On simple adaptive momentum

Simple Adaptive Momentum [1] was introduced as a simple means of speeding the training of multi-layer perceptrons (MLPs) by changing the momentum term depending on the angle between the current and previous changes in the weights of the MLP. In the original paper. the weight changes of the whole network are used in determining this angle. This paper considers adapting the momentum term using certain subsets of these weights. This idea was inspired by the author's object oriented approach to programming MLPs. successfully used in teaching students: this approach is also described. It is concluded that the angle is best determined using the weight changes in each layer separately.

Neural and wavelet network models for financial distress classification

This work analyzes the use of linear discriminant models, multi-layer perceptron neural networks and wavelet networks for corporate financial distress prediction. Although simple and easy to interpret, linear models require statistical assumptions that may be unrealistic. Neural networks are able to discriminate patterns that are not linearly separable, but the large number of parameters involved in a neural model often causes generalization problems. Wavelet networks are classification models that implement nonlinear discriminant surfaces as the superposition of dilated and translated versions of a single "mother wavelet" function. In this paper, an algorithm is proposed to select dilation and translation parameters that yield a wavelet network classifier with good parsimony characteristics. The models are compared in a case study involving failed and continuing British firms in the period 1997-2000. Problems associated with over-parameterized neural networks are illustrated and the Optimal Brain Damage pruning technique is employed to obtain a parsimonious neural model. The results, supported by a re-sampling study, show that both neural and wavelet networks may be a valid alternative to classical linear discriminant models.

Comparison of extrasystolic ECG signal classifiers using discrete wavelet transforms

This work compares and contrasts results of classifying time-domain ECG signals with pathological conditions taken from the MITBIH arrhythmia database. Linear discriminant analysis and a multi-layer perceptron were used as classifiers. The neural network was trained by two different methods, namely back-propagation and a genetic algorithm. Converting the time-domain signal into the wavelet domain reduced the dimensionality of the problem at least 10-fold. This was achieved using wavelets from the db6 family as well as using adaptive wavelets generated using two different strategies. The wavelet transforms used in this study were limited to two decomposition levels. A neural network with evolved weights proved to be the best classifier with a maximum of 99.6% accuracy when optimised wavelet-transform ECG data wits presented to its input and 95.9% accuracy when the signals presented to its input were decomposed using db6 wavelets. The linear discriminant analysis achieved a maximum classification accuracy of 95.7% when presented with optimised and 95.5% with db6 wavelet coefficients. It is shown that the much simpler signal representation of a few wavelet coefficients obtained through an optimised discrete wavelet transform facilitates the classification of non-stationary time-variant signals task considerably. In addition, the results indicate that wavelet optimisation may improve the classification ability of a neural network. (c) 2005 Elsevier B.V. All rights reserved.

Reversible ageing effects in cryogenically cooled infrared filter radiometers

In this paper we report the observation of drifts in the responsivity of cryogenically cooled InSb detector-based infrared filter radiometers which have very strong wavelength dependence. These drifts can result in the increase or decrease of the response of the filter radiometers by over 5%. The origin of these variations was investigated and was shown to arise due to a thin film of ice formed on the multi-layer bandpass filter used to define the spectral response of the filter radiometer. The thin layer of ice interacts with the characteristics of the filter (which itself consists of a number of thin layers) and modifies the filter spectral transmission thus modifying the response of the filter radiometer of which the filter is part of. These observations are particularly relevant to space instruments which use infrared filter radiometers for earth observation. Debris from the spacecraft engines is known to accumulate on cold surfaces of instruments carried on board. The deposition of this debris on cold filters can modify the spectral response of the instruments, which use these filters to define a spectral response. Crown Copyright (c) 2004 Published by Elsevier B.V. All rights reserved.

Improvements in or relating to the manufacture of multilayer interference filters

This invention relates to the manufacture of multi-layer interference filters for use with infra-red radiation, especially at wavelengths beyond 3.8 microns. A method of manufacturing a multi-layer interference filter comprising the steps of forming on a substrate successive layers of lead telluride and another material in alternation, under conditions in which sufficient oxygen is included in the lead telluride layers to reduce the apparent free charge carrier concentration therein, so that the resulting filter exhibits enhanced transparency to radiation of wavelengths greater than 3.8 microns and enhanced natural absorption to radiation of wavelength less than 3.8 microns.

Improvements relating to the manufacture of coated substrates

This invention relates to the manufacture of coated substrates, and particularly, but not exclusively, to the deposition of multi-layer coatings in the manufacture of interference filters consisting of multiple thin films. An object of the invention is to allow accurate control of the deposition of a succession of layers having good uniformity, for example during the manufacture by vacuum evaporation of multilayer interference filters for use with infrared radiation of particularly long wavelength, using a method which is self calibrating and which avoids the repetitive use of individual control layers.

