over large areas is highly desirable. Radar detection, if proven viable, could be the alternative technology. The concept of UHECR radar detection is based on the scattering of electromagnetic waves by the atmospheric ionization produced by extended air showers. The MARIACHI (Mixed Apparatus for Radar Investigation of Atmospheric Cosmic-ray of High Ionization) experiment set up on Long Island (NY) has several small shower detectors and parasitic bi-static radars to test the concept of radar detection of UHECRs. The analysis of the early data has shown coincidence candidates between radar and shower arrays over distances of 60 km, giving strong motivation to continue with this work.

This award will provide funds to support the implementation of a radar station at the Telescope Array facility at Delta, Utah. An early survey of the radio noise at Delta shows a nearly ideal site for the bi-static radar. They propose to carry out a coincidence search with the Telescope Array whose fluorescence detectors and ground array will provide the PI with information on cosmic ray energy as well as direction, thereby allowing him a direct comparison to the radar data.

The broader impact of the program includes activities involving teachers and students in various research programs. Several groups throughout the country have already shown interest to join this effort. The team is also committed to continue working with underrepresented groups.

A complex system is often defined as a system composed of interconnected parts whose properties cannot be predicted from the properties of its individual components. The investigator studies systems composed of many interacting agents that are not controlled by designated controlling agents and are self-organized. The agents are autonomous and have only local views of their system. They are endowed with the ability to learn from received signals and they share their knowledge with their neighbors by communicating it with agreed languages and rules. A typical feature of such systems is their tendency to display emergent behavior. An important instance of emergent behavior is the phenomenon of herding. Herding is a process where agents in a group ignore their own signals about the state of nature and follow the actions of their neighbors. The research is on the development of a methodology for understanding the interplay between learning and herding in complex systems.

The aim of this work is to study the emergence of herding in complex systems with distributed signal processing. Emergence of herding is defined in a mathematically precise way so that it can be detected in a meaningful way. The agents of the system are rational, i.e., they employ Bayesian learning. The main objective is to understand the emergence of herding in multi-agent systems which is due to diffusion of system knowledge through interactions of received signals, perceived actions of neighboring agents, and learning. Various models of sharing information are studied and scenarios where herding readily arises are identified. Improved methods for efficient diffusion of knowledge in multi-agent systems are developed and ways of preventing adverse herding are sought.

At the recommended level of support, the PI will make every attempt to meet the original scope and level of effort of the project.

The research addresses social networks of agents, where the agents learn about the state of nature not only from private signals (i.e., signals only available to the agents receiving them), but from neighboring agents too. The agents are rational and cooperative, and in forming their beliefs about the state of nature, they process all the information that is available to them. The research aims at finding how information and misinformation can diffuse over networks of agents. The objectives are to use the new theory to design better engineering systems and to influence biological systems in ways so that beneficial outcomes are attained.

The goals of the research are to understand the processes of belief evolution about the state of nature in time and/or space in networks of agents and in a wide variety of settings. The knowledge of the agents is expressed by their beliefs about the state of nature and is quantified by posterior probability distributions. Unlike in the majority of known studies where the agents want to get point estimates about the unknown state of nature, the agents in the addressed problems strive for obtaining complete beliefs about the unknown states as measured by posterior distributions. The state of nature can be static or dynamic and the information acquired from neighbors about it can be of continuous or discrete nature. For information processing, the agents use the Bayes' rule. Endowed with Bayesian reasoning, the agents carry out optimal information processing, and thereby it is expected that they beat the performance of agents that use competing methodologies.

This project will develop an initial version of a Programmable and Versatile Backscattering Tag (PVBT) platform that will drive the research on backscatter-based systems of the future. The idea is to develop a modular, extensible and powerful hardware architecture for backscatter-based tags that is programmable at every layer. The PVBT hardware will be extensible in that the RF, analog and digital sections will be modular and exchangeable. In order to demonstrate the power of PVBT the PIs will build a small set of limited prototypes that will drive several core research directions. These will include (1) development and evaluation of protocols in the PHY/MAC layers providing higher speed and better interference management, (2) exploration of tag-to-tag communication and related networking protocols, (3) sensor networks utilizing backscatter communications.

