Adaptive Pointing Theory (APT) Artificial Neural Network

The choice value and the testing process against the vigilance parameter, characteristic of ART Neural Network, are merged. Only, a single unique test is required to determine if a committed category node can represent the current input or not. Advantages of APT over ART are: 1-Avoid testing every committed category node before deciding to train a committed category node or a new node must be committed, 2-The vigilance parameter is fixed during training, and 3-The choice value parameter is eliminated.

Inteligencia artificial para la predicción y control del acabado superficial en procesos de fresado a alta velocidad

El objetivo principal de esta tesis doctoral es profundizar en el análisis y diseño de un sistema inteligente para la predicción y control del acabado superficial en un proceso de fresado a alta velocidad, basado fundamentalmente en clasificadores Bayesianos, con el prop´osito de desarrollar una metodolog´ıa que facilite el diseño de este tipo de sistemas. El sistema, cuyo propósito es posibilitar la predicción y control de la rugosidad superficial, se compone de un modelo aprendido a partir de datos experimentales con redes Bayesianas, que ayudar´a a comprender los procesos dinámicos involucrados en el mecanizado y las interacciones entre las variables relevantes. Dado que las redes neuronales artificiales son modelos ampliamente utilizados en procesos de corte de materiales, también se incluye un modelo para fresado usándolas, donde se introdujo la geometría y la dureza del material como variables novedosas hasta ahora no estudiadas en este contexto. Por lo tanto, una importante contribución en esta tesis son estos dos modelos para la predicción de la rugosidad superficial, que se comparan con respecto a diferentes aspectos: la influencia de las nuevas variables, los indicadores de evaluación del desempeño, interpretabilidad. Uno de los principales problemas en la modelización con clasificadores Bayesianos es la comprensión de las enormes tablas de probabilidad a posteriori producidas. Introducimos un m´etodo de explicación que genera un conjunto de reglas obtenidas de árboles de decisión. Estos árboles son inducidos a partir de un conjunto de datos simulados generados de las probabilidades a posteriori de la variable clase, calculadas con la red Bayesiana aprendida a partir de un conjunto de datos de entrenamiento. Por último, contribuimos en el campo multiobjetivo en el caso de que algunos de los objetivos no se puedan cuantificar en números reales, sino como funciones en intervalo de valores. Esto ocurre a menudo en aplicaciones de aprendizaje automático, especialmente las basadas en clasificación supervisada. En concreto, se extienden las ideas de dominancia y frontera de Pareto a esta situación. Su aplicación a los estudios de predicción de la rugosidad superficial en el caso de maximizar al mismo tiempo la sensibilidad y la especificidad del clasificador inducido de la red Bayesiana, y no solo maximizar la tasa de clasificación correcta. Los intervalos de estos dos objetivos provienen de un m´etodo de estimación honesta de ambos objetivos, como e.g. validación cruzada en k rodajas o bootstrap.---ABSTRACT---The main objective of this PhD Thesis is to go more deeply into the analysis and design of an intelligent system for surface roughness prediction and control in the end-milling machining process, based fundamentally on Bayesian network classifiers, with the aim of developing a methodology that makes easier the design of this type of systems. The system, whose purpose is to make possible the surface roughness prediction and control, consists of a model learnt from experimental data with the aid of Bayesian networks, that will help to understand the dynamic processes involved in the machining and the interactions among the relevant variables. Since artificial neural networks are models widely used in material cutting proceses, we include also an end-milling model using them, where the geometry and hardness of the piecework are introduced as novel variables not studied so far within this context. Thus, an important contribution in this thesis is these two models for surface roughness prediction, that are then compared with respecto to different aspects: influence of the new variables, performance evaluation metrics, interpretability. One of the main problems with Bayesian classifier-based modelling is the understanding of the enormous posterior probabilitiy tables produced. We introduce an explanation method that generates a set of rules obtained from decision trees. Such trees are induced from a simulated data set generated from the posterior probabilities of the class variable, calculated with the Bayesian network learned from a training data set. Finally, we contribute in the multi-objective field in the case that some of the objectives cannot be quantified as real numbers but as interval-valued functions. This often occurs in machine learning applications, especially those based on supervised classification. Specifically, the dominance and Pareto front ideas are extended to this setting. Its application to the surface roughness prediction studies the case of maximizing simultaneously the sensitivity and specificity of the induced Bayesian network classifier, rather than only maximizing the correct classification rate. Intervals in these two objectives come from a honest estimation method of both objectives, like e.g. k-fold cross-validation or bootstrap.

