Abstract Various works have been published around the optimization of Neural Networks that emphasize the significance of the learning rate. In this study we analyze the need for a different treatment for each layer and how this affects training. We propose a novel optimization technique, called AdaLip, that utilizes an estimation of the Lipschitz constant of the gradients in order to construct an adaptive learning rate per layer that can work on top of already existing optimizers, like SGD or Adam. A detailed experimental framework was used to prove the usefulness of the optimizer on three benchmark datasets. It showed that AdaLip improves the training performance and the convergence speed, but also made the training process more robust to the selection of the initial global learning rate.

As Monte Carlo Tree Search has been established as one of the most promising algorithms in the field of Game AI, several approaches have been proposed in an attempt to exploit as much information as possible during the tree search, most important of which include Rapid Action Value Estimation and its variants. These techniques estimate for each action in a node an additional value (AMAF), based on statistics of all simulations where the action was selected deeper in the search tree. In this study, a methodology for determining the most suitable node for using its AMAF scores during the selection phase is presented. Two different approaches are proposed under the scope of discovering similar nodes’ states based on the actions selected towards their paths; in the first one, N-grams are employed to detect similar paths, while in the second one a vectorized representation of the actions taken is used. The suggested algorithms are tested in the context of general game playing achieving quite satisfactory results in terms of both win rate and overall score.

Abstract Nowadays, most research conducted in the field of abstractive text summarization focuses on neural-based models alone, without considering their combination with knowledge-based approaches that could further enhance their efficiency. In this direction, this work presents a novel framework that combines sequence-to-sequence neural-based text summarization along with structure and semantic-based methodologies. The proposed framework is capable of dealing with the problem of out-of-vocabulary or rare words, improving the performance of the deep learning models. The overall methodology is based on a well-defined theoretical model of knowledge-based content generalization and deep learning predictions for generating abstractive summaries. The framework is composed of three key elements: (i) a pre-processing task, (ii) a machine learning methodology, and (iii) a post-processing task. The pre-processing task is a knowledge-based approach, based on ontological knowledge resources, word sense disambiguation, and named entity recognition, along with content generalization, that transforms ordinary text into a generalized form. A deep learning model of attentive encoder-decoder architecture, which is expanded to enable a coping and coverage mechanism, as well as reinforcement learning and transformer-based architectures, is trained on a generalized version of text-summary pairs, learning to predict summaries in a generalized form. The post-processing task utilizes knowledge resources, word embeddings, word sense disambiguation, and heuristic algorithms based on text similarity methods in order to transform the generalized version of a predicted summary to a final, human-readable form. An extensive experimental procedure on three popular data sets evaluates key aspects of the proposed framework, while the obtained results exhibit promising performance, validating the robustness of the proposed approach.

Collectible card games are an interesting testing ground for artificial intelligence algorithms, mainly because of their stochasticity and high branching factor. In this work, the performance of a monte carlo tree search-based agent, enhanced with a gradient boosting tree classifier on the simulation phase, is investigated on Hearthstone. Furthermore, the impact of the combination of random simulations and the classifier's predictions is studied, as well as its correlation with the action space and the tree's depth. The aforementioned approach has been implemented in the Metastone framework and has been tested against the vanilla approach and the state-of-the-art algorithm, both provided by the framework itself. Over a set of evaluation games, it is demonstrated that the examined methodology significantly outperforms the vanilla-MCTS and is even matched with the heuristic-driven minimax algorithm.

Abstract Figurative language (FL) seems ubiquitous in all social media discussion forums and chats, posing extra challenges to sentiment analysis endeavors. Identification of FL schemas in short texts remains largely an unresolved issue in the broader field of natural language processing, mainly due to their contradictory and metaphorical meaning content. The main FL expression forms are sarcasm, irony and metaphor. In the present paper, we employ advanced deep learning methodologies to tackle the problem of identifying the aforementioned FL forms. Significantly extending our previous work (Potamias et al., in: International conference on engineering applications of neural networks, Springer, Berlin, pp 164–175, 2019), we propose a neural network methodology that builds on a recently proposed pre-trained transformer-based network architecture which is further enhanced with the employment and devise of a recurrent convolutional neural network. With this setup, data preprocessing is kept in minimum. The performance of the devised hybrid neural architecture is tested on four benchmark datasets, and contrasted with other relevant state-of-the-art methodologies and systems. Results demonstrate that the proposed methodology achieves state-of-the-art performance under all benchmark datasets, outperforming, even by a large margin, all other methodologies and published studies.