Use of the PVBT will help researchers and manufacturers adopt PVBT-based evaluation in their workflow to develop new designs faster and at a lower cost, thereby helping RFIDs make more inroads in the marketplace. On the educational front, the PIs will initiate development of an online course on backscatter-based systems. The PVBT will also act as a platform for (i) summer projects of local high school students via several programs at SBU, (ii) research and inquiry-based courses in the WISE (Women In Science & Engineering) Program at SBU, (iii) senior projects of engineering undergraduates, and (iv) course/thesis projects for graduate students at SBU.

Objective: The objective of this project is to address the feasibility of a novel paradigm for the Internet of Things based on radio-less backscattering devices that can communicate with each other independently. The novel device is referred to as sense-a-tag, and it will allow for point-to-point and multihop communication. The work also involves development of protocols for the sense-a-tag. The research is exploratory and of transformative nature. If the feasibility of the envisioned sense-a-tag is demonstrated, one will have the long-sought device that can enable a vast range of applications within the framework of the Internet of Things.

Intellectual merit: The intellectual merit is in the design of a novel device and its protocol. The protocol will make seamless the device?s operation with commercial Radio Frequency Identification tags and readers. The device will play a central role in the Internet of Things where proximity-based sensing and tag-to-tag to communication are critical. The new sense-a-tag will also permit novel solutions for localization, tracking and monitoring.

Broader impacts: The broader impacts are in advancing the research and technological growth of Radio Frequency Identification as a key pillar to the Internet of Things. The device will facilitate low cost and high scalability deployments, a sine qua non for the growth of the Internet of Things, thus bringing it to reality much sooner. The latter will benefit society in innumerable ways. Furthermore, with the proposed device, better Radio Frequency Identification systems will be designed and consequently, more technical and societal benefits will be reaped.

The ?Internet of Things? (IoT) vision requires tiny, intelligent, battery-less devices communicating with one another. However, reliance on power-hungry, active radios for communications has been impeding progress in realizing this vision. The proposed research infrastructure enabled by this project is geared towards enabling a paradigm shift resulting in realization of systems wherein radio-less backscattering devices communicate directly with each other as opposed to an active radio based device (like in traditional reader-tag based RFID systems). This award helps design and develop the Research Infrastructure for Backscatter-Based Networks or RIBBN that responds to this need for an integrated experimental platform which is both flexible and programmable at every layer so that new protocols and systems can be developed and evaluated. Use of RIBBN should help researchers and manufacturers to adopt RIBBN-based evaluation in their workflow while developing new designs faster and at a lower cost, thereby helping RFIDs make more inroads in the marketplace. On the educational front, RIBBN hopes to contribute to courseware development, summer projects of local high school students, and the WISE (Women In Science & Engineering) program in the university.


RIBBN is unique in that it enables experimental research on a new paradigm for future IoT systems by providing three essential capabilities in the same platform ? 1) backscatter-based tag-to-tag communication, 2) on-board sensing and 3) computational ability. RIBBN is developed as a modular as well as programmable platform. It enables a design and experimentation with new tag architectures and protocols in a unified, open, reference system. RIBBN promotes and enhances research in several areas including (1) development and evaluation of various protocols in both lower and upper layers, (2) power harvesting, (3) security and privacy issues in backscattered tag-to-tag communications, (4) relevant signal and information processing methods, and (5) Internet of Things. The proposed effort leads to a setup that we refer to as the RIBBN Lab, an organized testbed facility for conducting research on backscatter-based networks.

The International Conference on Acoustics Speech and Signal Processing (ICASSP) is an annual event, and is the most widely-attended gathering of experts from the signal processing community. In 2014, it will take place in Florence, Italy. Historically, about 25% of the ICASSP participants have been students. This is a very healthy statistic because the future of this all-pervasive discipline is largely determined by the number of young people adopting signal processing as their field of endeavor. Many of the students are coauthors of accepted papers and presenters at the conference. For them, ICASSP is a unique opportunity to meet many of the leaders in the field and to broaden their horizons by being exposed to areas of research different from their own. The conference is also an excellent opportunity for them to start networking with people who will likely be their future collaborators.

Unfortunately is expected that the number of US-based students attending the conference in Florence will be smaller than usual. This is largely due to the high travel costs to get to Florence from North America. This proposal aims at providing travel support for some of these students, hence, insuring better representation of US students at ICASSP 2014.

