Visualization of learning in multilayer perceptron networks using principal component analysis

This paper is concerned with the use of scientific visualization methods for the analysis of feedforward neural networks (NNs). Inevitably, the kinds of data associated with the design and implementation of neural networks are of very high dimensionality, presenting a major challenge for visualization. A method is described using the well-known statistical technique of principal component analysis (PCA). This is found to be an effective and useful method of visualizing the learning trajectories of many learning algorithms such as back-propagation and can also be used to provide insight into the learning process and the nature of the error surface.

Measuring the decoherence rate in a semiconductor charge qubit

We describe a method by which the decoherence time of a solid-state qubit may be measured. The qubit is coded in the orbital degree of freedom of a single electron bound to a pair of donor impurities in a semiconductor host. The qubit is manipulated by adiabatically varying an external electric field. We show that by measuring the total probability of a successful qubit rotation as a function of the control field parameters, the decoherence rate may be determined. We estimate various system parameters, including the decoherence rates due to electromagnetic fluctuations and acoustic phonons. We find that, for reasonable physical parameters, the experiment is possible with existing technology. In particular, the use of adiabatic control fields implies that the experiment can be performed with control electronics with a time resolution of tens of nanoseconds.

Polarization vision and its role in biological signaling

Visual pigments, the molecules in photoreceptors that initiate the process of vision, are inherently dichroic, differentially absorbing light according to its axis of polarization. Many animals have taken advantage of this property to build receptor systems capable of analyzing the polarization of incoming light, as polarized light is abundant in natural scenes (commonly being produced by scattering or reflection). Such polarization sensitivity has long been associated with behavioral tasks like orientation or navigation. However, only recently have we become aware that it can be incorporated into a high-level visual perception akin to color vision, permitting segmentation of a viewed scene into regions that differ in their polarization. By analogy to color vision, we call this capacity polarization vision. It is apparently used for tasks like those that color vision specializes in: contrast enhancement, camouflage breaking, object recognition, and signal detection and discrimination. While color is very useful in terrestrial or shallow-water environments, it is an unreliable cue deeper in water due to the spectral modification of light as it travels through water of various depths or of varying optical quality. Here, polarization vision has special utility and consequently has evolved in numerous marine species, as well as at least one terrestrial animal. In this review, we consider recent findings concerning polarization vision and its significance in biological signaling.

Magnetic Investigation of CoFe(2)O(4) Nanoparticles Supported in Biocompatible Polymeric Microsphere

Magnetic investigation of spinel ferrite nanoparticles dispersed in biocompatible polymeric microspheres is reported in this study. X-ray diffraction data analysis confirms the presence of nanosized CoFe(2)O(4) particles (mean size of similar to 8 nm). This finding is corroborated by transmission electron microscopy micrographs. Magnetization isotherms suggest a spin disorder likely occurring at the nanoparticle`s surface. The saturation magnetization value is used to estimate particle concentration of 1.6 x 10(18) cm(-3) dispersed in the polymeric template. A T(1/2) dependence of the coercive field is determined in the low-temperature region (T < 30 K). The model of non-interacting mono-domains is used to estimate an effective magnetic anisotropy of K(eff) = 0.6 x 10(5) J/m(3). The K(eff) value we found is lower than the value reported for spherically-shaped CoFe(2)O(4) nanoparticles, though consistent with the low coercive field observed in the investigated sample.

Pressure-Frequency Sensing Subxiphoid Access System for Use in Percutaneous Cardiac Electrophysiology: Prototype Design and Pilot Study Results

We have designed, built, and tested an early prototype of a novel subxiphoid access system intended to facilitate epicardial electrophysiology, but with possible applications elsewhere in the body. The present version of the system consists of a commercially available insertion needle, a miniature pressure sensor and interconnect tubing, read-out electronics to monitor the pressures measured during the access procedure, and a host computer with user-interface software. The nominal resolution of the system is <0.1 mmHg, and it has deviations from linearity of <1%. During a pilot series of human clinical studies with this system, as well as in an auxiliary study done with an independent method, we observed that the pericardial space contained pressure-frequency components related to both the heart rate and respiratory rate, while the thorax contained components related only to the respiratory rate, a previously unobserved finding that could facilitate access to the pericardial space. We present and discuss the design principles, details of construction, and performance characteristics of this system.

