Jose Carlos Principe - US grants

Supervision of machine tools will be investigated from the perspective of a fault detect system. Recognition of faults mean that the system is able to make predictions and compare predicted with measured values. The comparison cannot be strict, the noise can cause false or missed conditions. A deeply coupled, numeric/symbolic model that incorporates heuristic and physical knowledge and is implemented in a multiprocessor architecture will be used. Real-time signal processing algorithms being developed will be implemented in digital signal processing chips, and will be validated with real data. Another aspect of this research deals with improvements on symbolic processing speed and a decision making strategy to service alarms within a maximum response time. For the inference mechanism, a neural network will be developed and evaluated, bringing distributed parallel processing to the most time consuming aspect of inference. In order to maximize the system reaction time interrupt driven specialists will be implemented. This will focus the symbolic reasoning and create belief levels in the alarm by integrating multiple sensor information. As an end result, a self-contained computer system capable of real-time knowledge-based supervision will be available for demonstrations.

This project is a collaborative effort between an electrical engineer and a neuroscientist, to develop a new sensor capable of an order of magnitude improvement in our ability to study living human brain tissue and how it works. The ultimate goal is the development of a chip, .3 cm by .3 cm, containing 16,000 independent electrodes capable of independent recording from brain cells. Another part of the research is the development of techniques to keep brain slices alive and functional in contact with the sensor, and to design a sensor which does not deteriorate in that environment. Tests will be done on brain slices already available as a byproduct of surgery to correct epilepsy; sensor data will be analyzed to see how patterns vary between normal tissue and tissue causing the epilepsy, in order to yield insight into normal and abnormal functioning of the brain.

The purpose of this research is to establish a test model of a highly comprehensive supervision system. It will be implemented on a 200 Series Omnimill with a Flex-Mate Controller. Sensors of tool wear, tool breakage, chatter, geometric errors due to cutter deflections, and of cutter overloads will be employed. Signals from the sensors will be processed together with predictions based on inputs derived from the numerical control (NC) program and on analytical models and simulations. Using these models and information theory, corresponding corrective actions will be derived and carried out through routines implemented in the NC Controller. Thus, this system will be comprehensive in that sense that apart from the sensor signals it includes also inputs from the Computer-Aided Design (CAD)/Computer-Aided Manufacturing (CAM) system for more than just the usual cutter location path data and it includes the Controller for the execution of the corrections. Some of these corrections will permanently modify the NC program.

A workstation and electronic test equipment will be provided for researchers at the University of Florida for research in the Department of Electrical Engineering. This equipment is provided under the Instrumentation Grants for Research in Computer and Information Science and Engineering program. The research for which the equipment is to be used will be in the area(s) of Modular arithmetic systems; attached array processors, and neural network simulators.

This research activity is designed to develop the theoretical and experimental knowledge base to enable the development of the next generation metal cutting machine tool. This machine tool is characterized by the properties of high speed, high power, high accuracy and high static and dynamic stability. The characteristics of high speed and power are necessary to reduce costs involved in machining. The key to achieving this goal is the development of statically and dynamically stable machine tool sub- systems, integrated to form a highly stable machine tool. Sub- systems such as stiff high speed spindles and new-material, light- weight, stiff structures will be researched and developed. Major control issues, such as spindle speed control to avoid chatter and servo systems with continuous updating of drive dynamics to reduce following error, will be addressed. Modelling and simulation of machining processes will be conducted, on a large scale, to allow the incorporation of these models into further development of Computer Aided Design and Manufacturing systems. The research is focused on developing models of machine tools for the aerospace and automotive industry, which represent the two largest segments of manufacturing. Systematic research of this nature cannot be provided by the domestic industry which is extremely fragmented and battered by overseas competition. This work will however, provide for a revival of the domestic machine tool industry, through cooperation and technology transfer, while greatly benefiting a major segment of the manufacturing sector of the United States.

This project will seek a unifying connectionist framework for the classification of time varying signals based on the gamma model proposed by (de Vries and Principe, 1990, 1991, 1992). The gamma model extends the conventional additive model because the multiplication of output activations with weights is subsituted by a temporal convolution, yielding a computationally improved additive model. This research will have theoretical and practical components. New insights on the role of temporal weights will be sought, using the theory of recursive estimation. Likewise, a paradigm to study how the neural network is representing the history of the input signal will be sought, using Hilbert spaces. The gamma memory projects a vector of unconstrained dimension onto a predefined subspace. This same principle will be extended to learning, producing a backpropagation through time algorithm that has memory requirements independent of the length of the input signal. As an application, the gamma network will be implemented in a multiprocessor system for speech recognition. The P.I. will contrast the gamma model with two of the best available paradigms, Lang's time delay neural network and Tank's and Hopfield concentration in time networks, in terms of performance, network complexity and learning effectiveness.

