Empirical evidence for ultrametric structure in multi-layer perceptron error surfaces

Combinatorial optimization problems share an interesting property with spin glass systems in that their state spaces can exhibit ultrametric structure. We use sampling methods to analyse the error surfaces of feedforward multi-layer perceptron neural networks learning encoder problems. The third order statistics of these points of attraction are examined and found to be arranged in a highly ultrametric way. This is a unique result for a finite, continuous parameter space. The implications of this result are discussed.

Micro-mechanical approximation to the stress field around an inclined fiber of FRCC

Matrix spalling or crushing is one of the important mechanisms of fiber-matrix interaction of fiber reinforced cementitious composites (FRCC). The fiber pullout mechanisms have been extensively studied for an aligned fiber but matrix failure is rarely investigated since it is thought not to be a major affect. However, for an inclined fiber, the matrix failure should not be neglected. Due to the complex process of matrix spalling, experimental investigation and analytical study of this mechanism are rarely found in literature. In this paper, it is assumed that the load transfer is concentrated within the short length of the inclined fiber from the exit point towards anchored end and follows the exponential law. The Mindlin formulation is employed to calculate the 3D stress field. The simulation gives much information about this field. The 3D approximation of the stress state around an inclined fiber helps to qualitatively understand the mechanism of matrix failure. Finally, a spalling criterion is proposed by which matrix spalling occurs only when the stress in a certain volume, rather than the stress at a small point, exceeds the material strength. This implies some local stress redistribution after first yield. The stress redistribution results in more energy input and higher load bearing capacity of the matrix. In accordance with this hypothesis, the evolution of matrix spalling is demonstrated. The accurate prediction of matrix spalling needs the careful determination of the parameters in this model. This is the work of further study. (C) 2002 Elsevier Science Ltd. All rights reserved.

Boundary-layer transition measurements in hypervelocity flows in a shock tunnel

Experiments to investigate the transition process in hypervelocity boundary layers were performed in the T4 free-piston shock tunnel. An array of thin-film heat-transfer gauges was used to detect the location and extent of the transitional region on a 1500 mm long x 120 turn wide flat plate, which formed one of the walls of a duct. The experiments were performed in a Mach 6 flow of air with 6- and 12-MJ/kg nozzle-supply enthalpies at unit Reynolds numbers ranging from 1.6 x 10(6) to 4.9 x 10(6) m(-1). The results show that the characteristics typical of transition taking place through the initiation, growth, and merger of turbulent spots are evident in the heat-transfer signals. A 2-mm-high excrescence located 440 turn from the leading edge was found to be capable of generating a turbulent wedge within an otherwise laminar boundary layer at a unit Reynolds number of 2.6 x 10(6) m(-1) at the 6-MJ/kg condition. A tripping strip, located 100 mm from the leading edge and consisting of a line 37 teeth of 2 rum height equally spaced and spanning the test surface, was also found to be capable of advancing the transition location at the same condition and at the higher enthalpy condition.

Antagonists of protein kinase C inhibit rat retinal glutamate transport activity in situ

Neuronal and glial high-affinity transporters regulate extracellular glutamate concentration, thereby terminating synaptic transmission and preventing neuronal excitotoxicity. Glutamate transporter activity has been shown to be modulated by protein kinase C (PKC) in cell culture. This is the first study to demonstrate such modulation in situ, by following the fate of the non-metabolisable glutamate transporter substrate, D-aspartate. In the rat retina, pan-isoform PKC inhibition with chelerythrine suppressed glutamate uptake by GLAST (glutamate/aspartate transporter), the dominant excitatory amino acid transporter localized to the glial Muller cells. This effect was mimicked by rottlerin but not by Go6976, suggesting the involvement of the PKCdelta isoform, but not PKCalpha, beta or gamma. Western blotting and immunohistochemical labeling revealed that the suppression of glutamate transport was not due to a change in transporter expression. Inhibition of PKCdelta selectively suppressed GLAST but not neuronal glutamate transporter activity. These data suggest that the targeting of specific glutamate transporters with isoform-specific modulators of PKC activity may have significant implications for the understanding of neurodegenerative conditions arising from compromised glutamate homeostasis, e.g. glaucoma and amyotrophic lateral sclerosis.

Distribution of two splice variants of the glutamate transporter GLT-1 in the developing rat retina

The distributions of a carboxyl terminal splice variant of the glutamate transporter GLT-1, referred to as GLT-1B, and the carboxyl terminus of the originally described variant of GLT-1, referred to hereafter as GLT-1alpha, were examined using specific antisera. GLT-1B was present in the retina at very early developmental stages. Labelling was demonstrable at embryonic day 14, and strong labelling was evident by embryonic day 18. Such labelling was initially restricted to populations of cone photoreceptors, the processes of which extended through the entire thickness of the retina and appeared to make contact with the retinal ganglion cells. During postnatal development the GLT-1B-positive photoreceptor processes retracted to form the outer plexiform layer, and around postnatal day 7, GLT-1B-immunoreactive bipolar cells appeared. The pattern of labelling of bipolar cell processes within the inner plexiform layer changed during postnatal development. Two strata of strongly immunoreactive terminals were initially evident in the inner plexiform layer, but by adulthood these two bands were no longer evident and labelling was restricted to the somata and processes (but not synaptic terminals) of the bipolar cells, as well as the somata, processes, and terminals of cone photoreceptors. By contrast, GLT-1alpha appeared late in postnatal development and was restricted mainly to a population of amacrine cells, although transient labelling was also associated with punctate elements in the outer plexiform layer, which may represent photoreceptor terminals, (C) 2002 Wiley-Liss, Inc.

