Search

Recruitment learning is prone to instability when a chain of concepts is recruited in cascade as seen in Figure 8.1. The statistical variance inherent in the recruitment method causes increasing perturbations to the recruited set size, and thus instability (Valiant, 1994). We previously proposed a boost-and-limit algorithm to improve recruitment stability, and verified the applicability of this method with a software model in a spiking neuroidal net simulator (Günay and Maida, 2001, 2006b), which was overviewed in Chapter 7. In that model, excess recruitment candidates were rejected to enforce a stable recruitment level. This chapter describes the improvements of the previous boost-and-limit method to use a more biologically realistic neural mechanism (Günay and Maida, 2006a). [ABSTRACT FROM AUTHOR]

In the brain, functionally and physically separate areas of cortex (e.g., visual, auditory, somatosensory) analyze different features of environmental objects. This raises the question of how physically distributed feature representations are combined to form coherent unitary percepts. This question of neural representation was first identified by Rosenblatt (1961) as the binding problem. A solution to this problem is temporal binding (von der Malsburg, 1981) that we showed in the previous chapter that it is consistent with recruitment learning when spiking neurons are employed. This solution has become known as the temporal correlation hypothesis (TCH), which posits that binding of disparate feature representations is achieved by synchronized neural firing across cortical areas (von der Malsburg, 1981, 1995; Gray et al, 1989; Singer, 1995; Singer and Gray, 1995; König and Engel, 1995; Ritz and Sejnowski, 1997; Engel et al, 1999; Treisman, 1996) based on what has more recently been termed the study of the functional connectivity of the brain (Baeg et al, 2007). The viability of TCH depends on maintaining the synchrony of neural spikes coding features from the same object and desynchrony of spikes coding features from different objects. [ABSTRACT FROM AUTHOR]

This book provides an overview of recruitment learning, a neural network learning system which exhibits high biological plausibility due to a number of core features - notably sparse connectivity levels and the use of weight updates which both respect synaptic constraints and support rapid concept acquisition. The earlier sections of this volume have given form to these ideas, for both the discrete time systems characteristic of the early wave of connectionism, and the more recent continuous time spiking neural systems - which allow closer alignment with neural observation. In this concluding discussion, it is worthwhile to summarise these ideas and to identify opportunities which remain unexplored, most notably those which arise through the emergence of adult neurogenesis. [ABSTRACT FROM AUTHOR]

Previous chapters of this book gave an overview of connectionist recruitment learning or the neuroidal network (Feldman, 1982, 1990; Valiant, 1988, 1994, 1998, 2000b, 2005) as a discrete-time algorithm. By extending recruitment into continuous-time to make it closer to the neuronal networks of the brain, this chapter puts it in the context of spiking neural networks (Hebb, 1949; Von der Malsburg, 1981, 1994; Abeles, 1991; Gerstner, 2001) and shows its similarity to temporal binding (Von der Malsburg, 1981; Singer and Gray, 1995). The chapter is organized as follows. First, classical neural network research is reviewed to describe the theory of spiking neural network models. Then, the development of concept representation in neural networks and the binding problem is reviewed to present the relationship between recruitment learning and the temporal binding hypothesis. A historical account of these developments is also given. [ABSTRACT FROM AUTHOR]

In the final chapter of Part I of this book, we examine the application of connectionist models of recruitment to problems at the core of cognitive science, the integration of visual and linguistic information and the acquisition of an elementary semantics of space and facial expressions. Most significantly, the tasks of this chapter transcend barriers of race and culture, forcing computational systems to confront deeper issues of representation and generalisation. In this respect they lie firmly within the framework of the L0 project of Feldman et al (1990), the ˵touchstone task for cognitive science″ devised in response to the excessively domain specific computational models of cognition prevalent at that time. [ABSTRACT FROM AUTHOR]

In the preceding chapter we were more concerned with the neural foundations of knowledge representation and the existence of structures capable of supporting non-trivial concepts than with actual recruitment of functional circuits, which so far has been limited to the case of one-shot learning (Chapter 2). This chapter makes explicit the linkages between structural plausibility, knowledge representation and recruitment, exploring the interactions between them and ultimately demonstrating the success of our approach when applied to a number of standard benchmark problems. Later in Part I of the book, in Chapter 5, we come somewhat closer to the biological context in which the ideas were developed, recruiting sparse random networks of this type to encode the spatial relations semantics of simple images. [ABSTRACT FROM AUTHOR]