DESCRIPTION (provided by applicant): For many years, there has been a concerted effort to automate the analysis of fetal heart rate (FHR) rhythms. However, despite significant advances in biomedical signal analysis, there has not been any significant improvement in automated decision support systems. FHR monitoring is now ubiquitous throughout delivery rooms, especially using the non-invasive Doppler monitor, but also using the fetal scalp electrode. Physician classification of fetal heart rate patterns is known to be a non-trivial problem because of significant inter and intra-observer variability of diagnosis. This has led to a marked increase in the number of caesarean deliveries, thereby increasing risk to the fetus and mother in many cases. This has further motivated the machine learning community to automate the classification procedure in the interest of accuracy and consistency as well as robustness with respect to noise. Usual approaches to this involve some type of supervised classification procedure, where the algorithm output on training data is compared with a gold-standard physician classification, followed by testing and validation on new datasets. However, since physician classification can be unreliable in the presence of the aforementioned diagnostic variability, as well as significant tracing noise, we propose the use of unsupervised algorithms to cluster FHR data records into clinically useful categories. We use nonparametric Bayes theory and Markov-time-dependence models for the evolution of feature sequences to propose methods that will achieve improved accuracy. The methods involve extraction of feature sequences from FHR time series data, which are modeled as samples from finite or infinite Dirichlet mixture models. We then use Gibbs sampling to obtain the cluster probabilities for each dataset. Clustering outcomes are compared against direct physician diagnosis and our current results are seen to be in broad agreement with them, while still giving new information on the character of different sub-groups of FHR records. With the proposed research, further gains in classification performance will be made.

With the explosion of mobile data, there is a growing realization that the radio frequency spectrum must be treated as an important resource that is in limited supply. Policy makers and researchers alike are promoting various forms of spectrum sharing models to improve spectrum utilization. Just like any other resource with mismatched demand and supply, all steps towards better utilization of radio spectrum have also increased the need for large scale spectrum monitoring. This serves two key purposes: (i) it helps identify available spectrum opportunities, making spectrum sharing systems more effective, (ii) it can help us develop deeper understanding of spectrum usage and demand over time and space. Large-scale spectrum monitoring can feed into multitudes of 'spectrum-aware' applications forming an entire ecosystem of spectrum data, analytics and apps. The proposed project develops an end-to-end enabling platform called SpecSense to support this vision. SpecSense (i) crowdsources spectrum monitoring using low-cost, low-power custom-designed hardware, and (ii) provides necessary library and interface support for spectrum-aware apps via a central spectrum server/database platform. This project is expected to foster interest in spectrum data marketplaces facilitated by crowdsourced spectrum sensing. This can engender commercial interests in various aspects of the spectrum data ecosystem. In many fields, e.g., healthcare, education, Internet-of-Things, there is a tremendous need for mobile bandwidth and innovation is stunted due to a lack of bandwidth. Success in this project will drive such innovations. The project will also contribute to various educational activities for students with a range of academic preparations.

This project addresses several of the core intellectual challenges in developing SpecSense, viz., (1) Exploration of FPGA-based sensors where sensing algorithms are built into the FPGA, with accompanying tools to automatically implement and optimize these algorithms so that they provide the desired trade-off between power and performance; (2) Novel interpolation techniques to estimate spectrum occupancy in both spatial and temporal domains; (3) Algorithms to support optimized selection of sensors to minimize overall sensing cost; (4) Development of an end-to-end testbed and evaluation over a range of spectrum-aware applications. The project team has a range of expertise in topics relevant to the proposal, such as automated hardware design, digital signal processing, detection and estimation, wireless networking, networking algorithms, and networked systems design.

Complex networks are all around us. They can be physical, biological, social, and virtual. All the species in the world live in societies that can be represented as networks. Almost all complex systems, natural or man-made, are networks of interconnected components. The principal ingredients of all networks are their nodes and the links among the nodes. Most networks change with time. New nodes can be created and old ones can be eliminated. Similarly, new links can be established and existing ones removed. The nodes can form communities which they may leave later in time. The nodes may join another community or create a new one. Communities can emerge and disappear. All these phenomena can create very rich network dynamics. Understanding the common principles of these dynamics, the time-varying network structures and the functionalities that regulate network behaviors is of foremost importance in many fields of science and engineering. The theory that addresses these phenomena is a part of Network Science.

