MOTIVATION: Recapitulating aspects of human organ functions using in vitro (e.g. plates, transwells, etc.), in vivo (e.g. mouse, rat, etc.), or ex vivo (e.g. organ chips, 3D systems, etc.) organ models is of paramount importance for drug discovery and precision medicine. It will allow us to identify potential side effects and test the effectiveness of new therapeutic approaches early in their design phase, and will inform the development of better disease models. Developing mathematical methods to reliably compare the 'distance/similarity' of organ models from/to the real human organ they represent is an understudied problem with important applications in biomedicine and tissue engineering. RESULTS: We introduce the Transcriptomic Signature Distance (TSD), an information-theoretic distance for assessing the transcriptomic similarity of two tissue samples, or two groups of tissue samples. In developing TSD, we are leveraging next-generation sequencing data as well as information retrieved from well-curated databases providing signature gene sets characteristic for human organs. We present the justification and mathematical development of the new distance and demonstrate its effectiveness and advantages in different scenarios of practical importance using several publicly available RNA-seq datasets. AVAILABILITY AND IMPLEMENTATION: The computation of both TSD versions (simple and weighted) has been implemented in R and can be downloaded from https://github.com/Cod3B3nd3R/Transcriptomic-Signature-Distance. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Time-lapse microscopy live-cell imaging is essential for studying the evolution of bacterial communities at single-cell resolution. It allows capturing detailed information about the morphology, gene expression, and spatial characteristics of individual cells at every time instance of the imaging experiment. The image analysis of bacterial "single-cell movies" (videos) generates big data in the form of multidimensional time series of measured bacterial attributes. If properly analyzed, these datasets can help us decipher the bacterial communities' growth dynamics and identify the sources and potential functional role of intra- and inter-subpopulation heterogeneity. Recent research has highlighted the importance of investigating the role of biological "noise" in gene regulation, cell growth, cell division, etc. Single-cell analytics of complex single-cell movie datasets, capturing the interaction of multiple micro-colonies with thousands of cells, can shed light on essential phenomena for human health, such as the competition of pathogens and benign microbiome cells, the emergence of dormant cells ("persisters"), the formation of biofilms under different stress conditions, etc. However, highly accurate and automated bacterial bioimage analysis and single-cell analytics methods remain elusive, even though they are required before we can routinely exploit the plethora of data that single-cell movies generate. RESULTS: We present visualization and single-cell analytics using R (ViSCAR), a set of methods and corresponding functions, to visually explore and correlate single-cell attributes generated from the image processing of complex bacterial single-cell movies. They can be used to model and visualize the spatiotemporal evolution of attributes at different levels of the microbial community organization (i.e., cell population, colony, generation, etc.), to discover possible epigenetic information transfer across cell generations, infer mathematical and statistical models describing various stochastic phenomena (e.g., cell growth, cell division), and even identify and auto-correct errors introduced unavoidably during the bioimage analysis of a dense movie with thousands of overcrowded cells in the microscope's field of view. CONCLUSIONS: ViSCAR empowers researchers to capture and characterize the stochasticity, uncover the mechanisms leading to cellular phenotypes of interest, and decipher a large heterogeneous microbial communities' dynamic behavior. ViSCAR source code is available from GitLab at https://gitlab.com/ManolakosLab/viscar .

Time-lapse microscopy provides 4D imaging data for monitoring and studying down to single-cell, the stochastic processes involved as bacterial colonies grow and interact under different stress conditions. Two main factors prevent high throughput analysis: a) cell segmentation and tracking are very time-consuming and error-prone and b) analytics tools are lacking to interpret the plethora of features extracted from a complex “cell-movie.” To address both limitations, we have recently developed a multi-resolution Bio-image Analysis & Single-Cell Analytics framework, called BaSCA. Our end-to-end pipeline segments accurately single-cells and tracks them across frames, identifying cell divisions and constructing each cell colony's lineage and division tree. Here we present the capabilities of an R- package, which enables users to correlate and visually explore the spatiotemporal trends of single-cell attributes, discover and study epigenetic effects across many cell generations and even identify and correct segmentation errors. All these unique tools empower research towards deciphering microbial community dynamic behaviors, such as inter-species competition/cooperation, “persister” (dormant) cells stochastic emergence, etc., with a huge impact on human health.

