Pruning artificial neural networks using neural complexity measures

This paper describes a new method for pruning artificial neural networks, using a measure of the neural complexity of the neural network. This measure is used to determine the connections that should be pruned. The measure computes the information-theoretic complexity of a neural network, which is similar to, yet different from previous research on pruning. The method proposed here shows how overly large and complex networks can be reduced in size, whilst retaining learnt behaviour and fitness. The technique proposed here helps to discover a network topology that matches the complexity of the problem it is meant to solve. This novel pruning technique is tested in a robot control domain, simulating a racecar. It is shown, that the proposed pruning method is a significant improvement over the most commonly used pruning method Magnitude Based Pruning. Furthermore, some of the pruned networks prove to be faster learners than the benchmark network that they originate from. This means that this pruning method can also help to unleash hidden potential in a network, because the learning time decreases substantially for a pruned a network, due to the reduction of dimensionality of the network.     

String matching artificial neural networks.

Three artificial neural networks (ANNs) are proposed for solving a variety of on- and off-line string matching problems. The ANN structure employed as the building block of these ANNs is derived from the harmony theory (HT) ANN, whereby the resulting string matching ANNs are characterized by fast match-mismatch decisions, low computational complexity, and activation values of the ANN output nodes that can be used as indicators of substitution, insertion (addition) and deletion spelling errors.     

Bioprocess model building using artificial neural networks

Artificial neural networks are made upon of highly interconnected layers of simple neuron-like nodes. The neurons act as non-linear processing elements within the network. An attractive property of artificial neural networks is that given the appropriate network topology, they are capable of learning and characterising non-linear functional relationships. Furthermore, the structure of the resulting neural network based process model may be considered generic, in the sense that little prior process knowledge is required in its determination. The methodology therefore provides a cost efficient and reliable process modelling technique. One area where such a technique could be useful is biotechnological systems. Here, for example, the use of a process model within an estimation scheme has long been considered an effective means of overcoming inherent on-line measurement problems. However, the development of an accurate process model is extremely time consuming and often results in a model of limited applicability. Artificial neural networks could therefore prove to be a useful model building tool when striving to improve bioprocess operability. Two large scale industrial fermentation systems have been considered as test cases; a fed-batch penicillin fermentation and a continuous mycelial fermentation. Both systems serve to demonstrate the utility, flexibility and potential of the artificial neural network approach to process modelling.     

Modeling of pain using artificial neural networks

In dealing with human nervous system, the sensation of pain is as sophisticated as other physiological phenomena. To obtain an acceptable model of the pain, physiology of the pain has been analysed in the present paper. Pain mechanisms are explained in block diagram representation form. Because of the nonlinear interactions existing among different sections in the diagram, artificial neural networks (ANNs) have been exploited. The basic patterns associated with chronic and acute pain have been collected and then used to obtain proper features for training the neural networks. Both static and dynamic representations of the ANNs were used in this regard. The trained networks then were employed to predict response of the body when it is exposed to special excitations. These excitations have not been used in the training phase and their behavior is interesting from the physiological view. Some of these predictions can be inferred from clinical experimentations. However, more clinical tests have to be accomplished for some of the predictions.     

Clustering proteins into families using artificial neural networks

An incorrect version of Figure 3 was published in the abovearticle, the corrected version is reproduced below.     

Reverberations and control of neural networks

Summary The simulation of neural networks, such as the brain cortex, which have a diffuse and rather uniform structure quite unlike the simple block-structure of extant computers, leads naturally to the study of functions and principles which only in part fall within the scope of Automata Theory. Systems of decision equations must be studied with a view especially to obtaining practical means for the prevision and computation of diffuse reverberations of wanted general characteristics, with the exclusion of all others. This amounts to deriving constraints on the allowed variability of the couplings among elements during learning processes, failing which the behavior of the simulator would become uncontrollable for practical purposes. A simple mathematical treatment is presented, which essentially linearizes these problems by an appropriate use of matrix algebra and permits a straightforward study of the wanted conditions, as well as of the controlling elements which may have to be added to the network.This work has been performed in part at the Laboratoire de Physique Théorique et Hautes Energies, Faculté des Sciences de Paris.This work has been performed with the joint sponsorship of the U.S.A.F. and their European Office of Aerospace Research under contracts no. AF EOAR 66-25 and AF 33(615)-2786.We wish to express our sincere thanks to Dr. F. Lauria for many illuminated discussions; and to Prof. M. Lévy for his kind hospitality at the Laboratoire de physique Théorique, in Paris, where part of this research was made.     