This project will attempt to develop more advanced designs for artifical neural networks which can learn from experience how to predict of filter streams of inputs over time. The work will build upon novel designs developed under a previous NSF grant, which are being tested in applications such as blind deconvolution and interference cancelling in nonstationary communication channels (e.g., making cellular phones work better), identification and control of nonlinear plants, prediction of financial time-series and classification of time varying patterns such as speech recognition and transient signal processing (radar, sonar, biological). The basic approach will be to deepen the theoretical understanding of the limitations of the existing designs in these applications, and to develop new general-purpose designs--rooted in an exploration of basic theory--to overcome these limitations. Among the theoretical issues given priority are the problem of mixing discrete and continuous variables in the context, the problem of time-warping (e.g. different speakers speaking at different speeds), and the problems of robustness over time.

This project equips a classroom with a network of PCs to teach neural systems in the undergraduate curriculum. Since the mathematics of adaptive systems are beyond the knowledge of the undergraduate, the project teaches the concepts through simulation. An undergraduate textbook integrating a hypertext and a commercial simulator is being written to explain the concepts. Students use the simulator in every lecture to comprehend the concepts of adaptation.

9713039 Fred Taylor University of Florida CISE Educational Innovation: Synergy: A CISE Learning Environment This CISE Educational Innovation award, involving both the University of Florida and the University of Texas at El Paso, supports the development of a new activity-based learning environment called Synergy. Synergy provides undergraduate students with the opportunity to engage in hands-on research-centric digital signal processing (DSP) studies within a simulated professional engineering framework. A significant part of the project includes development of a new capstone laboratory which is designed to provide students with an experience which requires demonstrated unstructured problem solving skills, clear and concise communications, planning and scheduling discipline, Internet utilization, and personal accountability. The laboratory is being developed with input from DSP projects from across areas of the institution including engineering, medicine, education, architecture, computer science, and others, and from industrial partners as well.

The goal of this CRCD project at the University of Florida, entitled, "A Net-Centric Undergraduate Course in Adaptive Systems," is to innovate the engineering undergraduate curriculum by developing a net-centric, WEB based upper division elective on Adaptive Systems. The ultimate goal is to establish a resource center for Adaptive Systems undergraduate instruction at the University of Florida which will integrate collaborations from experts worldwide and will deliver the course locally and distribute it across the WEB to learners at other Universities and in Industry. The software tools and the teaching methodologies developed during this project will be widely applicable to other technological courses, so they transcend the specific area of Adaptive Systems.

9900394
Principe

The focus of this research is the development and evaluation of a new class of algorithms for information
theoretic learning (ITL). Conventional learning designs are usually based on an effort to minimize either square error or a measure of entropy which can be computationally difficult to handle. This project will attempt to estimate and use a new measure of entropy, based on concepts from Renyi. The appeal of the algorithm is that it can be easily integrated with a Parzen window estimator yielding several practical criteria to adapt universal mappers, either under unsupervised or supervised paradigms.

If the research is successful a novel and quite general class of algorithms will be made available to
the scientific community interested in studying and applying learning systems. The natural goal of this research is to further develop the ITL algorithms, an study their application to a variety of important problems in learning. Specifically, the project will study the properties of the estimator for both entropy and mutual information, ways to decrease the computational complexity of the algorithm, extend it to time signals, set its parameter and access its scalability. It will also investigate new distance measures for mutual information optimization and the feasibility of imple-menting an "entropy chip" in analog VLSI which will use the laws of physics to do the computation.

The research team/will be investigating issues in information filtering, independent component analysis and blind source separation using the newly developed ITL class of algorithms. These areas are important in their own right, and have a momentum that will be further advanced in this research. But the team will also use the common computational infra-structure of estimating entropy and mutual information from examples to compare the proposed ITL algorithms' performance with the best available techniques in each field. Specifically, (1) They will extend the present state-of-the-art in the blind source separation of convolutive mixtures. They will apply the new
learning algorithm to the co-channel interference in mobile communication channels, and noise reduction in hearing aids. (2) They will apply ITL to system identification and dynamic modeling and com-pare it to our previous results using the mean square error. (3) The problem of sparse representa-tions is crucial to understand the brain and design intelligent artificial systems. We will be researching how to learn sparse representation from data, using both overcomplete bases and the ITL algorithm to implement independent component analysis.
***