One of the main objectives of Network Science is to exploit statistical signal processing for inferential modeling of physical, biological, and social phenomena. The aspirations are to improve the understanding and prediction of these phenomena. In this project the investigator proposes to advance Network Science by introducing novel models for dynamic networks and novel ways for making inference and learning about them. The PI proposes to work with a methodology where the complexity of the network model is not predefined a priori but instead, it is determined by the observed data. Furthermore, the investigator proposes to work with Monte Carlo-based methods that can meet the most difficult challenges of the models in terms of their nonlinearities and dimensionalities.

The essential role of electronic fetal monitoring (EFM) during labor is to prevent adverse outcomes due to fetal hypoxia and ischemia. Its established weaknesses include: 1) the obstetrician?s highly subjective visual interpretations of the signal patterns and 2) the widespread use of unproven surrogates for relevant fetal hypoxic and/or ischemic injury such as umbilical arterial pH, intrapartum stillbirth, newborn Apgar scores and neonatal seizures. This technology over the past 50 years has not been shown to decrease stillbirths or reduce the numbers of infants with cerebral palsy. EFM as it is presently used in the clinical setting has been associated with an extraordinary increase in the use of operative vaginal delivery and cesarean delivery. No functional algorithm has yet been developed that integrates clinical data collected in the antepartum period and during labor and any other patient specific data with the results of EFM. The main objective of the proposed research is to use recent breakthroughs in machine learning to drive the development of predictive analytics to support and improve the interpretation of EFM data, especially under real world conditions and in real time where clinicians must make timely decisions about interventions to prevent adverse outcomes. It is anticipated that the proposed research will result in significantly decreased use of operative vaginal delivery and cesarean delivery while more precisely defining the fetus at risk for developing metabolic acidosis and long term neurologic injury.

Radio Frequency Identification (RFID) tags have been used to identify tagged objects when they are interrogated by expensive devices known as RFID readers. An RFID reader sends an RF signal toward a tag, and by reflecting that signal the tag transmits information. In this project, the starting assumption is that RFID tags are enhanced and read the reflected RF signal from another transmitting tag. The signal reflected by the transmitting tag is either an ambient RF signal or signal generated by a dedicated inexpensive RF source. This eliminates the expensive RFID readers from the networks. Intelligence and collaboration are added to a network of such tags based on a novel method for estimation of tag-to-tag communication channels. This enables the network to localize and recognize dynamic events that take place in the proximity of the tags. The new capabilities of the tags to gather unique environmental data will unlock a large number of applications in the Internet-of-Things domain which in turn will bring benefits to society in various forms including improved safety, comfort, efficiency and better decision making. The team of researchers will contribute to enhancing the offerings of after-school activities for female high school students via Stony Brook's Women in Engineering and Science Honors Program. Further, the team will be engaged with the daily activities of high-school students through the engineering teaching laboratories in the Electrical and Computer Engineering department.

The goal and scope of the proposed work include the development of autonomous networks of battery-less, RF-powered tags, which besides performing basic operations such as localization and tracking, carry out RF sensing and fingerprinting. This is accomplished by novel techniques for channel measurement based only on passive backscattering and by processing these measurements to recognize surrounding activities. The research on this project brings the following contributions: 1) novel battery-less RF tag architectures that operate with a few microwatts of available power and that are capable of passive channel estimation and necessary communication and processing abilities; 2) effective distributed algorithms for tag selection for channel measurements, information fusion and learning techniques to obtain the best possible estimates given various constraints; 3) novel collaborative backscattering techniques that exploit channel estimation; 4) building discrete and integrated circuit versions of the RF tags which will be used for creating a tag network in the lab and in more realistic smart home environments for evaluating the proposed techniques.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

This project focuses on the physical footprints of consciousness. A convergent team of engineers, neurosurgeons and neuroscientists will focus on the fundamental problem of understanding what causes the emergence of consciousness. Using techniques from systems neuroscience, neurosurgery, signal processing and machine learning, and data collected with humans and non-human primates, the research aims to develop models of the brain circuit interactions supporting consciousness. Improved understanding of the mechanism of consciousness will facilitate the advancement of new therapeutic approaches for disorders of consciousness and cognitive problems (among other broader impacts). The project includes interdisciplinary research training, will integrate research findings into teaching, and includes outreach to high-school and middle-school students.