Synaptic vesicle transport and synaptic membrane transporter sites in excitatory amino acid nerve terminals in Alzheimer disease

The apparent L-[H-3]glutamate uptake rate (v') was measured in synaptic vesicles isolated from cerebral cortex synaptosomes prepared from autopsied Alzheimer and non-Alzheimer dementia cases, and age-matched controls. The initial synaptosome preparations exhibited similar densities of D-[H-3]aspartate membrane binding sites (B-MAX values) in the three groups. In control brain the temporal cortex D-[H-3]aspartate B-MAX was 132% of that in motor cortex, parallel with the L- [H-3]glutamate v' values (temporal = 139% of motor; NS). Unlike D- [H-3]aspartate B-MAX values, L- [H-3]glutamate v' values were markedly and selectively lower in Alzheimer brain preparations than in controls, particularly in temporal cortex. The difference could not be attributed to differential effects of autopsy interval or age at death. Non-Alzheimer dementia cases resembled controls. The selective loss of vesicular glutamate transport is consistent with a dysfunction in the recycling of transmitter glutamate.

Haemodynamic response to TENS applied to upper thoracic nerve roots in normal subjects

This study investigated the haemodynamic response to the 90-minute application of 85 Hz transcutaneous electrical nerve stimulation (TENS) to the T1 and T5 nerve roots. Comparison was made between 20 healthy subjects who had TENS stimulation and a separate group of 20 healthy subjects who rested for 90 minutes. Pulse and blood pressure were measured just prior to the start of TENS stimulation, after 30 minutes of stimulation, and after 90 minutes of stimulation (immediately after stopping TENS) or at completion of the rest time depending on group allocation. The rate pressure product was calculated from the pulse and systolic blood pressure data. Multivariate repeated measures analysis showed a significant group effect for TENS (p = 0.048). Univariate repeated measures analyses showed a significant group by time effect due to TENS on systolic blood pressure over the 90-minute time period (p = 0.028). Separate group repeated measures ANOVA showed a significant decline in heart rate (p = 0.000), systolic blood pressure (p = 0.013) and rate pressure product (p = 0.000) for the TENS group, while the control resting group showed a significant decline in heart rate only (p = 0.04). The application of 85 Hz TENS to the upper thoracic nerve roots causes no adverse haemodynamic effects in healthy subjects.

Nerve growth factor overexpression and autocrine loop in breast cancer cells

We show here that nerve growth factor (NGF), the canonical neurotrophic factor, is synthesized and released by breast cancer cells. High levels of NGF transcript and protein were detected in breast cancer cells by reverse transcription-PCR, Western blotting, ELISA assay and immunohistochemistry. Conversely, NGF production could not be detected in normal breast epithelial cells at either the transcriptional or protein level. Confocal analysis indicated the presence of NGF within classical secretion vesicles. Breast cancer cell-produced NGF was biologically active, as demonstrated by its ability to induce the neuronal differentiation of embryonic neural precursor cells. Importantly, the constitutive growth of breast cancer cells was strongly inhibited by either NGF-neutralizing antibodies or K-252a, a pharmacological inhibitor of NGF receptor TrkA, indicating the existence of an NGF autocrine loop. Together, our data demonstrate the physiological relevance of NGF in breast cancer and its potential interest as a marker and therapeutic target.

Biaxial failure model for fiber reinforced concrete

Steel fiber reinforced concrete (SFRC) is widely applied in the construction industry. Numerical elastoplastic analysis of the macroscopic behavior is complex. This typically involves a piecewise linear failure curve including corner singularities. This paper presents a single smooth biaxial failure curve for SFRC based on a semianalytical approximation. Convexity of the proposed model is guaranteed so that numerical problems are avoided. The model has sufficient flexibility to closely match experimental results. The failure curve is also suitable for modeling plain concrete under biaxial loading. Since this model is capable of simulating the failure states in all stress regimes with a single envelope, the elastoplastic formulation is very concise and simple. The finite element implementation is developed to demonstrate the conciseness and the effectiveness of the model. The computed results display good agreement with published experimental data.