Overtopping of harbour breakwaters: a comparison of semi-empirical equations, neural networks, and physical model tests

This paper reports extensive tests of empirical equations developed by different authors for harbour breakwater overtopping. First, the existing equations are compiled and evaluated as tools for estimating the overtopping rates on sloping and vertical breakwaters. These equations are then tested using the data obtained in a number of laboratory studies performed in the Centre for Harbours and Coastal Studies of the CEDEX, Spain. It was found that the recommended application ranges of the empirical equations typically deviate from those revealed in the experimental tests. In addition, a neural network model developed within the European CLASH Project is tested. The wind effects on overtopping are also assessed using a reduced scale physical model

Monitorización de estructuras aeronáuticas mediante técnicas de inteligencia artificial

Una de las barreras para la aplicación de las técnicas de monitorización de la integridad estructural (SHM) basadas en ondas elásticas guiadas (GLW) en aeronaves es la influencia perniciosa de las condiciones ambientales y de operación (EOC). En esta tesis se ha estudiado dicha influencia y la compensación de la misma, particularizando en variaciones del estado de carga y temperatura. La compensación de dichos efectos se fundamenta en Redes Neuronales Artificiales (ANN) empleando datos experimentales procesados con la Transformada Chirplet. Los cambios en la geometría y en las propiedades del material respecto al estado inicial de la estructura (lo daños) provocan cambios en la forma de onda de las GLW (lo que denominamos característica sensible al daño o DSF). Mediante técnicas de tratamiento de señal se puede buscar una relación entre dichas variaciones y los daños, esto se conoce como SHM. Sin embargo, las variaciones en las EOC producen también cambios en los datos adquiridos relativos a las GLW (DSF) que provocan errores en los algoritmos de diagnóstico de daño (SHM). Esto sucede porque las firmas de daño y de las EOC en la DSF son del mismo orden. Por lo tanto, es necesario cuantificar y compensar el efecto de las EOC sobre la GLW. Si bien existen diversas metodologías para compensar los efectos de las EOC como por ejemplo “Optimal Baseline Selection” (OBS) o “Baseline Signal Stretching” (BSS), estas, se emplean exclusivamente en la compensación de los efectos térmicos. El método propuesto en esta tesis mezcla análisis de datos experimentales, como en el método OBS, y modelos basados en Redes Neuronales Artificiales (ANN) que reemplazan el modelado físico requerido por el método BSS. El análisis de datos experimentales consiste en aplicar la Transformada Chirplet (CT) para extraer la firma de las EOC sobre la DSF. Con esta información, obtenida bajo diversas EOC, se entrena una ANN. A continuación, la ANN actuará como un interpolador de referencias de la estructura sin daño, generando información de referencia para cualquier EOC. La comparación de las mediciones reales de la DSF con los valores simulados por la ANN, dará como resultado la firma daño en la DSF, lo que permite el diagnóstico de daño. Este esquema se ha aplicado y verificado, en diversas EOC, para una estructura unidimensional con un único camino de daño, y para una estructura representativa de un fuselaje de una aeronave, con curvatura y múltiples elementos rigidizadores, sometida a un estado de cargas complejo, con múltiples caminos de daños. Los efectos de las EOC se han estudiado en detalle en la estructura unidimensional y se han generalizado para el fuselaje, demostrando la independencia del método respecto a la configuración de la estructura y el tipo de sensores utilizados para la adquisición de datos GLW. Por otra parte, esta metodología se puede utilizar para la compensación simultánea de una variedad medible de EOC, que afecten a la adquisición de datos de la onda elástica guiada. El principal resultado entre otros, de esta tesis, es la metodología CT-ANN para la compensación de EOC en técnicas SHM basadas en ondas elásticas guiadas para el diagnóstico de daño. ABSTRACT One of the open problems to implement Structural Health Monitoring techniques based on elastic guided waves in real aircraft structures at operation is the influence of the environmental and operational conditions (EOC) on the damage diagnosis problem. This thesis deals with the compensation of these environmental and operational effects, specifically, the temperature and the external loading, by the use of the Chirplet Transform working with Artificial Neural Networks. It is well known that the guided elastic wave form is affected by the damage appearance (what is known as the damage sensitive feature or DSF). The DSF is modified by the temperature and by the load applied to the structure. The EOC promotes variations in the acquired data (DSF) and cause mistakes in damage diagnosis algorithms. This effect promotes changes on the waveform due to the EOC variations of the same order than the damage occurrence. It is difficult to separate both effects in order to avoid damage diagnosis mistakes. Therefore it is necessary to quantify and compensate the effect of EOC over the GLW forms. There are several approaches to compensate the EOC effects such as Optimal Baseline Selection (OBS) or Baseline Signal Stretching (BSS). Usually, they are used for temperature compensation. The new method proposed here mixes experimental data analysis, as in the OBS method, and Artificial Neural Network (ANN) models to replace the physical modelling which involves the BSS method. The experimental data analysis studied is based on apply the Chirplet Transform (CT) to extract the EOC signature on the DSF. The information obtained varying EOC is employed to train an ANN. Then, the ANN will act as a baselines interpolator of the undamaged structure. The ANN generates reference information at any EOC. By comparing real measurements of the DSF against the ANN simulated values, the damage signature appears clearly in the DSF, enabling an accurate damage diagnosis. This schema has been applied in a range of EOC for a one-dimensional structure containing single damage path and two dimensional real fuselage structure with stiffener elements and multiple damage paths. The EOC effects tested in the one-dimensional structure have been generalized to the fuselage showing its independence from structural arrangement and the type of sensors used for GLW data acquisition. Moreover, it can be used for the simultaneous compensation of a variety of measurable EOC, which affects the guided wave data acquisition. The main result, among others, of this thesis is the CT-ANN methodology for the compensation of EOC in GLW based SHM technique for damage diagnosis.

Nepfix: on-demand cloud elastic simulation of networks of evolutionary processors