This work proposes a novel framework for enhancing abstractive text summarization based on the combination of deep learning techniques along with semantic data transformations.Initially, a theoretical model for semantic-based text generalization is introduced and used in conjunction with a deep encoder-decoder architecture in order to produce a summary in generalized form.Subsequently, a methodology is proposed which transforms the aforementioned generalized summary into human-readable form, retaining at the same time important informational aspects of the original text and addressing the problem of out-of-vocabulary or rare words.The overall approach is evaluated on two popular datasets with encouraging results.

Insufficient reasoning for their predictions has for long been a major drawback of neural networks and has proved to be a major obstacle for their adoption by several fields of application. This paper presents a framework for discriminative localization, which helps shed some light into the decision-making of Convolutional Neural Networks (CNN). Our framework generates robust, refined and high-quality Class Activation Maps, without impacting the CNN's performance.

This paper presents a new method for medical diagnosis of neurodegenerative diseases, such as Parkinson's, by extracting and using latent information from trained Deep convolutional, or convolutional-recurrent Neural Networks (DNNs). In particular, our approach adopts a combination of transfer learning, k-means clustering and k-Nearest Neighbour classification of deep neural network learned representations to provide enriched prediction of the disease based on MRI and/or DaT Scan data. A new loss function is introduced and used in the training of the DNNs, so as to perform adaptation of the generated learned representations between data from different medical environments. Results are presented using a recently published database of Parkinson's related information, which was generated and evaluated in a hospital environment.

This paper presents three conditions of Multi-Task Learning (MTL) model architectures based on Deep Neural Networks (DNNs) to predict the Parkinson's Disease (PD) from brain images. It also demonstrates the usefulness of incorporating additional patients' contextual epidemiological information (i.e. their age and sex). Our aim is to investigate which are the patients' clinical data, that when joined with the primary task could perform best and thus provide an improved computational PD prediction model. Our proposed model architectures are evaluated on a new medical dataset, which is presently under development. Our preliminary results suggest the robustness of our proposed systems to analyze and provide an accurate estimate of the status of the disease. Finally, we discuss the lessons learned from our experimental settings with respect to addressing several research questions such as the importance of selecting the auxiliary task(s) respectively, which is the best performing task combination to achieve such improved Parkinson's disease prediction, as well as whether all auxiliary tasks are equally effective.

Parkinson's disease (PD) is a long-term degenerative disorder of the central nervous system, with symptoms generally appearing slowly over time. Predicting the PD disease is critical as motor and non-motor manifestations occur many years after the onset of neurodegeneration, hence its early management of disease is a significant challenge in the field of PD therapeutics. While part of previous studies with respect to the prediction of Parkinson's Disease has been based mainly on brain images, dependencies between additional patients' information have not been taken into account. This observation suggests that prediction of Parkinson's Disease along with additional patients' data with a unified framework should outperform Machine Learning (ML) algorithms that treat different sources of patients' information separately. Our presented framework relies on Multi-Task Learning (MTL) implemented with Deep Neural Networks (DNNs) with shared hidden layers. Our preliminary experimental results confirm the benefits of MTL over Single-Task Learning (STL), underlying the capability of our proposed system to achieve an increased Area Under the Curve (AUC) as high as 92% and helping at the same time to reduce human error.

Integrated spiral inductors are a fundamental part of Radio-Frequency (RF) circuits. In certain scenarios, a solution to the inverse spiral inductor design problem is required; given the desired properties of an inductor, locate the most suitable geometric characteristics. This problem does not have a unique solution and current approaches approximate it through a number of differential equations and the subsequent application of optimization techniques that narrow down the set of feasible solutions. In this work, the Neural Network Specialists model is outlined; a preliminary approach to solving the aforementioned problem using fully connected neural network models. The obtained results on a first round of experiments are encouraging, especially in terms of the reduction in time complexity.

Usually, documents are given in textual form, accompanied by a set of terminological classifications (metadata), based on vocabularies of domain ontologies. This paper presents a novel method for advancing the above classification, by extracting more properties of the analyzed documents. We first extract additional roles from the textual part and together with roles extracted from the ontology statements, we construct an extended document vector representation. We then introduce a pruning algorithm that, for a given document collection, merges concepts of the ontology to produce classes with a sufficient number of corresponding instances. We then classify the documents to ontology classes using the Stanford linear Classifier. Finally, we propose an algorithm that assigns additional concept labels to documents, using the output of the classifier. Our system is evaluated in a set of real data and ontological descriptions and its performance is measured in terms of various accuracy and specificity measures indicates that the proposed approach for documents classification produces correct labels for the majority of items.