Actively shielded multi-layer gradient coil designs with improved cooling properties

In standard cylindrical gradient coils consisting of a single layer of wires, a limiting factor in achieving very large magnetic field gradients is the rapid increase in coil resistance with efficiency. This is a particular problem in small-bore scanners, such as those used for MR microscopy. By adopting a multi-layer design in which the coil wires are allowed to spread out into multiple layers wound at increasing radii, a more favourable scaling of resistance with efficiency is achieved, thus allowing the design of more powerful gradient coils with acceptable resistance values. Previously this approach has been applied to the design of unshielded, longitudinal, and transverse gradient coils. Here, the multi-layer approach has been extended to allow the design of actively shielded multi-layer gradient coils, and also to produce coils exhibiting enhanced cooling characteristics. An iterative approach to modelling the steady-state temperature distribution within the coil has also been developed. Results indicate that a good level of screening can be achieved in multi-layer coils, that small versions of such coils can yield higher efficiencies at fixed resistance than conventional two-layer (primary and screen) coils, and that performance improves as the number of layers of increases. Simulations show that by optimising multi-layer coils for cooling it is possible to achieve significantly higher gradient strengths at a fixed maximum operating temperature. A four-layer coil of 8 mm inner diameter has been constructed and used to test the steady-state temperature model. (C) 2003 Elsevier Inc. All rights reserved.

The Lake Tekapo experiment (LTEX): An investigation of atmospheric boundary layer processes in complex terrain

A research program on atmospheric boundary layer processes and local wind regimes in complex terrain was conducted in the vicinity of Lake Tekapo in the southern Alps of New Zealand, during two 1-month field campaigns in 1997 and 1999. The effects of the interaction of thermal and dynamic forcing were of specific interest, with a particular focus on the interaction of thermal forcing of differing scales. The rationale and objectives of the field and modeling program are described, along with the methodology used to achieve them. Specific research aims include improved knowledge of the role of surface forcing associated with varying energy balances across heterogeneous terrain, thermal influences on boundary layer and local wind development, and dynamic influences of the terrain through channeling effects. Data were collected using a network of surface meteorological and energy balance stations, radiosonde and pilot balloon soundings, tethered balloon and kite-based systems, sodar, and an instrumented light aircraft. These data are being used to investigate the energetics of surface heat fluxes, the effects of localized heating/cooling and advective processes on atmospheric boundary layer development, and dynamic channeling. A complementary program of numerical modeling includes application of the Regional Atmospheric Modeling System (RAMS) to case studies characterizing typical boundary layer structures and airflow patterns observed around Lake Tekapo. Some initial results derived from the special observation periods are used to illustrate progress made to date. In spite of the difficulties involved in obtaining good data and undertaking modeling experiments in such complex terrain, initial results show that surface thermal heterogeneity has a significant influence on local atmospheric structure and wind fields in the vicinity of the lake. This influence occurs particularly in the morning. However, dynamic channeling effects and the larger-scale thermal effect of the mountain region frequently override these more local features later in the day.

Identification of cholinoceptive glycinergic neurons in the mammalian retina

The light-evoked release of acetylcholine (ACh) affects the responses of many retinal ganglion cells, in part via nicotinic acetylcholine receptors (nAChRs). nAChRs that contain beta2alpha3 neuronal nicotinic acetylcholine receptors have been identified and localized in the rabbit retina; these nAChRs are recognized by the monoclonal antibody mAb210. We have examined the expression of beta2alpha3 nAChRs by glycinergic amacrine cells in the rabbit retina and have identified different subpopulations of nicotinic cholinoceptive glycinergic cells using double and triple immunohistochemistry with quantitative analysis. Here we demonstrate that about 70% of the cholinoceptive amacrine cells in rabbit retina are glycinergic cells. At least three nonoverlapping subpopulations of mAb210 glycine-immunoreactive cells can be distinguished with antibodies against calretinin, calbindin, and gamma-aminobutyric acid (GABA)(A) receptors. The cholinergic cells in rabbit retina are thought to synapse only on other cholinergic cells and ganglion cells. Thus, the expression of beta2alpha3 nAChRs on diverse populations of glycinergic cells is puzzling. To explore this finding, the subcellular localization of beta2alpha3 was studied at the electron microscopic level. mAb210 immunoreactivity was localized on the dendrites of amacrines and ganglion cells throughout the inner plexiform layer, and much of the labeling was not associated with recognizable synapses. Thus, our findings indicate that ACh in the mammalian retina may modulate glycinergic circuits via extrasynaptic beta2alpha3 nAChRs. (C) 2002 Wiley-Liss, Inc.

Phase equilibria and surface tension of pure fluids using a molecular layer structure theory (MLST) model

This paper presents a new model based on thermodynamic and molecular interaction between molecules to describe the vapour-liquid phase equilibria and surface tension of pure component. The model assumes that the bulk fluid can be characterised as set of parallel layers. Because of this molecular structure, we coin the model as the molecular layer structure theory (MLST). Each layer has two energetic components. One is the interaction energy of one molecule of that layer with all surrounding layers. The other component is the intra-layer Helmholtz free energy, which accounts for the internal energy and the entropy of that layer. The equilibrium between two separating phases is derived from the minimum of the grand potential, and the surface tension is calculated as the excess of the Helmholtz energy of the system. We test this model with a number of components, argon, krypton, ethane, n-butane, iso-butane, ethylene and sulphur hexafluoride, and the results are very satisfactory. (C) 2002 Elsevier Science B.V. All rights reserved.