System Python (SysPy) is a public domain design tool using Python to facilitate all prototyping phases of processor-centric SoCs for FPGAs. In previous work we used Python as a high-level description mechanism to design hardware modules and connect them to embedded processors. In this paper we show how SysPy can also facilitate high-level functional verification of a SoC when used as an Architectural Description Language (ADL), helping a designer make decisions about key architectural features early in the design phase. To the best of our knowledge our tool is unique in supporting algorithmic joint modeling of hardware and software elements of a SoC (processing and control logic) using popular Python libraries such as SciPy and NumPy following a Matlab-like syntax. The tool also supports C software development and hw/sw co-verification as it automatically compiles and executes the user's C application code for the processor core. Other important features are the generation of Value Change Dump (VCD) files for visualizing signal waveforms using popular simulation tools, and of IP-XACT metadata models for presenting hardware blocks using alternative views (Python simulation models, RTL descriptions, IP-XACT models).

Stochastic simulation of large-scale biochemical reaction networks, with thousands of reactions, is important for systems biology and medicine since it will enable the insilico experimentation with genome-scale reconstructed networks. FPGA based stochastic simulation accelerators can exploit parallelism, but have been limited on the size of biomodels they can handle. We present a high performance scalable System on Chip architecture for implementing Gibson and Bruck's Next Reaction Method efficiently in reconfigurable hardware. Our MPSoC uses aggressive pipelining at the core level and also combines many cores into a Network on Chip to also execute in parallel stochastic repetitions of complex biomodels, each one with up to 4K reactions. The performance of our NRM core depends only on the average outdegree of the biomodel's Dependencies Graph (DG) and not on the number of DG nodes (reactions). By adding cores to the NoC, the system's performance scales linearly and reaches GCycles/sec levels. We show that a medium size FPGA running at ~200 MHz deliver high speedup gains relative to a popular and efficient software simulator running on a very powerful workstation PC.

There is a fast growing interest in exploiting wireless sensor networks (WSNs) for tracking the boundaries and predicting the evolution properties of diffusive hazardous phenomena (e.g. wildfires, oil slicks etc.) often modeled as “continuous objects”. We present a novel distributed algorithm for estimating and tracking the local evolution characteristics of continuous objects. The hazard's front line is approximated as a set of line segments, and the spatiotemporal evolution of each segment is modeled by a small number of parameters (orientation, direction and speed of motion). As the hazard approaches, these parameters are re-estimated using ad-hoc clusters (triplets) of collaborating sensor nodes. Parameters updating is based on algebraic closed-form expressions resulting from the analytical solution of a Bayesian estimation problem. Therefore, it can be implemented by microprocessors of the WSN nodes, while respecting their limited processing capabilities and strict energy constraints. Extensive computer simulations demonstrate the ability of the proposed distributed algorithm to estimate accurately the evolution characteristics of complex hazard fronts under different conditions by using reasonably dense WSNs. The proposed in-network processing scheme does not require sensor node clocks synchronization and is shown to be robust to sensor node failures and communication link failures, which are expected in harsh environments.

Assessing how the performance of a decentralized wireless sensor network (WSN) algorithm's implementation scales, in terms of communication and energy costs, as the network size increases is an essential requirement before its field deployment. Simulations are commonly used for this purpose, especially for large-scale environmental monitoring applications. However, it is difficult to evaluate energy consumption, processing and memory requirements before the algorithm is really ported to a real WSN platform. We propose a method for emulating the operation of collaborative algorithms in large-scale WSNs by re-using a small number of available real sensor nodes. We demonstrate the potential of the proposed simulation-driven WSN emulation approach by using it to estimate how communication and energy costs scale with the network's size when implementing a collaborative algorithm we developed in [12] for tracking the spatiotemporal evolution of a progressing environmental hazard.