Una Red de Procesadores Evolutivos o NEP (por sus siglas en ingles), es un modelo computacional inspirado por el modelo evolutivo de las celulas, específicamente por las reglas de multiplicación de las mismas. Esta inspiración hace que el modelo sea una abstracción sintactica de la manipulation de information de las celulas. En particu¬lar, una NEP define una maquina de cómputo teorica capaz de resolver problemas NP completos de manera eficiente en tóerminos de tiempo. En la praóctica, se espera que las NEP simuladas en móaquinas computacionales convencionales puedan resolver prob¬lemas reales complejos (que requieran ser altamente escalables) a cambio de una alta complejidad espacial. En el modelo NEP, las cóelulas estóan representadas por palabras que codifican sus secuencias de ADN. Informalmente, en cualquier momento de cómputo del sistema, su estado evolutivo se describe como un coleccion de palabras, donde cada una de ellas representa una celula. Estos momentos fijos de evolucion se denominan configuraciones. De manera similar al modelo biologico, las palabras (celulas) mutan y se dividen en base a bio-operaciones sencillas, pero solo aquellas palabras aptas (como ocurre de forma parecida en proceso de selection natural) seran conservadas para la siguiente configuracióon. Una NEP como herramienta de computation, define una arquitectura paralela y distribuida de procesamiento simbolico, en otras palabras, una red de procesadores de lenguajes. Desde el momento en que el modelo fue propuesto a la comunidad científica en el año 2001, múltiples variantes se han desarrollado y sus propiedades respecto a la completitud computacional, eficiencia y universalidad han sido ampliamente estudiadas y demostradas. En la actualidad, por tanto, podemos considerar que el modelo teórico NEP se encuentra en el estadio de la madurez. La motivación principal de este Proyecto de Fin de Grado, es proponer una aproxi-mación práctica que permita dar un salto del modelo teórico NEP a una implantación real que permita su ejecucion en plataformas computacionales de alto rendimiento, con el fin de solucionar problemas complejos que demanda la sociedad actual. Hasta el momento, las herramientas desarrolladas para la simulation del modelo NEP, si bien correctas y con resultados satisfactorios, normalmente estón atadas a su entorno de ejecucion, ya sea el uso de hardware específico o implementaciones particulares de un problema. En este contexto, el propósito fundamental de este trabajo es el desarrollo de Nepfix, una herramienta generica y extensible para la ejecucion de cualquier algo¬ritmo de un modelo NEP (o alguna de sus variantes), ya sea de forma local, como una aplicación tradicional, o distribuida utilizando los servicios de la nube. Nepfix es una aplicacion software desarrollada durante 7 meses y que actualmente se encuentra en su segunda iteration, una vez abandonada la fase de prototipo. Nepfix ha sido disenada como una aplicacion modular escrita en Java 8 y autocontenida, es decir, no requiere de un entorno de ejecucion específico (cualquier maquina virtual de Java es un contenedor vólido). Nepfix contiene dos componentes o móodulos. El primer móodulo corresponde a la ejecución de una NEP y es por lo tanto, el simulador. Para su desarrollo, se ha tenido en cuenta el estado actual del modelo, es decir, las definiciones de los procesadores y filtros mas comunes que conforman la familia del modelo NEP. Adicionalmente, este componente ofrece flexibilidad en la ejecucion, pudiendo ampliar las capacidades del simulador sin modificar Nepfix, usando para ello un lenguaje de scripting. Dentro del desarrollo de este componente, tambióen se ha definido un estóandar de representacióon del modelo NEP basado en el formato JSON y se propone una forma de representation y codificación de las palabras, necesaria para la comunicación entre servidores. Adicional-mente, una característica importante de este componente, es que se puede considerar una aplicacion aislada y por tanto, la estrategia de distribution y ejecución son total-mente independientes. El segundo moódulo, corresponde a la distribucióon de Nepfix en la nube. Este de-sarrollo es el resultado de un proceso de i+D, que tiene una componente científica considerable. Vale la pena resaltar el desarrollo de este modulo no solo por los resul-tados