In this paper, we focus on Recommender Systems that are enhanced with social information in the form of trust statements between their users. The trust information may be processed in a number of ways, including the random walks in the Social Graph, where every step in the walk is chosen almost uniformly at random from the available choices. Even though this strategy yields satisfactory results, it still does not fully exploit the similarity information among users and items. Our work tries to model user-to-user and user-to-item relation as a probability distribution using a novel approach based on Rejection Sampling in order to decide on its next step (biased random walk). Some initial results on reference datasets reveal the potential of this idea.

In this paper, we propose a novel method for conceptual hierarchical clustering of documents using knowledge extracted from Wikipedia. The proposed method overcomes the classic bag-of-words models disadvantages through the exploitation of Wikipedia textual content and link structure. A robust and compact document representation is built in real-time using the Wikipedia application programmer's interface, without the need to store locally any Wikipedia information. The clustering process is hierarchical and extends the idea of frequent items by using Wikipedia article titles for selecting cluster labels that are descriptive and important for the examined corpus. Experiments show that the proposed technique greatly improves over the baseline approach, both in terms of F-measure and entropy on the one hand and computational cost on the other.

One of the most important problems in artificial intelligence, namely pattern classication, is essentially equivalent to finding hyper-surfaces seperating the different classes of data in a high-dimensional space. A method called raycasting is commonly used in computer game programming to locate and describe surfaces in a 2-dimesional map. A viewer, situated at some point on this map, casts rays of light towards various directions, and the rays extend until they hit a part of a surface. As a result, the surface can be described in polar coordinates as a set of vectors of varying lengths and angles. In this work we present a pattern classification system loosely based on the raycasting concept. We modify the method to create fuzzy pattern classification rules in 2 dimensions, and then generalize the rules to high-dimensional spaces. The resulting classifier is tested on a number of benchmark tests from the UCI Repository. © 2009 IEEE.

In this paper, we build a hybrid Web-based metric for computing semantic relatedness between words. The method exploits page counts, titles, snippets and URLs returned by a Web search engine. Our technique uses traditional information retrieval methods and is enhanced by page-count-based similarity scores which are integrated with automatically extracted lexico-synantic patterns from titles, snippets and URLs for all kinds of semantically related words provided by WordNet (synonyms, hypernyms, meronyms, antonyms). A support vector machine is used to solve the arising regression problem of word relatedness and the proposed method is evaluated on standard benchmark datasets. The method achieves an overall correlation of 0.88, which is the highest among other metrics up to date.

A new relevance feedback (RF) approach for content-based image retrieval is presented. This approach uses Gaussian mixture (GM) models of the image features and a query that is updated in a probabilistic manner. This update reflects the preferences of the user and is based on the models of both the positive and negative feedback images. The retrieval is based on a recently proposed distance measure between probability density functions, which can be computed in closed form for GM models. The proposed approach takes advantage of the form of this distance measure and updates it very efficiently based on the models of the user-specified relevant and irrelevant images. It is also shown that this RF framework is fairly general and can be applied in case other image models or distance measures are used instead of those proposed in this work. Finally, comparative numerical experiments are provided, which that demonstrate the merits of the proposed RF methodology and the use of the distance measure, and also the advantages of using GMs for image modelling.

Fuzzy logic programming has been lately used as a general framework for representing and handling imprecise knowledge. In this paper, we define the syntax and the semantics of definite weighted fuzzy logic programs, which extend definite fuzzy logic programs by allowing the inclusion of different significance weights in the individual atoms that make up the antecedent of a fuzzy logic rule. The weights add expressiveness to a fuzzy logic program and allow the determination of the level up to which an atom in the antecedent of a rule may affect the truth value of its consequent. In describing the semantics of definite weighted fuzzy logic programs we introduce the notion of the generalized weighted fuzzy conjunction operator, which can be regarded as a weighted t-norm based aggregation. We determine the properties of generalized weighted fuzzy conjunction operators and provide several examples. A methodology for constructing generalized weighted fuzzy conjunction operators using generator functions of existing t-norms is also introduced. Finally, a method for setting up a parametric weighted fuzzy logic program and automatically adapting the weights of its rules using a numerical dataset is developed.

