Ebook: Neural Nets WIRN09

Human Beings leave, the Science continues.

This volume collects contributions to the 19th Italian Workshop of the Italian Society for Neural Network (SIREN). The conference held a few days after the death of prof. Maria Marinaro, who was a founder and a solid leader of the society. The conference was sad for this, but more intense at the same time. With neural networks we are exploring thought mechanisms that share the two features of an efficient computational tool and a representative of the physics of our brain, having the loops of our thoughts as an ultimate product. It is not a duty of our discipline sentencing what happens when these loops stop, but is a fascinating goal shedding light on how these loops run and which tracks they leave.

The Science continues, and we dedicate these selected papers to Maria. We have grouped them within the five themes of: “modeling”, “signal processing”, “economy and complexity”, “biological aspects”, and “general applications”. They come from three regular sessions of the conference plus two specific workshops on “Computational Intelligence for Economics and Finance” and “COST 2102: Cross Modal Analysis of Verbal and Nonverbal Communications”, respectively. The editors would like to thank the invited speakers as well as all those who contributed to the success of the workshops with papers of outstanding quality. Finally, special thanks go to the referees for their valuable input.

Input selection is found as a part of several machine learning tasks, either to improve performance or as the main goal. For instance, gene selection in bioinformatics is an input selection problem. However, as we prove in this paper, the reliability of input selection in the presence of high-dimensional data is affected by a small-sample problem. As a consequence of this effect, even completely random inputs have a chance to be selected as very useful, even if they are not relevant from the point of view of the underlying model. We express the probability of this event as a function of data cardinality and dimensionality, discuss the applicability of this analysis, and compute the probability for some data sets. We also show, as an illustration, some experimental results obtained by applying a specific input selection algorithm, previously presented by the authors, which show how inputs known to be random are consistently selected by the method.

Since real networks are noisy systems, in this work we investigate the dynamics of a deterministic model of gene networks affected by small random fluctuations. In this case jumps among different attractors are possible, thereby leading to an asymptotic dynamics different from that of the underlying deterministic model. The significance of the jumps among attractors is discussed. A key control parameter of this phenomenon is the size of the network, a fact that could lead to interesting consequences for the theories of natural and artificial evolving systems.

We present a sensitivity study of a wait and chase scheme introduced in a previous work to model the contact times between people belonging to a social community. The membership presupposes that, besides purely occasional encounters, people are motivated to meet other members of the community, while the social character of the latter makes each person met an equivalent target. This calls for a mobility in the family of Lévy jumps alternating a wandering period within a limited environment – waiting phase – with jumping to a new site constituting the target of a chase phase. In this paper we aim to connect specific features of single individual dynamics to the overall evolution of the social community in the true thread of the Palm calculus philosophy. We base this study on a large mobility track dataset expressly collected with this objective.

Inference on factor graphs with loops with the standard forward-backward algorithm, can give unpredictible results as messages can travel indefinitely in the system with no guarantee on convergence. We apply the exact method of cutset conditioning to Factor Graphs with loops starting from a fully developed three-variable example and providing comments and suggestions for distributed implementations.

The main objective of this work is the comparison between metabolic networks and neural networks (ANNs) in terms of their robustness and fault tolerance capabilities. In the context of metabolic networks errors are random removal of network nodes, while attacks are failures in the network caused intentionally. In the contest of neural networks errors are usually defined configurations of input submitted to the network that are affected by noise, while the failures are defined as the removal of some network neurons. This study have proven that ANNs are very robust networks, with respect to the presence of noise in the inputs, and the partial removal of some nodes, until it reached a critical threshold; while, metabolic networks are very tolerant to random failures (absence of a critical threshold), but extremely vulnerable to targeted attacks.

This paper describes general specifications and current status of the COST 2102 Italian Audio and Video Emotional Database collected to support the research effort of the COST Action 2102: “Cross Modal Analysis of Verbal and Nonverbal Communication” (http://cost2102.cs.stir.ac.uk/). Emphasis is placed on stimuli selection procedures, theoretical and practical aspects for stimuli identification, characteristics of selected stimuli and progresses in their assessment and validation.

Biometric template security and privacy are a great concern of biometric systems, because unlike passwords and tokens, compromised biometric templates cannot be revoked and reissued. In this paper we present a protection scheme for a face verification system based on a user dependent pseudo-random ordering of the DCT template coefficients and MPL and RBF Neural Networks for classification. In addition to privacy enhancement, because a hacker can hardly match a fake biometric sample without knowing the pseudo-random ordering this scheme, the proposed system also increases the biometric recognition performance.