The theory that the thalamus supports the flexible formation/activation of cortical neural ensembles required for the content of consciousness will be refined, and methods that will restore consciousness in non-human primates will be developed. These advances will be enabled by building novel models, advancing machine learning methods, and researching their theoretical and practical underpinnings. The new methods will be scalable and it is anticipated that they will able to identify causal interactions between brain regions, i.e., whether activity in one area causes activity in another.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

The project aims to build an unobtrusive system to enable lip reading for patients with Amyotrophic Lateral Sclerosis (ALS, also known as Lou Gehrig's diseases) and individuals with speech and hearing disorders. Although there is rich literature on lip reading, the bulkiness, obtrusiveness, and/or immobility of these solutions impedes their applications in daily practice, especially for patients with neuromuscular disorders. There is an urgent need to develop novel lip-reading technologies to improve the communication capabilities of ALS patients with loved ones and healthcare providers. The proposed system can considerably improve on existing solutions for tracking and interpreting facial movements and more broadly, body movements, such as finger motions and body gestures. The ability to gather multimodal motion patterns from unobtrusive sensors and apply machine learning (ML) to interpret the acquired data would greatly facilitate diagnosis, treatment, and rehabilitation of motion-related disorders, such as stroke and Parkinson's disease. In addition, this work paves the way for the development of nonverbal communication interfaces enabled by facial/body gestures and opens new avenues for rehabilitation, robotics, and human-machine interfaces. This project presents an excellent opportunity for students to participate in cross-disciplinary research. Part of the research will be integrated into the PI's courses and capstone design projects. The PIs are committed to outreach activities and increasing the diversity through local minority organizations and the Vertically Integrated Program at Stony Brook University.

The overarching goal of this project is to build an unobtrusive hardware-software platform for ALS patients that can capture speech-relevant lip gestures and decode lip movements for speech. First, a skin-like multimodal strain and electromyography (EMG) sensing system will be designed to track both skin deformations and muscle activities associated with lip movements. Self-assembled structures will be introduced to render the sensors ultrathin, breathable, and semi-transparent. Second, the feasibility of converting the sensed lip signals to corresponding spoken words will be demonstrated. Modern ML methods, and in particular, ensemble Gaussian processes (GPs) will be exploited for speech recognition. In the proposed scheme, each GP serves as a classifier and the final decision is made by fusing the results of all the GPs by making use of methods within the Bayesian framework. The potential contributions of the proposed work include: 1) Design of skin-like strain and EMG sensors with high sensitivity and good skin compatibility through a scalable self-assembly process. 2) Integration of multimodal sensors for comprehensive in-vivo quantification of lip movements associated with speech. 3) Development of ML algorithms that precisely convert lip movements to speech. 4) Laying the grounds for developing a truly natural and unobtrusive hardware-software system for lip reading. Our proposed work can fill the gaps in the existing solutions by an intuitive and unobtrusive technology for lip reading.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

The dynamics of complex systems are often studied by processing multivariate time signals that are produced by these systems. Improved understanding of the systems from such signals hinges on working with accurate models of the systems. The rationale of the proposed methodology for making inference from multivariate time signals stands on three important principles: algorithmic compressibility, locality, and deep probabilistic modeling. With algorithmic compressibility, one interprets seemingly complex high-dimensional data in much lower dimensional spaces. With locality, one exploits the fact that in nature the most influential events to an event are its local events. With deep probabilistic modeling, one aims at finding algorithmic compressibilities. These principles are used for developing novel models with little prior knowledge about the dynamics of the observed system. Another challenging problem of interest in the project is discovering causes and effects based on the adopted models. The developed methods are tested on multivariate local field potentials acquired from patients with epilepsy. Based on these signals, the objective is to find the zones in the brain that cause seizures in the patients. Finding these zones and removing them by surgery often cures the patients. <br/> <br/>The project conceptualizes a principled approach to building deep state-space models with deep probabilistic modeling. The research includes the development of theory and methods for estimating the unknowns of these models, investigation of methods for estimating the structures of the models, extension of the new methods to models that capture regime switching, development of theory and methods for discovering causalities among multivariate time signals, and identification of states that cause seizures in patients with epilepsy. The research is based on minimal assumptions about the models and is carried out within the Bayesian framework. The methodology is not data hungry, and all the produced results are probabilistic in nature. This research on deep probabilistic models and causal discovery considerably extends the capabilities for modeling multivariate time signals, which not only facilitates our understanding of complex systems but also offers new paradigms that extend the horizons and scope of signal processing and machine learning. The applications in medicine and neurosurgery, such as identifying the pathological zones in the brain of epilepsy patients that cause seizures stand on their merit.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.