We have earlier considered the foundations of recruitment learning, an approach grounded in the understanding of neural organisation and psychological function which emerged during the second half of the twentieth century. While the revolutionary findings in respect of adult neurogenesis explored in section 1.5 have deep implications for recruitment learning, we focus initially on Arbib΄s earlier contention that ˵... a certain critical degree of structural complexity is required of a network before it can become self-modifying ... in a way that we could consider intelligent″. This chapter begins a computational exploration of these ideas, adopting the strategy of converging constraints as pioneered by Feldman and Valiant to help delineate the tasks which might be realised through the recruitment of general purpose circuits. [ABSTRACT FROM AUTHOR]

Most work in connectionist learning, and indeed in the present volume, is devoted to induction from examples, but instruction is as important in symbolic artificial intelligence as it is in nature. This chapter begins our exploration of recruitment learning by examining a connectionist learning system capable of rapidly integrating a new fact into an existing knowledge base in response to a single instruction expressed in a formal description language. Such instructions are transformed into an input for a connectionist knowledge representation system, which in turn changes the network in order to integrate new knowledge. In subsequent chapters we shall consider models based explicitly on neurally plausible circuits, with operations defined at the level of the individual neuron or assemblies of neurons. Here the representation is more abstract, in principle replacing the neural assembly by a single node – demonstrating both the power of recruitment learning as a knowledge acquisition strategy, and the potential alignment between connectionist systems and the established agenda of symbolic artificial intelligence. [ABSTRACT FROM AUTHOR]

Computer science is the science of abstraction. Aho & Ullman (1992, p.3) call computer science the ˵mechanization of abstraction″, i.e. an appropriate model is sought for a given problem and is realized by use of a computational device. Abstraction often means simplification: a complex and real problem is replaced by a model that can solve a given problem (Aho & Ullman, 1992, p.3). [ABSTRACT FROM AUTHOR]

We present a novel, web-accessible scientific workflow system which makes large-scale comparative studies accessible without programming or excessive configuration requirements. GPFlow allows a workflow defined on single input values to be automatically lifted to operate over collections of input values and supports the formation and processing of collections of values without the need for explicit iteration constructs. We introduce a new model for collection processing based on key aggregation and slicing which guarantees processing integrity and facilitates automatic association of inputs, allowing scientific users to manage the combinatorial explosion of data values inherent in large scale comparative studies. The approach is demonstrated using a core task from comparative genomics, and builds upon our previous work in supporting combined interactive and batch operation, through a lightweight web-based user interface. [ABSTRACT FROM AUTHOR]

Artificial neural networks play an important role for pattern recognition tasks. However, due to poor comprehensibility of the learned network, and the inability to represent explanation structures, they are not considered sufficient for the general representation of knowledge. This paper details a methodology that represents the knowledge of a trained network in the form of restricted first-order logic rules, and subsequently allows user interaction by interfacing with a knowledge based reasoner. [ABSTRACT FROM AUTHOR]

Consider a network of unreliable links, each of which comes with a certain price and reliability. Given a fixed budget, which links should be bought in order to maximize the system's reliability? We introduce a Cross-Entropy approach to this problem, which can deal effectively with the noise and constraints in this difficult combinatorial optimization problem. Numerical results demonstrate the effectiveness of the proposed technique. [ABSTRACT FROM AUTHOR]

Racing algorithms have recently been proposed as a general-purpose method for performing model selection in machine learning algorithms. In this paper, we present an empirical study of the Hoeffding racing algorithm for selecting the k parameter in a simple k-nearest neighbor classifier. Fifteen widely-used classification datasets from UCI are used and experiments conducted across different confidence levels for racing. The results reveal a significant amount of sensitivity of thek -nn classifier to its model parameter value. The Hoeffding racing algorithm also varies widely in its performance, in terms of the computational savings gained over an exhaustive evaluation. While in some cases the savings gained are quite small, the racing algorithm proved to be highly robust to the possibility of erroneously eliminating the optimal models. All results were strongly dependent on the datasets used. [ABSTRACT FROM AUTHOR]