A hybrid approach combining the self-organizing map (SOM) and the hidden Markov model (HMM) is presented. The self-organizing hidden Markov model map (SOHMMM) establishes a cross-section between the theoretic foundations and algorithmic realizations of its constituents. The respective architectures and learning methodologies are blended together in an attempt to meet the increasing requirements imposed by the deoxyribonucleic acid (DNA), ribonucleic acid (RNA), and protein chain molecules. Addressing many of the most intriguing biological sequence analysis problems is achieved through its automatic raw sequence data learning mechanism. Since the SOHMMM carries out probabilistic sequence analysis with little or no prior knowledge, it can have a variety of applications in clustering, dimensionality reduction and visualization of large-scale sequence spaces, and also, in sequence discrimination, search and classification. A comprehensive series of experiments based on the globin protein family demonstrates SOHMMMpsilas sophisticated characteristics and advanced capabilities.

Artificial immune systems (AIS) constitute an emerging and promising field, and have been applied to pattern recognition and classification tasks to a limited extent so far. This work is a first attempt of applying the clonal selection principle to the training of multi-layer perceptrons (MLPs). The clonal selection based neural classifier (CSNC) uses the basic concepts of clonal selection to evolve MLPs, which are represented as real-valued linear antibodies. The proposed system is actually a multi-classifier, consisting of multiple sets of MLPs, each one devoted to the recognition of a different class of the input data. The final trained classifier is comprised of the best MLPs from each set. The proposed classifier is tested against a set of benchmark problems and yields promising results.

This work presents the design and experimental verification of an original system architecture aiming at recognizing gestures based solely on the hand trajectory. Self organizing feature maps are used to model spatial information while Markov models encode the temporal aspect of hand position within a trajectory. A validated classification mechanism is produced through a set of models and a committee machine setup ensures robustness as indicated by the experimental results performed.

The self-organizing hidden Markov model map (SOHMMM) constitutes a cross-section between the theoretic foundations and algorithmic realizations of the self-organizing map (SOM) and the hidden Markov model (HMM). The intimate fusion and synergy of the SOM unsupervised training and HMM dynamic programming algorithms brings forth a novel on-line gradient descent learning algorithm, which is fully integrated into the SOHMMM. The model is presented from both a theoretical and algorithmic perspective. The SOHMMM can have a variety of applications in clustering, dimensionality reduction and visualization of large-scale sequence spaces, and also, in sequence discrimination, search and classification.

In this paper a relevance feedback (RF) approach for content based image retrieval (CBIR) is described and evaluated. The approach uses Gaussian mixture (GM) models of the image features and a query that is updated in a probabilistic manner. This update reflects the preferences of the user and is based on the models of both positive and negative feedback images. Retrieval is based on a recently proposed distance measure between probability density functions (pdfs), which can be computed in closed form for GM models. The proposed approach takes advantage of the form of this distance measure and updates it very efficiently based on the models of the user specified relevant and irrelevant images. For evaluation purposes, comparative experimental results are presented that demonstrate the merits of the proposed methodology.

A hybrid evolutionary technique is proposed for data mining tasks, which combines a principle inspired by the immune system, namely the clonal selection principle, with a more common, though very efficient, evolutionary technique, gene expression programming (GEP). The clonal selection principle regulates the immune response in order to successfully recognize and confront any foreign antigen, and at the same time allows the amelioration of the immune response across successive appearances of the same antigen. On the other hand, gene expression programming is the descendant of genetic algorithms and genetic programming and eliminates their main disadvantages, such as the genotype-phenotype coincidence, though it preserves their advantageous features. In order to perform the data mining task, the proposed algorithm introduces the notion of a data class antigen, which is used to represent a class of data, the produced rules are evolved by our clonal selection algorithm (CSA), which extends the recently proposed CLONALG algorithm. In CSA, among other new features, a receptor editing step has been incorporated. Moreover, the rules themselves are represented as antibodies that are coded as GEP chromosomes in order to exploit the flexibility and the expressiveness of such encoding. The proposed hybrid technique is tested on a set of benchmark problems in comparison to GEP. In almost all problems considered, the results are very satisfactory and outperform conventional GEP both in terms of prediction accuracy and computational efficiency.

