Todd P. Coleman - US grants

The research supported under this grant targets the thorough and effective understanding of message-passing algorithms which constitute a large and very potent class of estimation and detection techniques. Indeed, while message-passing (iterative) processing is very successful in practice, understanding its limitations and its sources of non-optimal behavior has been elusive. Despite the enormous impact that message-passing algorithms have, in particular in a communication scenario, practical systems currently rely almost exclusively on a simulation-based evaluation. In this situation, understanding the behavior and geometry of message-passing will not only reduce the necessity of simulations but provide powerful tools for system optimization.

This proposal draws on recent exciting developments that connect message-passing algorithms to the well established theory of convex optimization. As it turns out, message-passing algorithms are intimately related to a linear programming formulation of the inference task at hand. In fact, belief propagation algorithms may be interpreted as an efficient duality-based method to closely approximate the solution to a linear program. Once such connections are established the investigators will strive to understand message-passing algorithms from an entirely new and fruitful point of view. Also, the investigators have already shown that the connection to convex optimization is rooted in the basic property of message-passing algorithms, namely that they operate only locally in a given graphical model. Thus the findings resulting from the approach investigated here will apply to any reasonable locally-operating algorithm.

This project addresses some fundamental scientific questions in the areas of authenticity and trust for digital media. Such issues arise in applications such as forgery detection and characterization, digital fingerprinting for content protection, and transaction tracking. This project develops an analytical framework for solving challenging problems based upon fundamental principles and modern methods of statistical inference; develops novel algorithms; and assesses the reliability of the receiver's decisions.

The educational component of this project includes a summer research program for high school students and undergraduates that teaches them about the ethics and technology surrounding information digital rights management.

The research component of the project focuses on the following two thrusts. First, desynchronization-resilient authentication, exploiting recent advances in Bayesian recursive filtering and inference using graphical models. Second, blind fingerprinting (or traitor tracing), developing theory and codes for problems where the original signal is not available to the receiver.

This project concerns traffic analysis--the practice of learning sensitive
information from communication patterns, rather than their contents. As
encryption of data becomes more prevalent, a detailed study of traffic
analysis is necessary to understand the threats to privacy that patterns
of communication pose, and to design effective countermeasures. Traffic
analysis is also important for intrusion detection, to detect attacks and
abnormalities that are embedded in encrypted traffic.

The project will focus on two types of traffic analysis: flow linking,
where packet timings are used to discover causal relationships between
network streams, and semantic information extraction, where information
about the flow contents is leaked through packet sizes and timings. In
both cases, the goal of the project is to use information, detection, and
queuing theory to discover the fundamental limits of traffic analysis and
to design optimal defenses.

This Integrative Education and Research Traineeship (IGERT) project will educate a diverse cadre of neuroscientists and engineers at the University of Illinois with an advanced understanding of both neuroscience and engineering, enabling them to engage in both sophisticated collaboration and independent research across the traditional gap between these domains. Many of the most important and exciting scientific and technological challenges for the future are centered on neuroscience, the study of the brain. Many recent (and most future) advances in understanding the brain depend on engineering new technologies for sensing, imaging, and analyzing the brain and their innovative use by neuroscientists. Similarly, some of the greatest and most important technological challenges, such as creating neural prostheses for the disabled, require engineers with a profound understanding of neuroscience. IGERT students will thus carry out innovative interdisciplinary research on neuroscience areas of great scientific and engineering importance, such as speech and audition, brain and imaging, and neural implants that may lead to revolutionary advances in understanding the brain and in new technologies such as neural prostheses for the disabled. IGERT trainees will also receive training in leadership, communication skills, and the responsible conduct of research as well as preparation for academic or industrial careers. IGERT is an NSF-wide program intended to meet the challenges of educating U.S. Ph.D. scientists and engineers with the interdisciplinary background, deep knowledge in a chosen discipline, and the technical, professional, and personal skills needed for the career demands of the future. The program is intended to catalyze a cultural change in graduate education by establishing innovative new models for graduate education and training in a fertile environment for collaborative research that transcends traditional disciplinary boundaries.

Information theory, communication theory, and statistical signal
processing have proven spectacularly successful as innovation engines
for communication, data compression, and information processing
technologies. Although in principle these theories are quite general,
in practice they are most useful for a particular family of models,
such as discrete Markov sources and channels with additive
Gaussian noise. Although this family is quite rich and encompasses
many practical technologies, there are several important scenarios that
fall outside of these models. In particular, the timing of discrete events is
the modality of interest in many applications, such as neuroscience
and certain problems in network security, and this modality does
not fit neatly into the standard classes. This research involves
extending information theory, communication theory, and statistical signal
processing to develop a general theory of information transfer via timing.