prócticos esperados, sino por el proceso de investigation que se se debe abordar con esta nueva perspectiva para la ejecución de sistemas de computación natural. La principal característica de las aplicaciones que se ejecutan en la nube es que son gestionadas por la plataforma y normalmente se encapsulan en un contenedor. En el caso de Nepfix, este contenedor es una aplicacion Spring que utiliza el protocolo HTTP o AMQP para comunicarse con el resto de instancias. Como valor añadido, Nepfix aborda dos perspectivas de implementation distintas (que han sido desarrolladas en dos iteraciones diferentes) del modelo de distribution y ejecucion, que tienen un impacto muy significativo en las capacidades y restricciones del simulador. En concreto, la primera iteration utiliza un modelo de ejecucion asincrono. En esta perspectiva asincrona, los componentes de la red NEP (procesadores y filtros) son considerados como elementos reactivos a la necesidad de procesar una palabra. Esta implementation es una optimization de una topologia comun en el modelo NEP que permite utilizar herramientas de la nube para lograr un escalado transparente (en lo ref¬erente al balance de carga entre procesadores) pero produce efectos no deseados como indeterminacion en el orden de los resultados o imposibilidad de distribuir eficiente-mente redes fuertemente interconectadas. Por otro lado, la segunda iteration corresponde al modelo de ejecucion sincrono. Los elementos de una red NEP siguen un ciclo inicio-computo-sincronizacion hasta que el problema se ha resuelto. Esta perspectiva sincrona representa fielmente al modelo teórico NEP pero el proceso de sincronizacion es costoso y requiere de infraestructura adicional. En concreto, se requiere un servidor de colas de mensajes RabbitMQ. Sin embargo, en esta perspectiva los beneficios para problemas suficientemente grandes superan a los inconvenientes, ya que la distribuciín es inmediata (no hay restricciones), aunque el proceso de escalado no es trivial. En definitiva, el concepto de Nepfix como marco computacional se puede considerar satisfactorio: la tecnología es viable y los primeros resultados confirman que las carac-terísticas que se buscaban originalmente se han conseguido. Muchos frentes quedan abiertos para futuras investigaciones. En este documento se proponen algunas aproxi-maciones a la solucion de los problemas identificados como la recuperacion de errores y la division dinamica de una NEP en diferentes subdominios. Por otra parte, otros prob-lemas, lejos del alcance de este proyecto, quedan abiertos a un futuro desarrollo como por ejemplo, la estandarización de la representación de las palabras y optimizaciones en la ejecucion del modelo síncrono. Finalmente, algunos resultados preliminares de este Proyecto de Fin de Grado han sido presentados recientemente en formato de artículo científico en la "International Work-Conference on Artificial Neural Networks (IWANN)-2015" y publicados en "Ad-vances in Computational Intelligence" volumen 9094 de "Lecture Notes in Computer Science" de Springer International Publishing. Lo anterior, es una confirmation de que este trabajo mas que un Proyecto de Fin de Grado, es solo el inicio de un trabajo que puede tener mayor repercusion en la comunidad científica. Abstract Network of Evolutionary Processors -NEP is a computational model inspired by the evolution of cell populations, which might model some properties of evolving cell communities at the syntactical level. NEP defines theoretical computing devices able to solve NP complete problems in an efficient manner. In this model, cells are represented by words which encode their DNA sequences. Informally, at any moment of time, the evolutionary system is described by a collection of words, where each word represents one cell. Cells belong to species and their community evolves according to mutations and division which are defined by operations on words. Only those cells are accepted as surviving (correct) ones which are represented by a word in a given set of words, called the genotype space of the species. This feature is analogous with the natural process of evolution. Formally, NEP is based on an architecture for parallel and distributed processing, in other words, a network of language processors. Since the date when NEP was pro¬posed, several extensions and