We present SysPy (System Python) a tool which exploits the strengths of the popular Python scripting language to boost design productivity of embedded System on Chips for FPGAs. SysPy acts as a “glue” software between mature HDLs, ready-to-use VHDL components and programmable processor soft IP cores. SysPy can be used to: (i) automatically translate hardware components described in Python into synthesizable VHDL, (ii) capture top-level structural descriptions of processor-centric SoCs in Python, (iii) implement all the steps necessary to compile the user's C code for an instruction set processor core and generate processor specific Tcl scripts that import to the design project all the necessary HDL files of the processor's description and instantiate/connect the core to other blocks in a synthesizable top-level Python description. Moreover, we have developed a Hardware Abstraction Layer (HAL) in Python which allows user applications running in a host PC to utilize effortlessly the SoC's resources in the FPGA. SysPy's design capabilities, when complemented with the developed HAL software API, provide all the necessary tools for hw/sw partitioning and iterative design for efficient SoC's performance tuning. We demonstrate how SysPy's design flow and functionalities can be used by building a processor-centric embedded SoC for computational systems biology. The designed SoC, implemented using a Xilinx Virtex-5 FPGA, combines the flexibility of a programmable soft processor core (Leon3) with the high performance of an application specific core to simulate flexibly and efficiently the stochastic behavior of large size biomolecular reaction networks. Such networks are essential for studying the dynamics of complex biological systems consisting of multiple interacting pathways.

We designed a variety of k-nearest-neighbor parallel architectures for FPGAs in the form of parameterizable soft IP cores. We show that they can be used to solve large classification problems with thousands of training vectors, or thousands of vector dimensions using a single FPGA, and achieve very high throughput. They can be used to flexibly synthesize architectures that also cover: 1NN classification (vector quantization), multishot queries (with different k ), LOOCV cross-validation, and compare favorably to GPU implementations. To the best of our knowledge this is the first attempt to design flexible IP cores for the popular kNN classifier.

A learning scheme based on random forests is used to discriminate between different reach to grasp movements in 3-D space, based on the myoelectric activity of human muscles of the upper-arm and the forearm. Task specificity for motion decoding is introduced in two different levels: Subspace to move toward and object to be grasped. The discrimination between the different reach to grasp strategies is accomplished with machine learning techniques for classification. The classification decision is then used in order to trigger an EMG-based task-specific motion decoding model. Task specific models manage to outperform "general" models providing better estimation accuracy. Thus, the proposed scheme takes advantage of a framework incorporating both a classifier and a regressor that cooperate advantageously in order to split the task space. The proposed learning scheme can be easily used to a series of EMG-based interfaces that must operate in real time, providing data-driven capabilities for multiclass problems, that occur in everyday life complex environments.

Reaching and grasping of objects in an everyday-life environment seems so simple for humans, though so complicated from an engineering point of view. Humans use a variety of strategies for reaching and grasping anything from the simplest to the most complicated objects, achieving high dexterity and efficiency. This seemingly simple process of reach-to-grasp relies on the complex coordination of the musculoskeletal system of the upper limbs. In this paper, we study the muscular co-activation patterns during a variety of reach-to-grasp motions, and we introduce a learning scheme that can discriminate between different strategies. This scheme can then classify reach-to-grasp strategies based on the muscular co-activations. We consider the arm and hand as a whole system, therefore we use surface ElectroMyoGraphic (sEMG) recordings from muscles of both the upper arm and the forearm. The proposed scheme is tested in extensive paradigms proving its efficiency, while it can be used as a switching mechanism for task-specific motion and force estimation models, improving EMG-based control of robotic arm-hand systems.

Simulation of biomolecular networks with thousands of reactions is becoming essential for systems biology. We are presenting the design of a scalable System on Chip parallel architecture that implements Gillespie's First Reaction Method in reconfigurable FPGA hardware. Our SoC architecture can deliver performance (Mega-Reactions/sec) and throughput (M-Reaction cycles/sec) that is increasing linearly with the number of processors when simulating large biomolecular networks with up to m = 4096 reactions using a moderate size FPGA. We have synthesized and verified various SoC instances with up to N=8 Processing Elements for Xilinx Virtex 5 and Altera Cyclone III FPGAs, reaching clock frequencies up to 180 MHz and delivering simulation performance that is more than 2 order of magnitude higher than that of Intel Core 2 and i7 CPUs running at frequencies above 2GHz.

Predicting accurately the spatiotemporal evolution of a diffusive environmental hazard is of paramount importance for its effective containment. We approximate the front line of a hazard with a set of line segments (local front models). We model the progression characteristics of these front segments by appropriately modified 2D Gaussian functions. The modified Gaussian model parameters are adjusted based on the solution of a Kullback-Leibler (KL) divergence minimization problem. The whole scheme can be realized by a wireless sensor network by forming dynamically triplets of cooperating sensor nodes along the path of the hazard. It is shown that the algorithm can track effectively the front characteristics (in terms of direction and speed) even in the presence of faulty sensor nodes.