Chemically and morphologically identifiable glomeruli in the rat olfactory bulb

Primary olfactory neurons that express the same odorant receptor are distributed mosaically throughout the olfactory neuroepithelium lining the nasal cavity, yet their axons converge and form discrete glomeruli in the olfactory bulb. We previously proposed that cell surface carbohydrates mediate the sorting out and selective fasciculation of primary olfactory axons en route to glomeruli. If this were the case, then axons that terminate in the same glomerulus would express the same complement of cell surface carbohydrates. In this study, we examined the expression of a novel carbohydrate (NOC-3) on neural cell adhesion molecule in the adult rat olfactory system. NOC-3 was expressed by a subset of neurons distributed throughout the olfactory neuroepithelium. The axons of these neurons entered the nerve fiber layer and terminated in a subset of glomeruli. It is interesting to note that we identified three unusually large glomeruli in the lateral, ventrolateral, and ventromedial olfactory bulb that were innervated by axons expressing NOC-3. NOC-3-expressing axons sorted out and fasciculated into discrete fascicles prior to entering these glomeruli. Each of these glomeruli was in a topographically fixed position in the olfactory bulbs of the same animal as well as in different animals, and their lengths were approximately 10% of the total length of the bulb. They could be identified reliably by both their topographical position and their unique morphology. These results reveal that axons expressing the same cell surface carbohydrates consistently target the same topographically fixed glomeruli, which supports a role for these molecules in axon navigation in the primary olfactory nerve pathway. J. Comp. Neurol. 436: 497-507, 2001. (C) 2001 Wiley-Liss, Inc.

On the fast approximation of Green's functions in MPIE formulations for planar layered media

The numerical implementation of the complex image approach for the Green's function of a mixed-potential integralequation formulation is examined and is found to be limited to low values of k(0) rho (in this context k(0) rho = 2 pirho/ lambda(0), where rho is the distance between the source and the field points of the Green's function and lambda(0) is the free space wavelength). This is a clear limitation for problems of large dimension or high frequency where this limit is easily exceeded. This paper examines the various strategies and proposes a hybrid method whereby most of the above problems can be avoided. An efficient integral method that is valid for large k(0) rho is combined with the complex image method in order to take advantage of the relative merits of both schemes. It is found that a wide overlapping region exists between the two techniques allowing a very efficient and consistent approach for accurately calculating the Green's functions. In this paper, the method developed for the computation of the Green's function is used for planar structures containing both lossless and lossy media.

Analysis of a circular patch antenna radiating in a parallel-plate radial guide

A field matching method is described to analyze a recessed circular cavity radiating into a radial waveguide. Using the wall impedance approach, the analysis is divided into two separate problems of the cavity and its external environment. Based on this analysis, a computer algorithm is developed for determining wall admittances as seen at the edge of the patch in the cavity, the radial admittance matrix for the two-probe feed arrangement, and the input impedance as observed from the coaxial line feeding the cavity. This algorithm is tested against the general-purpose Hewlett-Packard finite-element High Frequency Structure Simulator as well as against measured results. Good agreement in all considered cases is noted.