Neural networks for linear control: an analysis

This paper considers the use of radial basis function and multi-layer perceptron networks for linear or linearizable, adaptive feedback control schemes in a discrete-time environment. A close look is taken at the model structure selected and the extent of the resulting parameterization. A comparison is made with standard, nonneural network algorithms, e.g. self-tuning control.

A critique of neural networks for discrete-time linear control

This paper discusses the use of multi-layer perceptron networks for linear or linearizable, adaptive feedback.control schemes in a discrete-time environment. A close look is taken at the model structure selected and the extent of the resulting parametrization. A comparison is made with standard, non-perceptron algorithms, e.g. self-tuning control, and it is shown how gross over-parametrization can occur in the neural network case. Because of the resultant heavy computational burden and poor controller convergence, a strong case is made against the use of neural networks for discrete-time linear control.

Optimising memory usage in n-tuple neural networks

The use of n-tuple or weightless neural networks as pattern recognition devices is well known (Aleksander and Stonham, 1979). They have some significant advantages over the more common and biologically plausible networks, such as multi-layer perceptrons; for example, n-tuple networks have been used for a variety of tasks, the most popular being real-time pattern recognition, and they can be implemented easily in hardware as they use standard random access memories. In operation, a series of images of an object are shown to the network, each being processed suitably and effectively stored in a memory called a discriminator. Then, when another image is shown to the system, it is processed in a similar manner and the system reports whether it recognises the image; is the image sufficiently similar to one already taught? If the system is to be able to recognise and discriminate between m-objects, then it must contain m-discriminators. This can require a great deal of memory. This paper describes various ways in which memory requirements can be reduced, including a novel method for multiple discriminator n-tuple networks used for pattern recognition. By using this method, the memory normally required to handle m-objects can be used to recognise and discriminate between 2^m — 2 objects.

The control of dynamical systems by neural networks

In this paper the use of neural networks for the control of dynamical systems is considered. Both identification and feedback control aspects are discussed as well as the types of system for which neural networks can provide a useful technique. Multi-layer Perceptron and Radial Basis function neural network types are looked at, with an emphasis on the latter. It is shown how basis function centre selection is a critical part of the implementation process and that multivariate clustering algorithms can be an extremely useful tool for finding centres.

Cloud statistics and cloud radiative effect for a low-mountain site

In 2007, the Atmospheric Radiation Measurement (ARM) Mobile Facility (AMF) was operated for a nine-month period in the Murg Valley, Black Forest, Germany, in support of the Convective and Orographically-induced Precipitation Study (COPS). The synergy of AMF and COPS partner instrumentation was exploited to derive a set of high-quality thermodynamic and cloud property profiles with 30 s resolution. In total, clouds were present 72% of the time, with multi-layer mixed phase (28.4%) and single-layer water clouds (11.3%) occurring most frequently. A comparison with the Cloudnet sites Chilbolton and Lindenberg for the same time period revealed that the Murg Valley exhibits lower liquid water paths (LWPs; median = 37.5 g m−2) compared to the two sites located in flat terrain. In order to evaluate the derived thermodynamic and cloud property profiles, a radiative closure study was performed with independent surface radiation measurements. In clear sky, average differences between calculated and observed surface fluxes are less than 2% and 4% for the short wave and long wave part, respectively. In cloudy situations, differences between simulated and observed fluxes, particularly in the short wave part, are much larger, but most of these can be related to broken cloud situations. The daytime cloud radiative effect (CRE), i.e. the difference of cloudy and clear-sky net fluxes, has been analysed for the whole nine-month period. For overcast, single-layer water clouds, sensitivity studies revealed that the CRE uncertainty is likewise determined by uncertainties in liquid water content and effective radius. For low LWP clouds, CRE uncertainty is dominated by LWP uncertainty; therefore refined retrievals, such as using infrared and/or higher microwave frequencies, are needed.