Foreign Exchange trading has emerged in recent times as a significant activity in many countries. As with most forms of trading, the activity is influenced by many random parameters so that the creation of a system that effectively emulates the trading process will be very helpful. In this paper we try to create such a system using Machine learning approach to emulate trader behaviour on the Foreign Exchange market and to find the most profitable trading strategy. [ABSTRACT FROM AUTHOR]

This paper investigates the cross-market informational dependence between these assets under disparate interest rate conditions of the U.S and Australia. With conditional variance as a proxy for volatility, we use the BEKK - a matricular decomposition of the bivariate GARCH (1,1) model to examine the cross-market contemporaneous effect of information arrival. Applying the model to the stock and bond indices of both countries, we find evidence of volatility spillover, thereby supporting the notion of informational dependence between each market. Keywords: Comovement, volatility transmission, conditional varaince, GARCH (1,1) JEL Classification:G11, G12 [ABSTRACT FROM AUTHOR]

In this paper, we propose a self Adaptive Migration Model for Genetic Algorithms, where parameters of population size, the number of points of crossover and mutation rate for each population are fixed adaptively. Further, the migration of individuals between populations is decided dynamically. This paper gives a mathematical schema analysis of the method stating and showing that the algorithm exploits previously discovered knowledge for a more focused and concentrated search of heuristically high yielding regions while simultaneously performing a highly explorative search on the other regions of the search space. The effective performance of the algorithm is then shown using standard testbed functions, when compared with Island model GA(IGA) and Simple GA(SGA). [ABSTRACT FROM AUTHOR]

Volatility modelling of asset returns is an important aspect for many financial applications, e.g., option pricing and risk management. GARCH models are usually used to model the volatility processes of financial time series. However, multivariate GARCH modelling of volatilities is still a challenge due to the complexity of parameters estimation. To solve this problem, we suggest using Independent Component Analysis (ICA) for transforming the multivariate time series into statistically independent time series. Then, we propose the ICA-GARCH model which is computationally efficient to estimate the volatilities. The experimental results show that this method is more effective to model multivariate time series than existing methods, e.g., PCA-GARCH. Keywords: Financial Engineering, GARCH, ICA, Multivariate Time Series, Volatility [ABSTRACT FROM AUTHOR]

This paper presents a new multicriteria sorting procedure in financial classification problems, based on the methodological framework of PROMETHEE method. The proposed procedure, called as PROMSORT, is applied to the business failure risk problem and compared to PROMETHEE TRI and ELECTRE TRI. The proposed methodology also identifies the differences in performances across risk groups, and assists in monitoring the firms' financial performances. The results showed that the proposed procedure can be considered as an effective alternative to existing methods in financial classification problems. [ABSTRACT FROM AUTHOR]

A novel Genetic Programming (GP) paradigm called Co-evolutionary Rule-Chaining Genetic Programming (CRGP) has been proposed to learn the relationships among attributes represented by a set of classification rules for multi-class problems. It employs backward chaining inference to carry out classification based on the acquired acyclic rule set. Its main advantages are: 1) it can handle more than one class at a time; 2) it avoids cyclic result; 3) unlike Bayesian Network (BN), the CRGP can handle input attributes with continuous values directly; and 4) with the flexibility of GP, CRGP can learn complex relationship. We have demonstrated its better performance on one synthetic and one real-life medical data sets. [ABSTRACT FROM AUTHOR]

The design, development, and use of complex systems models raises a unique class of challenges and potential pitfalls, many of which are commonly recurring problems. Over time, researchers gain experience in this form of modeling, choosing algorithms, techniques, and frameworks that improve the quality, confidence level, and speed of development of their models. This increasing collective experience of complex systems modellers is a resource that should be captured. Fields such as software engineering and architecture have benefited from the development of generic solutions to recurring problems, called patterns. Using pattern development techniques from these fields, insights from communities such as learning and information processing, data mining, bioinformatics, and agent-based modeling can be identified and captured. Collections of such ‘pattern languages' would allow knowledge gained through experience to be readily accessible to less-experienced practitioners and to other domains. This paper proposes a methodology for capturing the wisdom of computational modelers by introducing example visualization patterns, and a pattern classification system for analyzing the relationship between micro and macro behaviour in complex systems models. We anticipate that a new field of complex systems patterns will provide an invaluable resource for both practicing and future generations of modelers. [ABSTRACT FROM AUTHOR]