variants have appeared engendering a new set of models named Networks of Bio-inspired Processors (NBP). During this time, several works have proved the computational power of NBP. Specifically, their efficiency, universality, and computational completeness have been thoroughly investigated. Therefore, we can say that the NEP model has reached its maturity. The main motivation for this End of Grade project (EOG project in short) is to propose a practical approximation that allows to close the gap between theoretical NEP model and a practical implementation in high performing computational platforms in order to solve some of high the high complexity problems society requires today. Up until now tools developed to simulate NEPs, while correct and successful, are usu¬ally tightly coupled to the execution environment, using specific software frameworks (Hadoop) or direct hardware usage (GPUs). Within this context the main purpose of this work is the development of Nepfix, a generic and extensible tool that aims to execute algorithms based on NEP model and compatible variants in a local way, similar to a traditional application or in a distributed cloud environment. Nepfix as an application was developed during a 7 month cycle and is undergoing its second iteration once the prototype period was abandoned. Nepfix is designed as a modular self-contained application written in Java 8, that is, no additional external dependencies are required and it does not rely on an specific execution environment, any JVM is a valid container. Nepfix is made of two components or modules. The first module corresponds to the NEP execution and therefore simulation. During the development the current state of the theoretical model was used as a reference including most common filters and processors. Additionally extensibility is provided by the use of Python as a scripting language to run custom logic. Along with the simulation a definition language for NEP has been defined based on JSON as well as a mechanisms to represent words and their possible manipulations. NEP simulator is isolated from distribution and as mentioned before different applications that include it as a dependency are possible, the distribution of NEPs is an example of this. The second module corresponds to executing Nepfix in the cloud. The development carried a heavy R&D process since this front was not explored by other research groups until now. It's important to point out that the development of this module is not focused on results at this point in time, instead we focus on feasibility and discovery of this new perspective to execute natural computing systems and NEPs specifically. The main properties of cloud applications is that they are managed by the platform and are encapsulated in a container. For Nepfix a Spring application becomes the container and the HTTP or AMQP protocols are used for communication with the rest of the instances. Different execution perspectives were studied, namely asynchronous and synchronous models were developed for solving different kind of problems using NEPs. Different limitations and restrictions manifest in both models and are explored in detail in the respective chapters. In conclusion we can consider that Nepfix as a computational framework is suc-cessful: Cloud technology is ready for the challenge and the first results reassure that the properties Nepfix project pursued were met. Many investigation branches are left open for future investigations. In this EOG implementation guidelines are proposed for some of them like error recovery or dynamic NEP splitting. On the other hand other interesting problems that were not in the scope of this project were identified during development like word representation standardization or NEP model optimizations. As a confirmation that the results of this work can be useful to the scientific com-munity a preliminary version of this project was published in The International Work- Conference on Artificial Neural Networks (IWANN) in May 2015. Development has not stopped since that point and while Nepfix in it's current state can not be consid¬ered a final product the most relevant ideas, possible problems and solutions that were produced during the seven months development cycle are worthy to be gathered and presented giving a meaning to this EOG work.