EIA-0135946
Fortes, Jose A
University of Florida

ITR/SY: Design and Simulation of Biologically inspired Nanolattice






This joint project between the University of Florida and Purdue University is pursuing scientific principles for designing and engineering biologically inspired neuromorphic computing architectures using radically new molecular electronic devices and biologically inspired, ultra-dense, self-assembled systems. Examples of applications of these architectures include unprecedently small and inexpensive nanoscale intelligent sensors. The architectures can be used to implement neurocomputing models and are well suited for nanotechnologies, thus accelerating the development of useful nanotechnology by providing clear functional targets for nanodevices.
The team of investigators includes computer architects, neurocomputing experts and device physicists working in close collaboration along three highly synergistic thrusts. One of the two thrusts is focused on advancing the understanding of biologically-inspired dynamic information processing systems in order to understand the impact of constraints imposed by architectures and technologies on the properties of these systems. Another thrust investigates neurocomputing system architectures that can be engineered within the constraints of nanotechnologies. The third thrust develops a toolbox of novel mechanisms for integration, self-assembly and interconnection of nanoscale devices.
The architectures are investigated via formal methods and simulation. Internet resources are used to conduct simulation, and to disseminate models, software and other research results. A new course, summer internships and educational materials are being developed to educate students on the key interdisciplinary aspects and results of the project.

EIA 02-24442
Fortes, Jose A.
Figueiredo, Renato; George, Alan D.; Principe, Jose C.; Su, Stanley Y.
University of Florida

CISE RR: Collaborative Research on Wide-Area Network Computing Using Virtual Machines

This collaborative research project (with Dinda at Northwestern University, proposal 02-24449), requiring a wide-area test bed that enables experimentation with, access to, and running of applications on unique resources, requests PC clusters, an IBM server, and other ancillary hardware for projects in
1. Distributed grid computing and information processing systems using virtualization technologies and
2. Information grids with real users and research applications requiring capabilities enabled by virtual machines (VMs).
Deploying a distributed system based on clusters connected by local, metropolitan, and wide area networks, the work aims to provide a virtual computing and data storage interface to clients that access resources on the underlying "information grid." The test bed includes the following defining features.
1. Virtualization capabilities, i.e., the ability to instantiate independent logical machines that can be multiplexed on physical processors (or fractions of them), storage and network I/O channels, and can use distinct operating systems;
2. Wide-area distribution, i.e., Internet-linked test bed components in independently-administered geographically-apart network domains;
3. Scalable capacity for both scientific computing and information processing; and
4. Heterogeneity.
Interrelated projects enabled by the test bed towards the goal of developing VM-based middleware for grid computing include virtualized end resources, monitoring and prediction, interactive computing, virtual file systems, data management, cycle selling, and security. Information grids and web portals for use of CAD tools are also enabled by the infrastructure for dissemination of collaborative research results and data, and for digital government services. From the availability of the portals and grid-computing resources benefits are expected in brain-machine interfaces, biologically-inspired nanocomputing, auction-based computing, distributed knowledge applications, medical imaging and data archiving, light-scattering spectroscopy, and mixed non-linear optimization. Collaborations include the Sigmicro microarchitecture center, NETCARE and the Purdue-hosted Nanohub (enabling users to run tools for computer architecture and parallel computing, and nanoelectrnics). The project impacts some minority serving institutions such as Chicago State and Florida A& M Universities and enables a testbed for a transnational digital government projects involving Carnegie Mellon University, University of Belize, University of Colorado, University of Florida, University of Massachusetts, and Pontificia Universidad Catolica Madre y Maestra of the Dominican Republic.

This project will investigate a novel approach to model and design learning algorithms in the small sample case using entropy. It will seek a deeper mathematical understanding of the new method and its relations to the mean square error, still today the workhorse of learning. The PI will investigate the properties of the information field created by the information particles, and how to use it to improve generalization and construct bounds on system errors. He will depart from the static fields provided by statistics and incorporate time in the information forces to mimic the interactions in diffusion fields and oscillating fields. This will open the door to adapt systems in space-time. He will also study the robustness of entropy for parameter estimation and compare performance with SVMs. But the large appeal of the novel pair wise interaction model for learning is that it opens drastically new opportunities to understand, apply and implement adaptive systems. He will be studying the following applications of the pair wise interaction model:

1- Model the information processing in the dendritic tree as an spatio-temporal interaction field that estimates entropy. Seek on-line algorithms and hopefully show their relation to Hebbian. Our preliminary results are very encouraging. Dr. Henry Markram, a world expert in dynamic synapses will help establish the biological plausibility of the model.