The first phase of this research involves solving several concrete
problems involving information transfer via timing, such as the
secrecy capacity of timing channels and how to minimize information
transfer via timing side channels in both wireless and wireline
networks. The second phase involves interconnecting these disparate
problems to form a general theory of information transfer via timing.
By elevating the role of timing in information transfer,
this research better positions the fields of information theory,
communication theory, and statistical signal processing to
impact allied fields such as networking, neuroscience, and
operations research, where timing and delay are of fundamental
importance. This impact is facilitated through special topics
courses and invited sessions at conferences, both of which are
designed to attract participants from a range of areas.

This proposal aims to develop a stretchable and flexible sensor technology capable of transforming healthcare from reactive and hospital-centered to preventive, proactive, evidence-based, and person-centered. The goal is to offer 'skin-like' properties, to enable intimate, complete non-invasive integration with the patient. The resulting 'epidermal' electronic devices may allow clinicians to monitor their patients, and the general public to assess, continuously, their health and well-being. The proposed interface monitoring system, designed to promote residual limb health in persons who wear prostheses, in physical forms that are 'skin like', may demonstrate key technological and scientific advancements towards evidenced-based and person-centered prosthetic care. The work involves Development of 'skin-like' pressure, strain and temperature sensors, with wireless operation, as well as hydration and blood flow sensors. Development of computational modeling and algorithms for statistical signal processing of the sensor data and pattern recognition to create a user-friendly interface for clinicians and patients. Application of the proposed sensor technologies and data processing and pattern recognition techniques to prosthetic clinical practice. The continuous capture, storage and transmission of sensor data are critical to the design of lower limb prosthetics for improved health and healthcare. The proposed work is consonant with the mission of NIBIB to improve health by leading development of new biomedical imaging devices for early detection and prevention of health problems and assessment of health status. In addition to prosthetic care, the proposal may address an unmet need for a model system for individualized healthcare, in which continuous sensing, monitoring and assessment are performed using wireless epidermal sensors instead of traditional lab-based instrumentation.

Project Summary: This project represents an interdisciplinary approach to modeling the neural and physiological dynamics of prosocial behavior. Advances in wireless electronics, wearable sensors, computing power, machine learning, signal processing, robotics, information theory, and neuromorphic engineering combined with innovative behavioral design and social neuroanatomical theory make possible the launch of a novel and integrative effort to model the dynamics underlying prosociality. Leveraging the expertise of our highly collaborative cross-functional team we aim to: (1) characterize the physiological (brain/autonomic) signatures of committing a prosocial act; (2) characterize the physiological signatures of experiencing/receiving a prosocial act; (3) quantify the extent to which the physiological dynamics of the recipient, while experiencing a stressful behavioral epoch, predict their social decisions as the actor; and (4) develop the iRat as an ?ethnodroid?, designed to elicit social behaviors through its own behavior while acting as a videographer in close proximity to the rats. Success in achieving all or even a subset of the target capabilities will demonstrate the power of this innovative approach in a virtually limitless array of medical applications. Results of the proposed project will serve as a game-changing springboard for the development of models of coordination and regulation of the brain and body towards efficacious sociality. The work will provide a foundation for future development of techniques to modulate the system for the purpose of restoring balance, regulation, and prosociality towards improved mental and physical health, in addition to interventions aimed towards preventative health. A byproduct of the work will involve technological advances for measuring physiological signals in a relatively unobtrusive fashion and in the development of contextual robotic tools for assessing social behavior. Such advancements will spur additional research programs. Nationally and Internationally, the development and display of prosocial behavior has been associated with more functional family relationships, with stronger mental health, with higher success in school, with greater physical health, and with better empathic responding. Thus, engaging in the proposed research and understanding the predictive dynamics of the system can allow the development of quantitative metrics for diagnosis, treatment and understanding of stress-related mental health disorders ? that modulate or are modulated by ? social dynamics.

Project Summary/Abstract Approximately 1 out of 400 children are born with significant hearing loss, making congenital deafness one of the most common disorders affecting young children. Approximately 50% of congenital deafness is genetic in origin. Currently, the only available treatments for hearing loss are cochlear implants or hearing amplification. While these treatments are often life-changing, they are limited in their ability to restore hearing to normal, which results in lifelong struggles beginning acutely in childhood. Gene therapy approaches for treating recessive hearing loss presents a challenging but exciting opportunity. Viral delivery of functional genes to the ear is challenging, especially in mice ? current in vivo viral delivery methods only transduce a fraction of the sensory hair cells necessary for proper hearing function, and only works easily for smaller proteins. Moreover, multiple applications of viral vectors may be required to target the optimal timing and duration for therapy. Mice and humans lacking the actin-regulatory protein Eps8 are deaf, and Eps8 KO mice have very short stereocilia that fail to contact the tectorial membrane in the organ of Corti. A novel transgenic mouse line will be generated to study the potential of postnatal gene expression in a deaf Eps8 KO mouse model. Using the PhyB system, a mouse line will be created wherein any UAS controlled transgene?s expression can be activated with red light or inactivated with far-red light. Using this system, UAS-Eps8 expression in vivo will be induced by either red light or doxycycline. Systematically varying the initiation and duration of Eps8 expression, then testing for hearing function and stereocilia elongation will facilitate the restoration potential of stereocilia elongation and hearing restoration in vivo in postnatal mice. Furthermore, the role of Eps8 in both development and maintenance of stereocilia over the lifetime of the organism will be elucidated in future studies. Overall, this proposal will pave the way for many future projects probing the effects of gene expression modulation in vivo and will provide an innovative but practical model system for probing and expanding the critical period for hearing restoration.