Image analysis is still considered as the bottleneck in 2D-gel based expression proteomics analysis for biomarkers discovery. We are presenting a new end-to-end image analysis pipeline of operations that can be fully automated. The pipeline includes image denoising and enhancement based on contourlets, image segmentation into Regions of Interest (ROIs) based on active contours, followed by the analysis of the extracted ROIs for spot detection and quantification using mixture modeling, model selection and unsupervised machine learning methods. The proposed system is shown to match the sensitivity and exceed the precision of commercial spot detection software when analyzing real 2D gel images. It is suitable for high throughput proteomics analysis of image stacks since, unlike commercial software, it does not require any manual re-calibration of parameters every time a new image is to be processed.

We present a novel decentralized Wireless Sensor Network (WSN) algorithm which can estimate both the speed and direction of an evolving diffusive hazardous phenomenon (e.g. a wildfire, oil spill, etc.). In the proposed scheme we approximate a progressing hazard's front as a set of line segments. The spatiotemporal evolution of each line segment is modeled by a modified 2D Gaussian function. As the phenomenon evolves, the parameters of this model are updated based on the analytical solution of a Kullback - Leibler (KL) divergence minimization problem. This leads to an efficient WSN distributed parameters estimation algorithm that can be implemented by dynamically formed clusters (triplets) of collaborating sensor nodes. Computer simulations show that our approach is able to track the evolving phenomenon with reasonable accuracy even if a percentage of sensors fails due to the hazard and/or the phenomenon has a time varying speed.

BACKGROUND: The steps of a high-throughput proteomics experiment include the separation, differential expression and mass spectrometry-based identification of proteins. However, the last and more challenging step is inferring the biological role of the identified proteins through their association with interaction networks, biological pathways, analysis of the effect of post-translational modifications, and other protein-related information. RESULTS: In this paper, we present an integrative visualization methodology that allows combining experimentally produced proteomic features with protein meta-features, typically coming from meta-analysis tools and databases, in synthetic Proteomic Feature Maps. Using three proteomics analysis scenarios, we show that the proposed visualization approach is effective in filtering, navigating and interacting with the proteomics data in order to address visually challenging biological questions. The novelty of our approach lies in the ease of integration of any user-defined proteomic features in easy-to-comprehend visual representations that resemble the familiar 2D-gel images, and can be adapted to the user's needs. The main capabilities of the developed VIP software, which implements the presented visualization methodology, are also highlighted and discussed. CONCLUSIONS: By using this visualization and the associated VIP software, researchers can explore a complex heterogeneous proteomics dataset from different perspectives in order to address visually important biological queries and formulate new hypotheses for further investigation. VIP is freely available at http://pelopas.uop.gr/~egian/VIP/index.html.

We present the systematic design of two linear array IP cores for the k-nearest neighbor (k-NN) benchmark classifier. The need for real-time classification of data vectors with possibly thousands of features (dimensions) motivates the implementation of this widely used algorithm in hardware in order to achieve very high performance by exploiting block pipelining and parallel processing. The two linear array architectures that we designed have been described as soft IP cores in fully parameterizable VHDL that can be used to synthesize effortlessly different k-NN parallel architectures for any desirable combination of the problem size parameters. They have been evaluated for a large variety of parameter combinations and Xilinx FPGAs. It is shown that they can be used to solve efficiently very large size k-NN classification problems, even with thousands of training vectors or vector dimensions, using a single, moderate size FPGA device. Furthermore the FPGA implementations exceed by a factor of two the performance of optimized NVIDIA CUDA API software implementations for the powerful GeForce 8800GTX GPU.

The post-genomic era is characterized by the rapid data accumulation leading to unwieldy and large volumes of biological data. The proteomics results (large sets of identified proteins or peptides) that originate from several workflow steps, play an important role in the analysis of a proteomics experiment. As a result, the area of high-through-put proteomics created new visualization challenges in interpreting large-scale datasets. We present VIP, an interactive visualization tool for proteomics data, which integrates protein or peptide features emanating at every step of the proteomics analysis and combines them visually in synthetic maps. Our novel tool offers the flexibility to choose any desired features according to the analysis objectives, examine simultaneously more than one maps and interact with the visualization by querying and filtering the results. The synthetic maps visualization aims not only at summarizing proteomics experiments in a unified manner for both 2DE-MS and LC-MS based analyses, but also at providing a quick and combined overview of protein/peptide features, thus facilitating the data analysis and interpretation.