A hybrid evolutionary technique is proposed for data mining tasks, which combines the Clonal Selection Principle with Gene Expression Programming (GEP). The proposed algorithm introduces the notion of Data Class Antigens, which is used to represent a class of data. The produced rules are evolved by a clonal selection algorithm, which extends the recently proposed CLONALG algorithm. In the present algorithm, among other new features, a receptor editing step has been incorporated. Moreover, the rules themselves are represented as antibodies, which are coded as GEP chromosomes, in order to exploit the flexibility and the expressiveness of such encoding. The algorithm is tested on some benchmark problems of the UCI repository, and in particular on the set of MONK problems and the Pima Indians Diabetes problem. In both problems, the results in terms of prediction accuracy are very satisfactory, albeit slightly less accurate than those obtained by a standard GEP technique. In terms of convergence rate and computational efficiency, however, the technique proposed here markedly outperforms the standard GEP algorithm.

The usefulness of the results of logic programming in real-life applications is sometimes limited due to the inability of this theory to model the uncertain and dynamic character of real environments. Fuzzy logic programming has been lately considered as an important framework for handling uncertainty in logic programming systems. Still, there is a need for modelling adaptation of logic programs and the progress in this area is rather slow. In the present paper, we first extend fuzzy logic programs in a direction that brings them closer to the connectionist approach: we introduce weighted fuzzy programs, which allow the association of significance weights with the atoms that make up the body of a logic rule. The weights add expressiveness to the programs and allow the determination of the degree with which an antecedent affects the value of the rule consequent. Then, we propose a neural network implementation of weighted fuzzy programs that is capable of computing the minimal Herbrand model of a weighted fuzzy program.

This paper presents a hybrid intelligent system made up of two modules. The bottom sub-symbolic module is a multi-layer feed-forward neural network trained by a modified Q-learning methodology that employs the scaled conjugate gradient algorithm. The top module is a symbolic system (implemented with a neural network built on-line) where rules are extracted from the bottom module during training, in a fashion similar to the CLARION system. The two modules augment each other in an effort to obtain a better performance than both of the modules acting alone in solving a problem. The originality of this work lies in the use of the advanced scaled conjugate learning algorithm in such a hybrid system. It is expected that the use of this algorithm would provide significant improvements in the performance of the overall system and also make it less dependent on user-selected parameters. This paper emphasises the implementation details, since the system is currently under development, rather that concrete experimental results.

Recently, there has been a lot of speculation among the members of the artificial intelligence community concerning the way AI can help with the problem of successful information search in the reservoirs of knowledge of Internet. Recommender systems provide a solution to this problem by giving individualized recommendations. Content-based and collaborative filtering are usually applied to predict these recommendations. A combination of the results of these two techniques is used in this work in order to construct a system that provides more precise recommendations concerning movies. The MovieLens data set was used to test the proposed hybrid system.

A multi-clustering fusion method is presented based on combining several runs of a clustering algorithm resulting in a common partition. More specifically, the results of several independent runs of the same clustering algorithm are appropriately combined to obtain a distinct partition of the data which is not affected by initialization and overcomes the instabilities of clustering methods. Subsequently, a fusion procedure is applied to the clusters generated during the previous phase to determine the optimal number of clusters in the data set according to some predefined criteria.

Computer aided diagnosis is based on classification of medical data by intelligent classifiers. Especially for medical purposes, the classification must be very efficient, as diagnosis demands a high rate of reliability. Under most circumstances, single classifiers, such as neural networks, support vector machines and decision trees, exhibit worse performance than ensemble combinations of them such as bagging and boosting. In order to further enhance performance, we propose here a combination of these combination methods in a multi-level combination scheme. After experimentation by using four medical diagnosis problems, the proposed approach seems to be efficient in decreasing the error, compared to the best combining method standalone.

We propose an intelligent agent-mediated trading environment for business-to-consumer transactions. The generic architecture of the system enables it to be used with semistructured product ontologies. The buying behavior model consists of the following fundamental trading agent-mediated stages: product brokering, merchant brokering and negotiation brokering. The product and merchant brokering stages involve three agents: the indexing agent and the retrieval agent, that collaborate through machine learning, text classification techniques and fuzzy logic to match potential buyers with products and merchants, and the profile agent that creates the buyer's profile using his/her relevance feedback. Finally, the negotiation stage includes two transaction agents: the buyer negotiator agent and the merchant negotiator agent, that participate in an auction between the buyers and the "suggested" merchant acting according to the transaction strategies of the buyers and merchant respectively.