PROJECT SUMMARY/ABSTRACT Glaucoma affects more than 70 million people worldwide and is the world's leading cause of irreversible blindness. The only current method to delay its development and progression is by lowering intraocular pressure (IOP), achieved with topical administration of eyedrops. Adherence rates for glaucoma eyedrop administration are poor, in many cases below 50%, resulting in disease progression, eventual blindness, and a more than 2-fold increase in healthcare costs. African Americans and Latinos carry a significantly higher glaucoma burden compared with Caucasians. Minorities have additionally been found to have disproportionately lower rates of medication adherence. Previously studied interventions aimed at improving glaucoma adherence had key limitations that particularly affect minorities, including unreliable self-reported measures of adherence, lack of consideration of individual circumstances influencing glaucoma medication management, and developing/testing interventions in predominantly Caucasian populations. Health information technology has experienced rapid advancement in the last decade with the electronic health record (EHR), the proliferation of accessory mobile health technologies, and the advancement of artificial intelligence. Although their integration holds great promise to enable screening tools for diagnosis and risk prediction, successful integration to aid minority populations in real-world settings depends on: understanding how the collected information relates to the patient's other (e.g. clinical) data and the patient's socio-cultural context; seamless information exchange and interoperability with the EHR, the central portal of healthcare delivery; and integration of algorithmic findings into workflows involving both providers and patients to deliver information and/or recommendations in a simple, actionable manner. Glaucoma is a complex chronic disease, spanning decades of patients' lives and requiring ongoing monitoring and evaluation, thus making it an ideal application for the use of health IT to reduce racial disparities. In this proposal, we aim to accomplish this by: demonstrating the effectiveness of a flexible electronic eyedrop sensor to generate granular digital signatures of an individual's adherence and contextualizing this data in a socio- cultural context with patient interviews (Aim 1), combining adherence data with EHR variables to construct machine learning models to predict IOP control and enhance clinical risk stratification (Aim 2), and prototyping a dynamic dashboard for intervention coordination (Aim 3). Altogether, success of this innovative, comprehensive, culturally-tailored, and scalable health IT framework will improve medication adherence and slow disease progression among minorities, therefore narrowing this important racial health disparity.

Project Summary Gastrointestinal (GI) problems are the second leading cause for missing work or school, giving rise to 10 percent of reasons a patient visits their primary care physician and costing $142 billion annually in the US. A majority of such cases are referred to GI specialists, where endoscopy, imaging, and blood tests allow for easy diagnosis of blockages and infections. However, more than half of GI disorders involve abnormal neuromuscular functioning of the GI tract, occurring in a majority of Parkinson?s and diabetes patients for instance. Diagnosis of such GI disorders typically entails subjective symptom-based questionnaires or objective but invasive procedures in specialized centers. Symptom-based diagnosis is problematic because many GI functional disorders with different treatment regimens have overlapping symptoms. Invasive approaches performed in specialized centers can differentiate between myopathic and neuropathic functional disorders and can change the diagnosis/treatment of 15% to 20% of patients with upper GI symptoms. However, they have drawbacks of cost and invasiveness: gastric scintigraphy with its radioactive imaging; manometry, which involves a catheter inserted through the mouth or nose with fluoroscopic or endoscopic guidance. Long wait times and intermittent monitoring associated with assessment of neuromuscular GI disorders, coupled with a strong preference by patients for non-invasive testing instead of current approaches, pinpoints the non-trivial challenges associated with scaling up GI assessment with specialized centers. Altogether, the non-existence of an objective, non-invasive, way to monitor functional GI disorders and their association with transient symptoms is a significant drawback that has vast economic, social, and healthcare consequences. We have developed and demonstrated a procedure that uses a non-invasive multi-electrode sensor array along with a suite of statistical signal processing methods that objectively provide wave propagation descriptions of GI neuromuscular functions that correlate with symptoms. Additionally, using this multi-electrode array we have developed novel Bayesian inference methods to source localize the gastric slow wave on the stomach surface. In this project, we will advance our source localization method to reduce the requirement of human intervention and then apply our method to an existing set of subjects for whom we have already collected data. This project is an important step towards validation of a quantifiable non-invasive measure for gastric health that can help modernize functional gastroenterology. It promotes an inexpensive, non-invasive technology coupled with novel signal processing methods that may lead to transformational clinical approaches that allow for understanding disease etiology, assessing disease progression, and predicting treatment response.