In this paper, a speech-interfaced system for fostering group conversations is presented. The system captures conversation keywords and shows visual stimuli on a tabletop display. A stimulus can be a feedback to the current conversation or a cue to discuss new topics. This work describes the overall system architecture and highlights details about the design choices of the overall system, with a particular focus on the real-time implementation issues. A suitable speech enhancement front-end and a keyword spotter have been integrated on a common software platform for real-time audio processing, namely Nu-Tech, resulting in a helpful and flexible architecture for real-world applications in group conversation modeling scenarios. Such system characteristics, jointly with some experimental results obtained from simulations on recorded speech data, seem to confirm the efficacy of the approach motivating the development of further features and the experimentation in new scenarios.

In this paper a blind source separation algorithm in reverberant environment is presented. The algorithms working in such adverse environments are usually characterized by a huge computational cost. In order to reduce the computational complexity of this kind of algorithms a partitioned frequency domain approach is proposed. Several experimental results are shown to demonstrate the effectiveness of the proposed method.

Music transcription consists in transforming the musical content of audio data into a symbolic representation. The objective of this study is to investigate a transcription system for polyphonic piano. The proposed method focuses on temporal musical structures, note events and their main characteristics: the attack instant and the pitch. Onset detection exploits a time-frequency representation of the audio signal. Note classification is based on constant Q transform (CQT) and support vector machines (SVMs). Finally, to validate our method, we present a collection of experiments using a wide number of musical pieces of heterogeneous styles.

We propose a 3D self organizing neural model for modeling both the background and the foreground in video, helping in distinguishing between moving and stopped objects in the scene. Our aim is to detect foreground objects in digital image sequences taken from stationary cameras and to distinguish them into moving and stopped objects by a model based approach. We show through experimental results that a good discrimination can be achieved for color video sequences that represent typical situations critical for vehicles stopped in no parking areas.

Before training a feed forward neural network, one needs to define its architecture. Even in simple feed-forward networks, the number of neurons of the hidden layer is a fundamental parameter, but it is not generally possible to compute its optimal value a priori. It is good practice to start from an initial number of neurons, then build, train and test several different networks with a similar hidden layer size, but this can be excessively expensive when the data to be learned are simple, while some real-time constraints have to be satisfied. This paper shows a heuristic method for dimensioning and initializing a network under such assumptions. The method has been tested on a project for waste water treatment monitoring.

We applied and compared two supervised pattern recognition techniques, namely the Multilayer Perceptron (MLP) and Support Vector Machine (SVM), to classify seismic signals recorded on Stromboli volcano. The available data are firstly preprocessed in order to obtain a compact representation of the raw seismic signals. We extract from data spectral and temporal information so that each input vector is made up of 71 components, containing both spectral and temporal information extracted from the early signal. We implemented two classification strategies to discriminate three different seismic events: landslide, explosion-quake, and volcanic microtremor signals. The first method is a two-layer MLP network, with a Cross-Entropy error function and logistic activation function for the output units. The second method is a Support Vector Machine, whose multi-class setting is accomplished through a 1vsAll architecture with gaussian kernel. The experiments show that although the MLP produces very good results, the SVM accuracy is always higher, both in term of best performance, 99.5%, and average performance, 98.8%, obtained with different sampling permutations of training and test sets.

In this paper we consider a Multi Person Multi Attribute decision problem in which a finite number of alternatives has to be scored on the basis of a finite number of criteria, using different Stakeholders' judgments. In particular, first we propose a new individual preferences aggregation method which takes possible Stakeholder's inconsistencies into account, measured by a new consistency measure we call μ-consistency, then we focus the attention on the aggregation process of the Stakeholders' preference structure about the criteria, preference structure which is represented by non additive measures defined over the space of the criteria.

In this contribution we consider a dynamic portfolio optimization problem where the manager has to deal with the presence of minimum guarantee requirements on the performance of the portfolio. We briefly discuss different possibilities for the formulation of the problem and present a quite general formulation which includes transaction costs, cardinality constraints and buy-in thresholds. The presence of realistic and operational constraints introduces binary and integer variables greatly increasing the complexity of the problem.