2- Formulate state space models of linear and nonlinear systems with our entropy estimator. This work will extend the field of minimum entropy control for non-Gaussian processes and will pro-vide a new approach to implement robust control.

3- Apply the entropy estimator to blind deconvolution of wireless channels as well as multiuser detection. The PI will extend his current work on blind source separation for convolutive mixtures, the more realistic (but also harder) case. In particular he will target the cocktail party effect in teleconferencing.

This project will attempt the first true "reverse engineering" of the behavior of cells from the cerebral cortex studied outside the brain, grown on a special chip which allows dense read-ins and read-outs from a network of these cells. The cerebral cortex is the largest part of the human brain (and of many other mammalian brains). Reverse engineering, if successful, would give us the first understanding of how these cells work as a functional, engineering system at the circuit level.
Because of its uniqueness and novelty, there will be serious risks in this work.
One of the collaborators will be comparing the behavior of these networks of cells against his new model of neurons, the liquid state model (LSM). Another will be comparing the cells and the LSM against general-purpose learning challenges which have been addressed using artificial neural networks, for real-world engineering tasks such as prediction and control and classification and so on. The crossdisciplinary and international collaborations play a crucial role in making this new direction for research possible.

Two related DDDAS application areas considered in this project are (1) cognitive brain modeling from experiments with live subjects and (2) the design of brain-inspired assistive systems to help human beings with severe motor behavior limitations (e.g. paraplegics) through brain-machine interfaces (BMIs). Simply stated, a BMI uses brain signals to directly control devices such as computers and robots. Today's BMI designs are extremely primitive and are a far cry from the seamless interface between brain and body in animals. In a healthy animal, the brain constantly learns and adapts to the needs of new physical movement, in addition to providing perfectly timed signals to the motor system. In this process, the brain receives and uses sensory feedback to both learn and generate the signals that lead to purposeful motion. In order to inch closer to BMI designs that are of use to humans, better models of brain motor control and movement planning are needed along with the necessary adaptive algorithms and computational architecture needed for their execution in real time. In light of such goals, this project's activities aim to significantly advance the state of the art of BMI research by developing the models, algorithms and computational architecture of dynamically-data-driven BMIs (DDDBMIs) that implement recently proposed advanced brain models of motor control. Achieving this goal in the proposed approaches will also allow to address a chief problem in current BMI research: The fact that paraplegics cannot train their own network models because they cannot move their limbs.

The research on DDDBMI systems conducted under this project is a drastic departure of the conventional BMI paradigm. The control interface architecture is distributed and borrowed from recent models of neurophysiology of movement, which will enable better overall performance. Learning occurs simultaneously for the subject and the control models in a synergistic manner, which requires more powerful adaptation schemes. Selective use of many computational models is the reason why a dynamically data-driven system is needed to provide the computational needs of a DDDBMI. The project interdisciplinary activities are closely intertwined around the development and integration of the DDDBMI components into a platform for BMI research. Research on middleware addresses the need for dynamic aggregation of Grid-resources with Quality-of-Service guarantees, and support for dynamic computation steering. Research on adaptive algorithms focuses on new data models and learning algorithms. Research on brain modeling concentrates on cognitive models of motor control and advancing our understanding of the neurobiology of movement. In the long run, the BMI experimental research platform will have a dual role: it will help validate the brain models under investigation and it will provide insights on to how to design BMIs for use by paraplegic patients.

Principe Abstract

The objective of this research is to understand further the optimization of nonlinear systems with cost functions based on information theory. The recent interpretation of entropy as the mean square of the projected samples in feature space provides a link to the theory of reproducing kernels Hilbert spaces (RKHS), and raises the hypothesis that it may be possible to analytically compute the optimal solution of nonlinear systems, unlike the current estimates that use search procedures. The solution of the two most widely used models in filtering, the Wiener and the Kalman filters will be addressed. The approach is also novel because it will exploit both the inner product structure of the RKHS and the geometry of the estimation process using a differential geometry approach.

Intellectual Merits. The intellectual merit of the proposal is to propose a new methodology based on information geometry to adapt systems with cost functions that directly manipulate information in the data, with the expected outcome of improving performance over the conventional methods in creating models from data and providing understanding of data.