We present ProtCV (protein clustering and visualization) a new software tool for grouping samples (mass spectra peak-lists) emanating from a high throughput proteomics analysis based on their spectral similarities, and summarizing effectively for the user the clustering results using advanced visualization methods. A unique feature of ProtCV is that it can compare clustering methods applied to the same data set based on several validity in dices to assist the user identify groupings that seem to capture best the underlying structure of the data set. Moreover, ProtCV can assist in formulating hypotheses about the potential role of unidentified proteins in a cluster, identify sets of proteins which act jointly at a specific biological state etc. All these data set mining and exploration operations are very useful for interpreting the results of high throughput MS based proteomics analyses that are commonly used for biomarkers discovery.

A dynamically self-reconfigurable master-slaves MPSoC architecture framework is introduced which can be fully embedded into a single FPGA device. The master core can request a configuration manager module to add, or remove, a slave core at runtime. If the request can be satisfied, self-reconfiguration commences, implemented by a pipeline of light-weight specialized blocks. The M-S architecture utilizes a simple and general token-based bus control mechanism that is reconfiguration aware. All system modules have been described in synthesizable VHDL. A first system prototype has been built and validated using the affordable XUP XC2VP30 board. Even when using CRC check of bitstreams dynamic reconfiguration can proceed at the maximum speed that can be supported by the ICAP Xilinx interface. The reconfiguration support logic consumes as little as 1012 slices on the Virtex II Pro FPGA.

We have designed a modular SOM systolic architecture that can classify data vectors with thousands of elements in real time. The architecture is described as a soft IP core in synthesizable VHDL. The SOM neural network size, the input data vectors dimension, the weight and data element bitwidth precision etc. are all designer tunable parameters. Several SOM neural network instances have been synthesized and their performance evaluated for different Xilinx Virtex-II and Virtex II-Pro FPGAs. Moderate to large size SOM networks that can process data vectors with as many as 4096 elements can fit into a single FPGA device and clocked at frequencies as high as 150 MHz. This makes the architecture a useful co-processor candidate for the real-time categorization of large-size genomics and proteomics datasets.

2D gel electrophoresis (2DGE) is the most commonly used method for protein separation. After gel scanning, images with a plethora of spot features are generated. In this paper we propose the use of the contourlet transform (CT) for 2D gels image denoising and compare it to the wavelet transform (WT). We show that contourlets not only achieve better average PSNR performance, but also preserve better, relatively to wavelets, the spot boundaries and alter less the intensities of significant spot features. Proper denoising of 2DGE images is essential in order to extract reliable spot features in proteomics workflows for biomarkers discovery.

Many scientists and engineers have signal and image processing applications that involve large data sets and could benefit from parallel processing on readily-available clusters of workstations (COWs). Unfortunately these applications often exist as legacy code, such as Matlab functions, which are not easily parallelizable. The goal of the JavaPorts project is to provide a framework for flexible development of parallel and distributed component-based applications for heterogeneous COWs. The latest version of the package supports the integration of Java and Matlab components into the same application. The main features of the JavaPorts framework and its tools are discussed and application examples are presented which highlight its ability to support the rapid prototyping of parallel and distributed computing strategies with communicating Java and Matlab components.

Summary form only given. As clusters of workstations become increasingly popular the need for component frameworks that facilitate the high level modeling and rapid prototyping of parallel and distributed applications on such systems is becoming pressing. Many scientists and engineers have image and signal processing applications that could benefit from cluster computing. However, these applications often exist as legacy code, such as serial Matlab functions, which are not easily parallelizable. The goal of the JavaPorts project is to provide a framework and a set of tools that make it easy to develop component-based parallel and distributed applications for networks of heterogeneous computing nodes. The latest version of the package supports the integration of Java and Matlab components into the same application and provides a mechanism for incorporating legacy Matlab functions into parallel processing applications. The design and salient features of the framework and associated tools are discussed here, and application examples are presented which highlight how JavaPorts can be used to model, develop, launch and restructure applications with any number of interacting Java and Matlab components.

We present tools that support the run-time execution of applications that mix software running on networks of workstations and reconfigurable hardware. We use JHDL to describe the reconfigurable hardware, and JavaPorts to handle the communications between nodes in the network. The heterogeneous resources are handled by interposing a communication layer between the application and the hardware. The communication layer provides (i) the ability to modify the target hardware without modifying the application, (ii) co-design of the application and hardware, (iii) simulation of the entire system before the hardware design is complete, and (iv) remote execution so the application can reside on a different host from the hardware. We demonstrate the feasibility of this approach with a Java-based system which has a communication layer called the packet exchange platform (PEP). We present the system, describe the PEP and its implementation, and show how this approach has been applied to an image processing application.