Thrips (Thysanoptera) identification using artificial neural networks

We studied the use of a supervised artificial neural network (ANN) model for semi-automated identification of 18 common European species of Thysanoptera from four genera: Aeolothrips Haliday (Aeolothripidae), Chirothrips Haliday, Dendrothrips Uzel, and Limothrips Haliday (all Thripidae). As input data, we entered 17 continuous morphometric and two qualitative two-state characters measured or determined on different parts of the thrips body (head, pronotum, forewing and ovipositor) and the sex. Our experimental data set included 498 thrips specimens. A relatively simple ANN architecture (multilayer perceptrons with a single hidden layer) enabled a 97% correct simultaneous identification of both males and females of all the 18 species in an independent test. This high reliability of classification is promising for a wider application of ANN in the practice of Thysanoptera identification.     

Application of artificial neural networks for yeast cells classification

The present paper describes a method for automatic classification of yeast cells in four groups: active with oval form, budding, weakened and dead. This method can be used in the previously developed structural mathematical model of the yeast cultivation process described in [1].     

Prediction of problematic wine fermentations using artificial neural networks

Artificial neural networks (ANNs) have been used for the recognition of non-linear patterns, a characteristic of bioprocesses like wine production. In this work, ANNs were tested to predict problems of wine fermentation. A database of about 20,000 data from industrial fermentations of Cabernet Sauvignon and 33 variables was used. Two different ways of inputting data into the model were studied, by points and by fermentation. Additionally, different sub-cases were studied by varying the predictor variables (total sugar, alcohol, glycerol, density, organic acids and nitrogen compounds) and the time of fermentation (72, 96 and 256 h). The input of data by fermentations gave better results than the input of data by points. In fact, it was possible to predict 100% of normal and problematic fermentations using three predictor variables: sugars, density and alcohol at 72 h (3 days). Overall, ANNs were capable of obtaining 80% of prediction using only one predictor variable at 72 h; however, it is recommended to add more fermentations to confirm this promising result.     

Modeling of phenol degradation system using artificial neural networks

Pseudomonas pictorum (NICM-2077) an effective strain used in the biodegradation of phenol was grown on various nutrient compounds which protect the microbes while confronting shock loads of concentrated toxic pollutants during waste water treatment. In the present study the effect of glucose, yeast extract, (NH₄)₂SO₄ and NaCl on phenol degradation has been investigated and a Artificial Neural Network (ANN) Model has been developed to predict degradation. Also the learning, recall and generalization characteristics of neural networks has been studied using phenol degradation system data. The network model was then compared with a Multiple Regression Analysis model (MRA) arrived from the same training data. Further, these two models were used to predict the percentage degradation of phenol for a blind test data. Though both the models perform equally well ANN is found to be better than MRA due to its slightly higher coefficient of correlation, lower RMS error value and lower average absolute error value during prediction.     

Prediction of N-linked glycan branching patterns using artificial neural networks

A model was developed for novel prediction of N-linked glycan branching pattern classification for CHO-derived N-linked glycoproteins. The model consists of 30 independent recurrent neural networks and uses predicted quantities of secondary structure elements and residue solvent accessibility as an input vector. The model was designed to predict the major component of a heterogeneous mixture of CHO-derived glycoforms of a recombinant protein under normal growth conditions. Resulting glycosylation prediction is classified as either complex-type or high mannose. The incorporation of predicted quantities in the input vector allowed for theoretical mutant N-linked glycan branching predictions without initial experimental analysis of protein structures. Primary amino acid sequence data were effectively eliminated from the input vector space based on neural network prediction analyses. This provided further evidence that localized protein secondary structure elements and conformational structure may play more important roles in determining glycan branching patterns than does the primary sequence of a polypeptide. A confidence interval parameter was incorporated into the model to enable identification of false predictions. The model was further tested using published experimental results for mutants of the tissue-type plasminogen activator protein [J. Wilhelm, S.G. Lee, N.K. Kalyan, S.M. Cheng, F. Wiener, W. Pierzchala, P.P. Hung, Alterations in the domain structure of tissue-type plasminogen activator change the nature of asparagine glycosylation. Biotechnology (N.Y.) 8 (1990) 321-325].     

Sleep classification in infants based on artificial neural networks.

The study reports on the possibility of classifying sleep stages in infants using an artificial neural network. The polygraphic data from 4 babies aged 6 weeks, 6 months and 1 year recorded over 8 hours were available for classification. From each baby 22 signals were recorded, digitized and stored on an optical disc. Subsets of these signals and additional calculated parameters were used to obtain data vectors, each of which represents an interval of 30 sec. For classification, two types of neural networks were used, a Multilayer Perceptron and a Learning Vector Quantizer. The teaching input for both networks was provided by a human expert. For the 6 sleep classes in babies aged 6 months, a 65% to 80% rate of correct classification (4 babies) was obtained for the testing data not previously seen.     