Broader Benefits. Our technology driven world is creating data at alarming rates. However, humans are interested in information, not data, and this is creating a tremendous bottleneck in medicine, business and even in engineering. Adaptive systems are one of the most promising methods to create models from data and provide understanding. The PI will also educate a breed of graduate students in the new area of differential geometry applied to signal processing who are needed to help solve this information bottleneck.

0541241
PI: John G. Harris
University of Florida

A Spike-Based Computer Architecture for Sensory Processing

The PIs will develop, study, and build a neurobiologically inspired architecture based on neuronal spikes (or action potentials). The proposed architecture combines previous spike-based sensory processing ideas developed by the PIs with a compelling model of brain computation called the Liquid State Machine (LSM). This model provides a conceptual framework for working with biologically realistic pulsed neuron models (integrate-and-fire neurons) as the basic computational element within a recurrent nonlinear architecture. The PIs propose three key steps for developing networks of spiking processing elements that are useful for computation:
1. The PIs will develop a sampling theory for converting continuous variable inputs into aperiodic spike trains (and likewise transform spikes trains back to continuous amplitude signals).
2. Although interesting, the architecture of the LSM can be largely improved once the characteristics of the computation are better understood. In particular, while the interconnect of the liquid is arbitrary and fixed, the PIs plan to develop a theory of adaptation for such spike based representations.
3. The PIs will map the spike-based architecture to todays silicon electronics. This hybrid analog/digital architecture is fundamentally different from both conventional digital architectures and past analog computing devices. However, the proposed architecture will be of only scientific interest if it cannot be built with competitive performance measures in terms of computational capability, power consumption, noise immunity and dynamic range. Therefore, in order to design the architecture in silicon, the PIs will identify key computational principles, design signal transformations, and fabricate the resulting building blocks with huge numbers of sufficiently small components in CMOS technology.
Finally, in order to ground the theoretical research and design in engineering practice, the PIs propose to develop a prototype system for chemical sensing using an off-the-shelf electronic nose (or e-nose) as the front end. This application was chosen due to its importance in homeland security, and also because it provides a natural input to the proposed spike-based architecture.

This Partnerhships for Innovation (PFI) project seeks to create entrepreneurial activities around a well established research program in Brain Machine Interfaces at the University of Florida. Neural interfaces have the potential to revolutionize the present interface with computers and man-made appliances, as well as play a crucial role in rehabilitation medicine. Due to the novelty of this field, it is timely to invest now in entrepreneurial and innovative activities, because they provide the opportunity to leverage the innovations and intellectual property, mobilize investors, and develop lines of products that will guarantee self-sufficiency for the Center of Innovative Brain Machine Interfaces. Project collaborators within the university are the College of Engineering, McKnight Brain Institute (UFMBI), and Shands and VA Medical Centers. The College of Engineering has extensive experience in entrepreneurial and technology transfer programs and has created a successful undergraduate student training program designed around the concept of "virtual start-up" companies. Capitalizing on these two experiences, the Partnerships project proposes to restructure the ongoing research in Brain Machine Interfaces around technology test beds. The goal of this restructuring is translation of the research mission into core technologies and competencies with a wrapper of immersive, experimental entrepreneurship education for the graduate students that are engaged in the research. Specifically, six high-technology test beds (multi electrode arrays, ultra low power bio amplifiers, wireless delivery of data/power, portable DSP systems and algorithms, brain computer interfaces) will be created. Technology advances in neural interfaces will be accelerated.

The broader impact will be felt at several levels: 1) this is a new educational experience since graduate students will be engaged in entrepreneurial activities while at the university; 2) a more transparent way of implementing technology transfer between universities and industry that can be widely applied is being prototyped; 3) the creation of a high tech industry in the state of Florida is being seeded; and 4) an underrepresented institution and minority students are being involved in hands-on courses that teach the technology components of an emerging industry. A plan has been outlined to make the center financially self-sufficient after the end of NSF support.

Partners: Partners include University of Florida (Lead Institution); Florida International University in Miami, FL, a designated minority institution, where one of the test beds will reside; four Companies, which have joined the Industrial Board (Advanced Neuromodulation Systems (ANS), Convergent Engineering, Tucker-Davis Technologies, and Motorola Labs); Sid Martin Development Incubator; City of Gainesville Department of Economic Development; Gainesville Chamber of Commerce; Technology Enterprise Center of Gainesville, Alachua County; Inflexion Partners(Venture Capitalists); and Saliwanchik, Lloyd & Saliwanchik (law firm).