Development of axon pathways in the zebrafish central nervous system

The zebrafish has a number of distinct advantages as an experimental model in developmental biology. For example, large numbers of embryos can be generated in each lay, development proceeds rapidly through a very precise temporal staging which exhibits minimal batch-to-batch variability, embryos are transparent and imaging of wholemounts negates the need for tedious histological preparation while preserving three-dimensional spatial relationships. The zebrafish nervous system is proving a convenient model for studies of axon guidance because of its small size and highly stereotypical trajectory of axons. Moreover, a simple scaffold of axon tracts and nerves is established early and provides a template for subsequent development. The ease with which this template can be visualized as well as the ability to spatially resolve individual pioneer axons enables the role of specific cell-cell and molecular interactions to be clearly deciphered. We describe here the morphology and development of the earliest axon pathways in the embryonic zebrafish central nervous system and highlight the major questions that remain to be addressed with regard to axon guidance.

Expression and role of roundabout-1 in embryonic Xenopus forebrain

The receptor Roundabout-1 (Robo1) and its ligand Slit are known to influence axon guidance and central nervous system (CNS) patterning in both vertebrate and nonvertebrate systems. Although Robo-Slit interactions mediate axon guidance in the Drosophila CNS, their role in establishing the early axon scaffold in the embryonic vertebrate brain remains unclear. We report here the identification and expression of a Xenopus Robo1 orthologue that is highly homologous to mammalian Robo1. By using overexpression studies and immunohistochemical and in situ hybridization techniques, we have investigated the role of Robo1 in the development of a subset of neurons and axon tracts in the Xenopus forebrain. Robo1 is expressed in forebrain nuclei and in neuroepithelial cells underlying the main axon tracts. Misexpression of Robo1 led to aberrant development of axon tracts as well as the ectopic differentiation of forebrain neurons. These results implicate Robo1 in both neuronal differentiation and axon guidance in embryonic vertebrate forebrain. (C) 2002 Wiley-Liss, Inc.

Review of time-domain polarization measurements for assessing insulation condition in aged transformers

This paper presents a review of the time-domain polarization measurement techniques for the condition assessment of aged transformer insulation. The polarization process is first described with appropriate dielectric response theories and then commonly used polarization methods are described with special emphasis on the most widely used return voltage(rv) measurement. Most recent emphasis has been directed to techniques of determining moisture content of insulation indirectly by measuring rv parameters. The major difficulty still lies with the accurate interpretation of return voltage results. This paper investigates different thoughts regarding the interpretation of rv results for different moisture and ageing conditions. Other time domain polarization measurement techniques and their results are also presented in this paper.

On the induced electric field gradients in the human body for magnetic stimulation by gradient coils in MRI

Prior theoretical studies indicate that the negative spatial derivative of the electric field induced by magnetic stimulation may he one of the main factors contributing to depolarization of the nerve fiber. This paper studies this parameter for peripheral nerve stimulation (PNS) induced by time.-varying gradient fields during MRI scans. The numerical calculations are based on an efficient, quasi-static, finite-difference scheme and an anatomically realistic human, full-body model. Whole-body cylindrical and planar gradient sets in MRI systems and various input signals have been explored. The spatial distributions of the induced electric field and their gradients are calculated and attempts are made to correlate these areas with reported experimental stimulation data. The induced electrical field pattern is similar for both the planar coils and cylindrical coils. This study provides some insight into the spatial characteristics of the induced field gradients for PNS in MRI, which may be used to further evaluate the sites where magnetic stimulation is likely to occur and to optimize gradient coil design.

An optical flow-based sensing system for reactive mobile robot navigation

This work discusses the use of optical flow to generate the sensorial information a mobile robot needs to react to the presence of obstacles when navigating in a non-structured environment. A sensing system based on optical flow and time-to-collision calculation is here proposed and experimented, which accomplishes two important paradigms. The first one is that all computations are performed onboard the robot, in spite of the limited computational capability available. The second one is that the algorithms for optical flow and time-to-collision calculations are fast enough to give the mobile robot the capability of reacting to any environmental change in real-time. Results of real experiments in which the sensing system here proposed is used as the only source of sensorial data to guide a mobile robot to avoid obstacles while wandering around are presented, and the analysis of such results allows validating the proposed sensing system.