One of the main advantages of the Java programming language is that the source code can be executed in any type of machine or embedded system that implements a Java Virtual Machine (JVM). The JavaPorts (JP)framework adds another level of portability by extending this capability to multi-threaded distributed applications consisting of many concurrent and interacting components. In this paper we primarily, discuss the aspects of the JavaPorts framework that pertain to the generation and reuse of components. We discuss how a distributed JP application is defined, using a Task Graph abstraction, and then captured and represented internally as an AMTP tree data structure. We present a suite of tools designed to help in simplifying the application development process. We explain how the JP graphical user interface can be used to assemble distributed applications by combining new and existing reusable software components.

In order to rapidly produce, using high level synthesis, quality silicon implementations of Signal Flow Graphs (SFGs) for large size, real-world signal/image processing problems, the Hardware Description Language (HDL) representations of SFG nodes should possess certain desirable characteristics. We have embedded in DG2VHDL, a design tool developed by the authors which translates automatically an algorithm's Dependence Graph into synthesizable VHDL models for SFG arrays, an algorithm that formulates the minimal design complexity nested loop structure (to be defined herein) for each SFG processor. This representation will, in all but some pathological cases, produce post-synthesis hardware whose area scales near optimally with increasing problem size. Furthermore, the time and memory required for the synthesis of such models does not increase with the problem size. A polynomial time heuristic is presented which finds (almost always) the minimal design complexity loop representation of SFG nodes.

We present a design case study using DG2VHDL a tool which bridges the gap between an abstract graphical description of a DSP algorithm and its concrete hardware description language (HDL) representation, DG2VHDL automatically translates a dependence graph (DG) into a synthesizable, behavioral VHDL entity that can be input to industrial strength behavioral compilers for producing silicon implementations of the algorithm (FPGAs, ASICs). Full search block matching motion estimation was selected for its current applications (MPEG, HDTV, video conferencing) as well as for the richness of literature and architectural exploration over the last decade. We will not only demonstrate here that the behavioral VHDL code produced automatically by the tool leads, after behavioral synthesis, to an efficient distributed memory and control modular array architecture, but will also provide comparative statistics for several new FS-BMA architectures derived for real-time motion estimation.

One of the long standing problems in passive stereo vision is that of constructing a range map using only two images providing two views of the same 3-D world scene. It amounts to identifying pairs of corresponding pixels, one pixel in each image, that are associated with the same point on the 3-D world. We are introducing ART-1 neural networks as a primitive capable of addressing the stereo correspondence problem. Using a multi-pass approach it is possible to increase gradually the density of matched points, while at the same time false matches are filtered by requiring close agreement between disparity estimates in a neighborhood. At the end, a reasonably dense disparity map is obtained, to the extent that it allows the reconstruction of the scene using interpolation methods. Our scheme has been tested on stereograms, virtual world scenes generated by computer programs (where the "ground truth" is known) as well as real-world scenes. In all cases the scene reconstructions are shown to be quite realistic.

Model-based joint segmentation and recognition of objects is proposed in the framework of parameter estimation of hierarchical mixture densities. The maximum a posteriori (MAP) estimate of the parameters is computed by the application of a modified version of the expectation-maximization algorithm (EM with regularizing constraints applied to multiple level hierarchies). The approach is flexible in the sense that it allows for non-stationary pixel statistics, different noise models and is translation and scale invariant. Simulation results suggest that the scheme is well suited for recognition of partially occluded objects and recognition in complex and poorly modeled background.

We propose an architecture for multiresolution classification based on a wavelet decomposition and a hierarchical neural network. Each layer of the neural network is a "frequency expert" associated with a single frequency channel. Higher layers in the hierarchy integrate the information provided by the layers below, leading to a network where different "experts" cooperate to explain the input data. We illustrate the performance of this architecture on edge detection, a problem which is well known to be best addressed in a multiresolution framework.