Proposal #: CNS 08-21622
PI(s): Fortes, Jose A.
McIntyre, Lauren M.; Moroz, Leonid L.; Principe, Jose C.; Sanchez, Justin C.
Institution: University of Florida
Gainesville, FL 32611-2002
Title: MRI/Acq.: Instrumentation for Coupled Experimental-Computational Neuroscience and Biology Research

Project Proposed:
This project, acquiring a virtualized multicomputer instrument with shared-memory subsystems and storage capacity, intends to use this instrument as a shared instrument whose resources can be virtualized, reserved, and configured on demand for different research activities related to neuroscience, computational biology, and cyberinfrastructure. Its configurations can also be dedicated and tightly coupled to in vivo experiments using network connections to in situ instrumentation used for experiments. It can simultaneously support multiple research activities because of its unique capability to support real-time computer-in-the-loop experiments, its ability to run many concurrent computation threads, its shared memory and storage subsystems, and its use of virtualization technology to manage the coexistence of multiple computing environments. To develop and validate autonomic computing nodes and techniques for multiuser virtualized computational systems, the instrument provides traces of performance and other needed monitored data. Research activities in brain-machine interfaces, neurogenesis, genomics, bioinformatics, signal processing, cyberinfrastructure, autonomic computing and other areas include:
- Brain-machine interfaces where cortex models for motor control are dynamically learned and applied in real-time,
- Experimental drug discovery through real-time analysis of large amounts of genetic data and many thousands of compounds,
- Analysis of Terabytes of genetic data captured in real-time as a neuron grows, learns, and remembers, and
- Online learning algorithms using dynamic filter topologies with online computation requirements that increase over time.
The activities have transformative goals, including the introduction of real-time, and address the high performance computation into closed-loop experiments and/or systems whose behavior is driven by complex processing of sensed data. Another goal involves providing off-line computing capabilities to match the unprecedented rate and volume of genetic data produced by massively parallel DNA sequencing technology.

Recently, there has been an increased interest within the machine learning and signal processing communities in kernel methods because they offer an attractive alternative to design nonlinear systems for the demanding current applications of information processing. Recursive state estimation and in particular the Kalman filter, would benefit from such an effort because there are many important applications in aerospace, automotive, surveillance, and medical fields that are intrinsically nonlinear, and the current nonlinear models have drawbacks.
This proposal will study the feasibility of a (nonlinear) mapping to a linear functional space (a Reproducing Kernel Hilbert Space) to implement there the Kalman filter equations and achieve performance commensurate with nonlinear models. One of the difficulties of this approach that will be investigated is the growing memory requirements that will be dealt with novel sparsification criteria and algorithms based on information theory.
The approach will be tested in the design of brain machine interfaces to help quadraplegics and also in automotive engine control for which there are data available and benchmarks. Two graduate students will be working on this important topic.

A key hurdle in studies of brain function is to be able to measure not only what signals are correlated with one another, but also how they are causally related. Correlation quantifies linear dependence, while causality is capable of distinguishing which brain area is leading the correlated counterparts; causality puts an arrow into correlation. Causality is a difficult problem in data analysis and here a novel measure of conditional statistical dependence to evaluate causality is proposed. The ultimate practical goal is to elucidate the principles of cognitive processing and provide online cognitive feedback to human subjects performing complex tasks.

The objective of this project is to use a recently developed paradigm for electroencephalogram (EEG) quantification based on periodic visual stimulation to improve the signal to noise ratio of visual stimulation on a pre-determined EEG frequency band (here around 10 Hz). The goal is to develop advanced signal processing techniques based on instantaneous frequency (Hilbert transform) to quantify the instantaneous amplitude of a visual stimulus in 32 channels over the scalp.

A recently developed measure of local statistical dependence in the joint space called correntropy will be utilized to evaluate the dependency among instantaneous amplitude time series collected over the scalp. The maximum value of correntropy is a measure of statistical dependence, which is the first step towards causality. To achieve a causality measure, conditional dependence will be evaluated by extending correntropy to conditional correntropy, first for triplets of variables and them to subspaces of arbitrary dimensions. Correntropy is a nonparametric measure of dependence; hence, the new method will be compared to linear and nonlinear Granger causality methods implemented in reproducing kernel Hilbert spaces.