Simple neural networks for the amplification and utilization of small changes in neuron firing rates

I describe physiologically plausible “voter-coincidence” neural networks such that secondary “coincidence” neurons fire on the simultaneous receipt of sufficiently large sets of input pulses from primary sets of neurons. The networks operate such that the firing rate of the secondary, output neurons increases (or decreases) sharply when the mean firing rate of primary neurons increases (or decreases) to a much smaller degree. In certain sensory systems, signals that are generally smaller than the noise levels of individual primary detectors, are manifest in very small increases in the firing rates of sets of afferent neurons. For such systems, this kind of network can act to generate relatively large changes in the firing rate of secondary “coincidence” neurons. These differential amplification systems can be cascaded to generate sharp, “yes–no” spike signals that can direct behavioral responses.

Self-organized phase transitions in neural networks as a neural mechanism of information processing.

Transitions between dynamically stable activity patterns imposed on an associative neural network are shown to be induced by self-organized infinitesimal changes in synaptic connection strength and to be a kind of phase transition. A key event for the neural process of information processing in a population coding scheme is transition between the activity patterns encoding usual entities. We propose that the infinitesimal and short-term synaptic changes based on the Hebbian learning rule are the driving force for the transition. The phase transition between the following two dynamical stable states is studied in detail, the state where the firing pattern is changed temporally so as to itinerate among several patterns and the state where the firing pattern is fixed to one of several patterns. The phase transition from the pattern itinerant state to a pattern fixed state may be induced by the Hebbian learning process under a weak input relevant to the fixed pattern. The reverse transition may be induced by the Hebbian unlearning process without input. The former transition is considered as recognition of the input stimulus, while the latter is considered as clearing of the used input data to get ready for new input. To ensure that information processing based on the phase transition can be made by the infinitesimal and short-term synaptic changes, it is absolutely necessary that the network always stays near the critical state corresponding to the phase transition point.