Eye tracking using artificial neural networks for human computer interaction

This paper describes an ongoing project that has the aim to develop a low cost application to replace a computer mouse for people with physical impairment. The application is based on an eye tracking algorithm and assumes that the camera and the head position are fixed. Color tracking and template matching methods are used for pupil detection. Calibration is provided by neural networks as well as by parametric interpolation methods. Neural networks use back-propagation for learning and bipolar sigmoid function is chosen as the activation function. The user's eye is scanned with a simple web camera with backlight compensation which is attached to a head fixation device. Neural networks significantly outperform parametric interpolation techniques: 1) the calibration procedure is faster as they require less calibration marks and 2) cursor control is more precise. The system in its current stage of development is able to distinguish regions at least on the level of desktop icons. The main limitation of the proposed method is the lack of head-pose invariance and its relative sensitivity to illumination (especially to incidental pupil reflections).     

Variable site-occupancy classification of N-linked glycosylation using artificial neural networks

A novel neural-network-based model has been developed for the prediction of N-linked glycosylation characteristics related to glycosylation site-occupancy. Intracellular oligosaccharide transfer to a polypeptide is known to be either robust or dependent upon culture conditions during pharmaceutical production. This glycan attachment is classified by the model as robust or variable and is based on an input of the polypeptide primary sequence around the site of glycosylation. The glycosylation model utilizes multiple recurrent neural networks followed by a perceptron classifier. The input length of the polypeptide chain around the site of glycosylation (glycosylation window) was optimized through multiple independent training sessions. Incorporation of five residues prior (n - 5) to the site of glycosylation (n) and four residues beyond (n + 4) the glycan attachment site led to optimal network performance. The size of the glycosylation window for site-occupancy determination is much larger than has been previously reported. This model was developed to evaluate the effects of theoretical polypeptide mutations on glycosylation site-occupancy characteristics. Following correct prediction of the model testing data set, 20 independent networks were used to predict site-occupancy characteristics of wild-type and mutants of the rabies virus glycoprotein (rgp). Simulation results strongly correlated with previously published experimental results (Kasturi, L.; Hegang, C.; Shakin-Eshleman, S. H. Regulation of N-linked core glycosylation: use of a site-directed mutagenesis approach to identify Asn-Xaa-Ser/Thr sequons that are poor oligosacchride acceptors. Biochem. J. 1997, 323, 415-419. Mellquist, J. L.; Kasturi, L.; Spitalnik, S. L.; Shakin-Eshleman, S. H. The amino acid following an Asn-X-Ser/Thr sequon is an important determinant of N-linked core glycosylation efficiency. Biochemistry 1998, 37, 6833-6837). Further simulations on purely theoretical sequences suggested that influences of charged residues were a subset of multiple mechanisms in the determination of glycosylation site-occupancy.     

PROSHIFT: protein chemical shift prediction using artificial neural networks

The importance of protein chemical shift values for the determination of three-dimensional protein structure has increased in recent years because of the large databases of protein structures with assigned chemical shift data. These databases have allowed the investigation of the quantitative relationship between chemical shift values obtained by liquid state NMR spectroscopy and the three-dimensional structure of proteins. A neural network was trained to predict the ¹H, ¹³C, and ¹⁵N of proteins using their three-dimensional structure as well as experimental conditions as input parameters. It achieves root mean square deviations of 0.3 ppm for hydrogen, 1.3 ppm for carbon, and 2.6 ppm for nitrogen chemical shifts. The model reflects important influences of the covalent structure as well as of the conformation not only for backbone atoms (as, e.g., the chemical shift index) but also for side-chain nuclei. For protein models with a RMSD smaller than 5 Å a correlation of the RMSD and the r.m.s. deviation between the predicted and the experimental chemical shift is obtained. Thus the method has the potential to not only support the assignment process of proteins but also help with the validation and the refinement of three-dimensional structural proposals. It is freely available for academic users at the PROSHIFT server: www.jens-meiler.de/proshift.html     

Application of model-predictive control based on artificial neural networks to optimize the fed-batch process for riboflavin production

The fed-batch process for commercial production of riboflavin (vitamin B2) was optimized on-line using model-predictive control based on artificial neural networks (ANNs). The information required for process models was extracted from both historical data and heuristic rules. After each cultivation the process model was readapted off-line to include the most recent process data. The control signal (feed rate), however, was optimized on-line at each sampling interval. An optimizer simulated variations in the control signal and assessed the forecasted model outputs according to an objective function. The optimum feed profile for increasing the product yield (YB2/S) and the amount of riboflavin at the time of harvesting was adjusted continuously and applied to the process. In contrast to the control by set-point profiles, the novel ANN-control is able to react on-line to variations in the process and also to incorporate the new process information continuously. As a result, both the total amount of riboflavin produced and the product yield increased systematically by more than 10% and the reproducibility of seven subsequently optimized batches was enhanced.