This paper presents the results of a comparative study on the impact of various error functions in multilayer feedforward neural networks used for classification problems. The objective is the comparison of the properties of the error functions, both in terms of the training speed and the generalization ability. In an effort to avoid complexities introduced by more advanced learning algorithms, the simple backpropagation with momentum algorithm has been employed. A member of classification problems were solved with neural networks that have been trained with the usual mean square error function, the mean absolute error function, the cross-entropy or maximum likelihood function, the Kalman-Kwasny error function (1991), as well as a novel error function designed by the authors. The results indicate that, in most problems examined, an error function other than the usual mean square gives a better performance, both in terms of the number of epochs needed for training, as well as the obtained generalization ability of the trained network.

We propose a neural network method for the generation of symbolic expressions using reinforcement learning. According to the proposed method, a human decides on the kind and number of primitive functions which, with the appropriate composition (in the mathematical sense), can represent a mapping between two domains. The appropriate composition is achieved by an agent which tries many compositions and receives a reward depending on the quality of the composed function.

A neural network classifier is presented, which is based on geometrical fuzzy sets. Starting from the construction of the Voronoi diagram of the training patterns, an aggregation of Voronoi regions is performed leading to the identification of larger regions belonging exclusively to one of the pattern classes. The resulting scheme is a constructive algorithm that defines fuzzy clusters of patterns. Based on observations concerning the grade of membership of the training patterns to the created regions, decision probabilities are computed through which the final classification is performed. Experimental results concerning several classification problems indicate that the proposed method achieves high classification rates and compares favorably with other well-known approaches.

In this paper the Bipolar Random Network is described, which constitutes an extension of the Random Neural Network model and exhibits autoassociative memory capabilities. This model is characterized by the existence of positive and negative nodes and symmetrical behavior of positive and negative signals circulating in the network. The network's ability of acting as autoassociative memory is examined and several techniques are developed concerning storage and reconstruction of patterns. These approaches are either based on properties of the network or constitute adaptations of existing neural network techniques. The performance of the network under the proposed schemes has been investigated through experiments showing very good storage and reconstruction capabilities. Moreover, the scheme exhibiting the best behavior seems to outperform other well-known associative neural network models, achieving capacities that exceed 0.5n where n is the size of the network.

A neural network architecture is proposed and shown to be very effective in performing lossy compression of medical images. A novel ROI-JPEG technique is introduced as the coding platform, in which the neural architecture adoptively selects regions of interest (ROI's) in the images. By letting the selected ROI's be coded with high quality, in contrast to the rest of image areas, high compression ratios are achieved, while retaining the significant (from medical point of view) image content. The performance of the method is illustrated by means of experimental results in real life problems taken from pathology and telemedicine applications. © 1996 IEEE.

A multiscale method is described in the context of binary Hopfield-type neural networks. The appropriateness of the proposed technique for solving several classes of optimization problems is established by means of the notion of group update which is introduced here and investigated in relation to the properties of multiscaling. The method has been tested in the solution of partitioning and covering problems, for which an original mapping to Hopfield-type neural networks has been developed. Experimental results indicate that the multiscale approach is very effective in exploring the state-space of the problem and providing feasible solutions of acceptable quality, while at the same it offers a significant acceleration.

A technique is developed based on the use of a neural network model for performing information retrieval in a pictorial information system. The neural network provides autoassociative memory operation and allows the retrieval of stored symbolic images using erroneous or incomplete information as input. The network used is based on an adaptation of the random neural network model featuring positive and negative nodes and symmetrical behavior of positive and negative signals. The network architecture considered has hierarchical structure and allows two-level operation during learning and recall. An experimental software prototype, including an efficient graphical interface, has been implemented and tested. The performance of the system has been investigated through experiments under several schemes concerning storage and reconstruction of patterns. These schemes are either based on properties of the random network or constitute adaptations of known neural network techniques.>

It is of interest to determine whether loosely coupled multiprocessors can be profitably used for the solution of larger numerical problems. We present here a performance evaluation of the gain obtained by solving partial differential equation systems on such an architecture. The experimental setting is an LSI 11 based multimicroprocessor system using a fiber optics local area network designed and implemented at Laboratoire de Recherche en Informatique, Université Paris-Sud. The paper includes a discussion of the numerical methods and of their implementation, a performance model of the parallel processing system, and measurements taken on the experimental system. The experimentally validated theoretical results confirm the interest of our approach based on performance models.