Growing Neural Gas approach for obtaining homogeneous maps by restricting the insertion of new nodes

The Growing Neural Gas model is used widely in artificial neural networks. However, its application is limited in some contexts by the proliferation of nodes in dense areas of the input space. In this study, we introduce some modifications to address this problem by imposing three restrictions on the insertion of new nodes. Each restriction aims to maintain the homogeneous values of selected criteria. One criterion is related to the square error of classification and an alternative approach is proposed for avoiding additional computational costs. Three parameters are added that allow the regulation of the restriction criteria. The resulting algorithm allows models to be obtained that suit specific needs by specifying meaningful parameters.

Improving 3D Keypoint Detection from Noisy Data Using Growing Neural Gas

3D sensors provides valuable information for mobile robotic tasks like scene classification or object recognition, but these sensors often produce noisy data that makes impossible applying classical keypoint detection and feature extraction techniques. Therefore, noise removal and downsampling have become essential steps in 3D data processing. In this work, we propose the use of a 3D filtering and down-sampling technique based on a Growing Neural Gas (GNG) network. GNG method is able to deal with outliers presents in the input data. These features allows to represent 3D spaces, obtaining an induced Delaunay Triangulation of the input space. Experiments show how the state-of-the-art keypoint detectors improve their performance using GNG output representation as input data. Descriptors extracted on improved keypoints perform better matching in robotics applications as 3D scene registration.

Human Behaviour Recognition based on Trajectory Analysis using Neural Networks

Automated human behaviour analysis has been, and still remains, a challenging problem. It has been dealt from different points of views: from primitive actions to human interaction recognition. This paper is focused on trajectory analysis which allows a simple high level understanding of complex human behaviour. It is proposed a novel representation method of trajectory data, called Activity Description Vector (ADV) based on the number of occurrences of a person is in a specific point of the scenario and the local movements that perform in it. The ADV is calculated for each cell of the scenario in which it is spatially sampled obtaining a cue for different clustering methods. The ADV representation has been tested as the input of several classic classifiers and compared to other approaches using CAVIAR dataset sequences obtaining great accuracy in the recognition of the behaviour of people in a Shopping Centre.

Typing mineral deposits using their grades and tonnages in an artificial neural network

A test of the ability of a probabilistic neural network to classify deposits into types on the basis of deposit tonnage and average Cu, Mo, Ag, Au, Zn, and Pb grades is conducted. The purpose is to examine whether this type of system might serve as a basis for integrating geoscience information available in large mineral databases to classify sites by deposit type. Benefits of proper classification of many sites in large regions are relatively rapid identification of terranes permissive for deposit types and recognition of specific sites perhaps worthy of exploring further. Total tonnages and average grades of 1,137 well-explored deposits identified in published grade and tonnage models representing 13 deposit types were used to train and test the network. Tonnages were transformed by logarithms and grades by square roots to reduce effects of skewness. All values were scaled by subtracting the variable's mean and dividing by its standard deviation. Half of the deposits were selected randomly to be used in training the probabilistic neural network and the other half were used for independent testing. Tests were performed with a probabilistic neural network employing a Gaussian kernel and separate sigma weights for each class (type) and each variable (grade or tonnage). Deposit types were selected to challenge the neural network. For many types, tonnages or average grades are significantly different from other types, but individual deposits may plot in the grade and tonnage space of more than one type. Porphyry Cu, porphyry Cu-Au, and porphyry Cu-Mo types have similar tonnages and relatively small differences in grades. Redbed Cu deposits typically have tonnages that could be confused with porphyry Cu deposits, also contain Cu and, in some situations, Ag. Cyprus and kuroko massive sulfide types have about the same tonnages. Cu, Zn, Ag, and Au grades. Polymetallic vein, sedimentary exhalative Zn-Pb, and Zn-Pb skarn types contain many of the same metals. Sediment-hosted Au, Comstock Au-Ag, and low-sulfide Au-quartz vein types are principally Au deposits with differing amounts of Ag. Given the intent to test the neural network under the most difficult conditions, an overall 75% agreement between the experts and the neural network is considered excellent. Among the largestclassification errors are skarn Zn-Pb and Cyprus massive sulfide deposits classed by the neuralnetwork as kuroko massive sulfides—24 and 63% error respectively. Other large errors are the classification of 92% of porphyry Cu-Mo as porphyry Cu deposits. Most of the larger classification errors involve 25 or fewer training deposits, suggesting that some errors might be the result of small sample size. About 91% of the gold deposit types were classed properly and 98% of porphyry Cu deposits were classes as some type of porphyry Cu deposit. An experienced economic geologist would not make many of the classification errors that were made by the neural network because the geologic settings of deposits would be used to reduce errors. In a separate test, the probabilistic neural network correctly classed 93% of 336 deposits in eight deposit types when trained with presence or absence of 58 minerals and six generalized rock types. The overall success rate of the probabilistic neural network when trained on tonnage and average grades would probably be more than 90% with additional information on the presence of a few rock types.