These algorithms will be tested on data collected from human subjects in a study of affective visual perception. The goal is to study and quantify the re-entry hypothesis of emotional perception -- that re-entrant modulation originating from higher-order cortices is responsible for enhanced activation in the occipital cortex when emotionally arousing stimuli are perceived. The signal processing and statistical methods developed here will provide a way to identify dependent EEG channels and causal relationships amongst them during the presentation of the stimulus, effectively tracing the flow of neural activity from the stimulated visual areas to frontal areas and back to the visual cortex.

This project proposes to establish the NSF I/UCR Center for Big Learning (CBL). The mission of CBL is to pioneer in large-scale deep learning algorithms, systems, and applications through unified and coordinated efforts in the CBL consortium. The vision of CBL is to create intelligence enablers towards intelligence-driven society. With the explosive big data generated from natural systems, engineered systems, and human activities, we need intelligent algorithms and systems to facilitate our decision making with distilled insights automatically at scale. The proposed CBL center is a timely initiative as our society moves towards intelligence-enabled world of opportunities. The CBL consortium is expected to become the magnet of deep learning research and applications and attract leading researchers, enthusiastic entrepreneurs, IT and industry giants working together on accomplishing the promising mission and vision. This planning grant will lead to a successful proposal for the establishment of the NSF I/UCR Center for Big Learning with a solid consortium across multiple campuses and a large number of industry partners.

CBL has the following broader impacts. (1) Making significant contributions and impacts to the deep learning community on pioneering research and applications to address a broad spectrum of real-world challenges. (2) Making significant contributions and impacts to promote products and services of industry in general and our members in particular. (3) Making significant contributions and impacts to the urgently-needed education of our next-generation talents with real-world settings and world-class mentors from both academia and industry. (4) Our meetings, forums, conferences, and planned training sessions will greatly promote and broaden the research and materialization of DL.



The proposed project aims to establish the NSF I/UCR Center for Big Learning (CBL). With dramatic breakthroughs in multiple modalities of challenges (e.g., image, video, speech, text, and Q&A), the renaissance of machine intelligence is looming.The mission of CBL is to pioneer in large-scale deep learning (DL) algorithms, systems, and applications through unified and coordinated efforts in the CBL consortium via fusion of broad expertise from our large number of faculty members, students, and industry partners. The vision of CBL is to create intelligence enablers towards intelligence-driven society. CBL possesses the pioneering intellectual merit in the following key research themes. (1) Novel algorithms. This theme focuses on novel DL algorithms and architectures, such as deep architecture, complex deep neural networks, brain-inspired components, optimization, deep reinforcement learning, and unsupervised learning. (2) Novel systems. We propose novel architectures, resource management, and software frameworks for enabling large-scale DL platforms and applications on desktops, mobiles, clusters, and clouds. (3) Novel applications in health, mobile/IoT, and surveillance. During the planning phase, we will establish a solid center strategic plan, marketing plan, and the CBL consortium that consists of four academic sites and a large number of industrial members.

CBL has the following broader impacts. (1) Making significant contributions and impacts to the deep learning community on pioneering research and applications to address a broad spectrum of real-world challenges. (2) Making significant contributions and impacts to promote products and services of industry in general and our members in particular. (3) Making significant contributions and impacts to the urgently-needed education of our next-generation talents with real-world settings and world-class mentors from both academia and industry. (4) Our meetings, forums, conferences, and planned training sessions will greatly promote and broaden the research and materialization of DL.

Most animals survive in turbulent air or water environments and are living proof that it is possible to quantify odor signals in complex turbulent flow conditions to track and find sources of odors (such as food, mates, etc.). However, our engineering knowledge is still incapable of formulating simple and effective measurements that will enable man-made systems to predict, navigate and utilize properties of this turbulent flow to locate sources of chemical release. This project builds on recent exciting computational modeling of the neurobiology of organisms by the proposers, which predict that lobsters are capable of estimating not only the concentration of odors but also the time since the last odor was encountered. Lobsters accomplish this by using ensemble competition across a population of olfactory receptor neurons (ORNs), called "bursting ORNs". Bursting ORNs function to compute the time since last encounter of an odor that, along with concentration, can provide a measure of the distance to the odor source. This research will seek to increase understanding of how ORNs perceive odor concentration and intermittency measured within an odor plume, and how this information is integrated within the lobster?s brain. An additional goal is to develop new neurobiology-based theories in the search for odor sources that can be implemented within human-engineered autonomous underwater vehicles that have the ability to navigate in turbulent chemical plumes.