We investigate the application of Hopfield neural networks (HNN's) to the problem of multiuser detection in spread spectrum/CDMA (code division multiple access) communication systems. It is shown that the NP-complete problem of minimizing the objective function of the optimal multiuser detector (OMD) can be translated into minimizing an HNN "energy" function, thus allowing to take advantage of the ability of HNN's to perform very fast gradient descent algorithms in analog hardware and produce in real-time suboptimal solutions to hard combinatorial optimization problems. The performance of the proposed HNN receiver is evaluated via computer simulations and compared to that of other suboptimal schemes as well as to that of the OMD for both the synchronous and the asynchronous CDMA transmission cases. It is shown that the HNN detector exhibits a number of attractive properties and that it provides a powerful generalization of a well-known and extensively studied suboptimal scheme, namely the multistage detector.

This paper presents an advanced concurrent simulation technique for performing Multiple Stuck-at Fault Simulation based on Multiple Domain Concurrent and Comparative Simulation (MDCCS). It efficiently compresses multiple experiments in a single simulation and requires no pre-analysis of the circuit. In addition, MDCCS has a unique feature that allows experiments to interact with each other and spawn offspring experiments should new behaviors arise. MDCCS is based on discrete event concurrent simulation (CS) and gains efficiency by utilizing the similarity among experiments without resorting to parallel hardware. It provides a mechanism for managing the complexity of interactions such that the implementation of this methodology is both storage and CPU time efficient. The intention of this paper is to present the MDCCS framework and report on its effectiveness for digital logic fault simulation.

We discuss the parallel implementation of multidimensional FFTs on distributed memory multiprocessor machines. We introduce a compact notation to describe four equivalent parallel algorithms and discuss their advantages and disadvantages. Two algorithms, suitable for the case when initial and final data are distributed either row- or column-wise, the traditional row-column (RC) and a variation of the vector radix (VR) that we call partial vector radix (PVR) are presented and their efficiency on the Paragon is compared. It is shown that the PVR, although it requires larger amount of interprocessor communication, results in more efficient implementations due to the regularity of local and distributed memory accesses. For the case in which data are partitioned along both dimensions, two suitable parallel algorithms, the collect-distribute (CD) and the general full vector-radix (FVR), are presented. Again, it is shown that regularity in memory accesses for the case of the FVR, results in more efficient implementations.

The design of efficient parallel processing implementations for speeding up the computationally intensive estimation of higher order statistics (HOS) has been recognized as an important task by the signal processing community. We report on the synthesis of minimum running time (latency) data-parallel algorithms that can be employed to compute all moment lags, up to the 3rd or 4th-order, on the MasPar-1 single instruction multiple data (SIMD) parallel system. By construction the synthesized SIMD algorithms require constant memory per processing element (PE), thus allowing the processing of 1-D input data sequences with as many as M=2/sup 10/ data samples. Simulation results are presented showing the gain in speedup and execution times, as compared to optimized versions of the serial estimation algorithm running in powerful workstations.

Nonlinear adaptive filters based on a variety of neural network models have been used successfully for system identification and noise-cancellation in a wide class of applications. An important problem in data communications is that of channel equalization, i.e., the removal of interferences introduced by linear or nonlinear message corrupting mechanisms, so that the originally transmitted symbols can be recovered correctly at the receiver. In this paper we introduce an adaptive recurrent neural network (RNN) based equalizer whose small size and high performance makes it suitable for high-speed channel equalization. We propose RNN based structures for both trained adaptation and blind equalization, and we evaluate their performance via extensive simulations for a variety of signal modulations and communication channel models. It is shown that the RNN equalizers have comparable performance with traditional linear filter based equalizers when the channel interferences are relatively mild, and that they outperform them by several orders of magnitude when either the channel's transfer function has spectral nulls or severe nonlinear distortion is present. In addition, the small-size RNN equalizers, being essentially generalized IIR filters, are shown to outperform multilayer perceptron equalizers of larger computational complexity in linear and nonlinear channel equalization cases.

CORP, a procedure for recovery from transient faults in real-time sensitive medical applications, is introduced. CORP is more effective than the traditional approach because it relies on concurrent retries executed by two neighbor processors in the array (faulty and assistant), instead of successive retries executed only by the faulty processor. Techniques to analyze how the occurrence of transient/intermittent faults disturbs the execution of a parallel algorithm in linear arrays are discussed. An optimal assistant assignment policy is constructed that maximizes the array performance in the presence of faults. The adaptive implementation of the optimal policy in linear-wavefront arrays using local distributed control and near-neighbor communications is presented.>