Introduction to Special issue on the Symposium on the Application of Neural Networks to the Earth Sciences

Papers in this issue of Natural Resources Research are from the “Symposium on the Application of Neural Networks to the Earth Sciences,” held 20–21 August 2002 at NASA Moffet Field, Mountain View, California. The Symposium represents the Seventh International Symposium on Mineral Exploration (ISME-02). It was sponsored by the Mining and Materials Processing Institute of Japan (MMIJ), the US Geological Survey, the Circum-Pacific Council, and NASA. The ISME symposia have been held every two years in order to bring together scientists actively working on diverse quantitative methods applied to the earth sciences. Although the title, International Symposium on Mineral Exploration, suggests exclusive focus on mineral exploration, interests and presentations always have been wide-ranging—talks presented at this symposium are no exception.

GANN: Genetic algorithm neural networks for the detection of conserved combinations of features in DNA

Background: The multitude of motif detection algorithms developed to date have largely focused on the detection of patterns in primary sequence. Since sequence-dependent DNA structure and flexibility may also play a role in protein-DNA interactions, the simultaneous exploration of sequence-and structure-based hypotheses about the composition of binding sites and the ordering of features in a regulatory region should be considered as well. The consideration of structural features requires the development of new detection tools that can deal with data types other than primary sequence. Results: GANN ( available at http://bioinformatics.org.au/gann) is a machine learning tool for the detection of conserved features in DNA. The software suite contains programs to extract different regions of genomic DNA from flat files and convert these sequences to indices that reflect sequence and structural composition or the presence of specific protein binding sites. The machine learning component allows the classification of different types of sequences based on subsamples of these indices, and can identify the best combinations of indices and machine learning architecture for sequence discrimination. Another key feature of GANN is the replicated splitting of data into training and test sets, and the implementation of negative controls. In validation experiments, GANN successfully merged important sequence and structural features to yield good predictive models for synthetic and real regulatory regions. Conclusion: GANN is a flexible tool that can search through large sets of sequence and structural feature combinations to identify those that best characterize a set of sequences.

The applicability of recurrent neural networks for biological sequence analysis

Selection of machine learning techniques requires a certain sensitivity to the requirements of the problem. In particular, the problem can be made more tractable by deliberately using algorithms that are biased toward solutions of the requisite kind. In this paper, we argue that recurrent neural networks have a natural bias toward a problem domain of which biological sequence analysis tasks are a subset. We use experiments with synthetic data to illustrate this bias. We then demonstrate that this bias can be exploitable using a data set of protein sequences containing several classes of subcellular localization targeting peptides. The results show that, compared with feed forward, recurrent neural networks will generally perform better on sequence analysis tasks. Furthermore, as the patterns within the sequence become more ambiguous, the choice of specific recurrent architecture becomes more critical.