The broader implications of this work stem from the large potential market for defense and civilian applications of a new generation of electronic noses for tracking chemicals in natural or man-initiated disasters. Through this project, there are also excellent resources and outreach opportunities for integrated education and training of students at the intersection of fluid dynamics, neuroscience, computer engineering and information processing. Outreach will be coordinated through the Center of Innovative Brain Machine Interfaces at the University of Florida and will provide opportunities for undergraduate and graduate research, promote neurotechnology innovations, and foster entrepreneurship activities in order to create potential future start-up companies.

The research will include laboratory experiments of chemical plume mixing and ORN responses to odor encounters by lobsters, theoretical analysis of search optimization, as well as numerical simulations and novel system architecture for electronic noses. This research brings together a multidisciplinary and complementary team of experts, including a fluid dynamicist, a neurobiologist, and an electrical engineer with the very clear goal of understanding and exploiting olfactory scene analysis in turbulent flow. In this new light, neurobiologists will understand new sensing strategies for olfaction, and engineers can improve the quantification of turbulent mixing and replicate these sensory strategies to propose novel bio-inspired sensors capable of quantifying the dispersion of chemical plumes and improve the search for the source.

This award studies the ways to monitor the most common physiological variables (heart rate, blood pressure, respiration and brain activity) with miniature devices that can harvest energy from the body instead of being powered by batteries, and using electrodes that can be applied to the skin as a "tattoo". This next generation of Mobile Health (mHealth) devices will improve further the national health and wellbeing. A major bottleneck towards the goal is how to decrease the power consumption of algorithms required to extract information from the collected signals. The aim of this project is to design, implement and validate a new ultra-low power signal processing solution that does not require digital computers, but much simpler digital devices driven by input pulse trains. The project will also train two graduate students in the theory and technology to design the next generation of biomedical devices.

The award will develop new pulse based algorithms and a reconfigurable hardware platform that amplifies, converts and quantifies structure of the signals in real time. More specifically, the research plan includes two synergestic aims: the first aim develops a mathematical framework based on signal processing and a statistical-syntactic approach to learn directly from data the structure of the input. A nonlinear state model called KAARMA (kernel adaptive autoregressive moving average model) will be trained statistically from data to recognize events with clinical significance. Once trained, KAARMA can be converted in a combination of finite state machines and memory tables that can easily be implemented in ultra-low power reconfigurable digital logic platform to design ambulatory monitoring of physiological variables for mHealth. No digital signal processors are needed in the deployed proposed device, lowering power consumption, maintaining programmability and the quality of the digital extraction of information. The second aim is to design an ultra-low power reconfigurable analog front-end sensing integrated circuit using mainly digital standard cells to implement a variable number of channels, multipurpose analog amplification and filtering, and the finite state machines. The expected goal is to demonstrate power consumption of less than 5 microwatts to analyze one channel of electrocardiogram (ECG). The KAARMA will be extended to blood pressure, respiration and brain activity. Validation with competing technologies will be conducted in the Physionet database.

The goal of this project is to develop technology that can improve patient care and increase patient safety during pandemics such as COVID-19 that flood medical facilities. This project will develop an inexpensive, portable, in-circuit respiratory monitoring technology that enables safe and effective care of multiple patients needing ventilation during a crisis when facilities and clinicians are stretched thin. The system can provide remote alarms and monitoring of any ventilated patient -- saving clinical time and precious personal protective equipment (PPE) required to enter a COVID-19 isolation room, as well as dramatically increasing safety when using emergency ventilators. The system will also be designed to provide decision support in a rapidly changing environment and can collect important data for analysis of patient physiology to also improve care. The system also includes the ability to individualize treatment and monitoring to provide safe and effective ventilation of multiple patients on a single ventilator.

The planned design is the result of the team's long experience with ventilator and respiratory monitor design and is based on existing medical device software and proven ventilator technology. The system has been designed to use novel algorithms and sensing, while being rapidly manufactured with off-the-shelf components where possible. Other critical components that may be unavailable due to the crush of the pandemic have been designed to be easily 3D printed or injection molded, based on existing, proven designs. The team propose to build a monitoring and control device using an inexpensive tablet as the user interface and a smart sensor using Internet of Things (IOT)-based processor boards to monitor and collect data and provide safety advice to the over-worked clinicians. A safe, easily manufactured, modified pneumatic exhalation valve controlled by the monitoring system with pressure from the airway circuit itself will be added to individualize treatment for patients sharing a ventilator. This project provides the monitoring and control to safely and effectively use existing tools to ventilate patients. Beyond its pandemic uses, next generation versions of the technology can be used in isolation rooms of the future and also in resource-limited areas of the country and the world.

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.