Up to now, our research, based on the theory of fuzzy subsets, leads to an alternative implementation of a series of techniques arising from multivalent logics [1] to decision making in management of human resources, using known algorithms (like Hamming [2], Euclean [3], Tran y Duckstein [4] distances, adequation coefficient [5] or weighted mean hemimetric for fuzzy numbers [6]) even creating new instruments (such as the “maximum and minimum level index” [7] or “discarding by overcoming rate-distance index” [8]) which would allow us to adapt in an even more reliable manner to the always complex and unstable reality. In many cases, in order to avoid uncontrollable errors, the expert was advised, if in doubt, to work with confidence intervals in stead of crisp numbers which, allowed to limit the uncertainty to a greater degree and to do calculations that ensured that we avoided errors in estimates or valuations.

The results were satisfactory and, besides, it was possible to approach a new question: if experts could carry out their valuations in a more meticulous way if possible (by providing several possibilities between a minimum and maximum), valuing, for example, by means of fuzzy numbers [9], would it be feasible to operate in order to find in the end, Hamming distances with weighting? The answer is yes. Many manuals provide us with the techniques to, on the one hand, calculate distances and, on the other hand, operate with fuzzy numbers. But to find distances between fuzzy sub-sets of a degree α (for example) was a challenge which we have allowed ourselves to tackle.

In last year there has been a growing interest in computational intelligence techniques applied to economics, providing support for financial decisions. In this paper we propose an intelligent decision support system, aimed at suggesting the best managing strategies for a game-based model of a virtual city. Two knowledge representation areas characterize the intelligent agent. The first one is a “deterministic” area, which deals with descriptions and deterministic events. The second one is a “decisional” area, which deals with decisions taken under conditions of uncertainty. The agent is capable of reasoning in order to prospect the future evolutions of particular choices taken by the user. The interaction is conducted through a natural language interface built as an Alice-based conversational agent.

In this paper we present the results of a research project promoted by Italian Ministry of Labour devoted to assign a formal certify to Italian private firms who respect Equal Opportunities (EO) between man and woman. The research project has been implemented in the framework of the 2007 European Year of Equal Opportunities for All and has been co-financed by the European Commission. Italian Ministry made a public call, to choose a sample of private firms. The sample was formed by: 14 joint-stock companies, 2 Ltds, 3 cooperatives Ltd, 10 social cooperatives and 5 others form of enterprise. The group of researchers involved is makes by sociologists, mathematicians and economists. The project started in June 2007 when the experts set up the self-assessment questionnaire. The test was carried out from November 2007 until February 2008. The selected firms filled in the self-assessment questionnaires and the sociologists conducted in-depth interviews to the relevant union representatives. A Fuzzy Expert System (FES) is used. The scope for using FES is connected not only to the multidimensional nature of EO and to the need of providing a synthetic indicator of firm's EO without losing the its complexity, but even with the composite group of experts involved. The presence of sociologists, economists trade unions members, not used to a mathematics language, carried us to propose an instrument more user friendly like a FES.

Several approaches have been developed for forecasting mortality using stochastic model. In particular, the Lee Carter model (1992) has become widely used and there have been various extensions and modifications proposed to attain a broader interpretation and to capture the main features of the dynamics of the mortality intensity. Hyndman and Ullah (2005). introduce a particular version of the Lee Carter methodology, the so-called Functional Demographic Model – FDM, the most accurate approach as regards some mortality data, particularly for longer forecast horizons where the benefit of a damped trend forecast is greater. The paper objective is properly to single out the most suitable model between the basic Lee Carter and the FDM to the Italian mortality data. A comparative assessment is made. Moreover, we provide information on the uncertainty affecting the forecasted quantities by using bootstrap technique. The empirical results are presented using a range of graphical analyses.

Will mathematics be useful in overcoming the crisis in the Middle-East Area? The classical decisions methods are based only on numbers and do not include qualitative aspects which can add important and significant meanings to the subject. Furthermore, Mathematics is often reviled and considered unpleasant or difficult to be comprehended and so social researchers prefer to have nothing to do with it. The recent development of Fuzzy Logic has opened the possibility of new applications in situations in which both numbers and adjectives are important and have to be treated, simultaneously and on an equal standing; a mathematical theory where the traditional concept of true or false can be enriched by true and false. A Fuzzy Cognitive Map of the conflicts in the Middle-East has been realized and has been used to evaluate the failure of diplomatic and political efforts with the consequent onset of war and conflict. The paper is the result of the uncommon but joint effort between mathematicians, social researchers, sociologists and information experts.