Global retrieval of ATSR cloud parameters and evaluation (GRAPE): dataset assessment

The Along-Track Scanning Radiometers (ATSRs) provide a long time-series of measurements suitable for the retrieval of cloud properties. This work evaluates the freely-available Global Retrieval of ATSR Cloud Parameters and Evaluation (GRAPE) dataset (version 3) created from the ATSR-2 (1995�2003) and Advanced ATSR (AATSR; 2002 onwards) records. Users are recommended to consider only retrievals flagged as high-quality, where there is a good consistency between the measurements and the retrieved state (corresponding to about 60% of converged retrievals over sea, and more than 80% over land). Cloud properties are found to be generally free of any significant spurious trends relating to satellite zenith angle. Estimates of the random error on retrieved cloud properties are suggested to be generally appropriate for optically-thick clouds, and up to a factor of two too small for optically-thin cases. The correspondence between ATSR-2 and AATSR cloud properties is high, but a relative calibration difference between the sensors of order 5�10% at 660 nm and 870 nm limits the potential of the current version of the dataset for trend analysis. As ATSR-2 is thought to have the better absolute calibration, the discussion focusses on this portion of the record. Cloud-top heights from GRAPE compare well to ground-based data at four sites, particularly for shallow clouds. Clouds forming in boundary-layer inversions are typically around 1 km too high in GRAPE due to poorly-resolved inversions in the modelled temperature profiles used. Global cloud fields are compared to satellite products derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) measurements, and a climatology of liquid water content derived from satellite microwave radiometers. In all cases the main reasons for differences are linked to differing sensitivity to, and treatment of, multi-layer cloud systems. The correlation coefficient between GRAPE and the two MODIS products considered is generally high (greater than 0.7 for most cloud properties), except for liquid and ice cloud effective radius, which also show biases between the datasets. For liquid clouds, part of the difference is linked to choice of wavelengths used in the retrieval. Total cloud cover is slightly lower in GRAPE (0.64) than the CALIOP dataset (0.66). GRAPE underestimates liquid cloud water path relative to microwave radiometers by up to 100 g m�2 near the Equator and overestimates by around 50 g m�2 in the storm tracks. Finally, potential future improvements to the algorithm are outlined.

The interaction of flexural-gravity waves with periodic geometries

A periodic structure of finite extent is embedded within an otherwise uniform two-dimensional system consisting of finite-depth fluid covered by a thin elastic plate. An incident harmonic flexural-gravity wave is scattered by the structure. By using an approximation to the corresponding linearised boundary value problem that is based on a slowly varying structure in conjunction with a transfer matrix formulation, a method is developed that generates the whole solution from that for just one cycle of the structure, providing both computational savings and insight into the scattering process. Numerical results show that variations in the plate produce strong resonances about the ‘Bragg frequencies’ for relatively few periods. We find that certain geometrical variations in the plate generate these resonances above the Bragg value, whereas other geometries produce the resonance below the Bragg value. The familiar resonances due to periodic bed undulations tend to be damped by the plate.

Improvement of the GenTHREADER method for genomic fold recognition

Motivation: In order to enhance genome annotation, the fully automatic fold recognition method GenTHREADER has been improved and benchmarked. The previous version of GenTHREADER consisted of a simple neural network which was trained to combine sequence alignment score, length information and energy potentials derived from threading into a single score representing the relationship between two proteins, as designated by CATH. The improved version incorporates PSI-BLAST searches, which have been jumpstarted with structural alignment profiles from FSSP, and now also makes use of PSIPRED predicted secondary structure and bi-directional scoring in order to calculate the final alignment score. Pairwise potentials and solvation potentials are calculated from the given sequence alignment which are then used as inputs to a multi-layer, feed-forward neural network, along with the alignment score, alignment length and sequence length. The neural network has also been expanded to accommodate the secondary structure element alignment (SSEA) score as an extra input and it is now trained to learn the FSSP Z-score as a measurement of similarity between two proteins. Results: The improvements made to GenTHREADER increase the number of remote homologues that can be detected with a low error rate, implying higher reliability of score, whilst also increasing the quality of the models produced. We find that up to five times as many true positives can be detected with low error rate per query. Total MaxSub score is doubled at low false positive rates using the improved method.

Chitosan coated alginate beads for the survival of microencapsulated Lactobacillus plantarum in pomegranate juice

This work studied the effect of multi-layer coating of alginate beads on the survival of encapsulated Lactobacillus plantarum in simulated gastric solution and during storage in pomegranate juice at 4 °C. Uncoated, single and double chitosan coated beads were examined. The survival of the cells in simulated gastric solution (pH 1.5) was improved in the case of the chitosan coated beads by 0.5–2 logs compared to the uncoated beads. The cell concentration in pomegranate juice after six weeks of storage was higher than 5.5 log CFU/mL for single and double coated beads, whereas for free cells and uncoated beads the cells died after 4 weeks of storage. In simulated gastric solution, the size of the beads decreased and their hardness increased with time; however, the opposite trend was observed for pomegranate juice, indicating that there is no correlation between cell survival and the hardness of the beads.