In this paper, a new crossover operator based on Latin square design is presented at first. This crossover operator can generate a set of uniformly scattered offspring around their parents, and it is of the ability of local search and thus can explore the search space efficiently. Then the level set of the objective function is evolved successively by crossover and mutation operators such that it gradually approaches to global optimal solution set. Based on these, a new evolutionary algorithm for nondifferentiable unconstrained global optimization is proposed and its global convergence is proved. At last, the numerical simulations are made for some standard test functions. The performance of the proposed algorithm is compared with that of two widely-cited algorithms. The results indicate the proposed algorithm is effective and has better performance than the compared algorithms for these test functions. [ABSTRACT FROM AUTHOR]

In many data mining projects the data to be analysed contains personal information, like names and addresses. Cleaning and pre-processing of such data likely involves deduplication or linkage with other data, which is often challenged by a lack of unique entity identifiers. In recent years there has been an increased research effort in data linkage and deduplication, mainly in the machine learning and database communities. Publicly available test data with known deduplication or linkage status is needed so that new linkage algorithms and techniques can be tested, evaluated and compared. However, publication of data containing personal information is normally impossible due to privacy and confidentiality issues. An alternative is to use artificially created data, which has the advantages that content and error rates can be controlled, and the deduplication or linkage status is known. Controlled experiments can be performed and replicated easily. In this paper we present a freely available data set generator capable of creating data sets containing names, addresses and other personal information. [ABSTRACT FROM AUTHOR]

Agents make up an important part of game worlds, ranging from the characters and monsters that live in the world to the armies that the player controls. Despite their importance, agents in current games rarely display an awareness of their environment or react appropriately, which severely detracts from the believability of the game. Some games have included agents with a basic awareness of other agents, but they are still unaware of important game events or environmental conditions. This paper presents an agent design we have developed, which combines cellular automata for environmental modeling with influence maps for agent decision-making. The agents were implemented into a 3D game environment we have developed, the EmerGEnT system, and tuned through three experiments. The result is simple, flexible game agents that are able to respond to natural phenomena (e.g. rain or fire), while pursuing a goal. [ABSTRACT FROM AUTHOR]

There is increasing demand for the self-diagnosis and self-healing of problems or errors arising in systems operating in the ubiquitous computing environment. In this paper, we propose a self-healing system that monitors, diagnoses and heals its own problems. The proposed system consists of multi agents that analyze the log context in order to perform self-diagnosis and self-healing. To minimize the resources used by the Adapters in an existing system, we place a single process in memory. By this, we mean that a single Monitoring Agent monitors the context of the logs that are generated by the different components of the system. For rapid and efficient self-healing, we use a 6-step process. The effectiveness of the proposed system is confirmed through experiments conducted with a prototype system. [ABSTRACT FROM AUTHOR]

Coalition formation in multi-agent systems (MAS) is becoming increasingly important as it increases the ability of agents to execute tasks and maximize their payoffs. This paper proposes a novel dependence theory namely transitive dependence theory for dynamic coalition formation in multi-agent system. Based on the proposed transitive dependence theory, a reasoning mechanism for searching coalition partners has been worked out which includes dependence tree generation, dependence tree reduction, plan optimization and action optimization. [ABSTRACT FROM AUTHOR]

Resources or devices on the network might be unavailable due to serious network partitioning. To provide a robust network connectivity, this paper presents an autonomous and dependable recovery scheme using teamwork in UPnP network settings. For our scheme, we introduce a team of recoverable control points and corresponding recovery device, and, in case of network failures among devices, the recoverable control points autonomously take care of the devices to achieve their mutual goal as a team member. In the experiments, we tested our recovery scheme in terms of the recovery effectiveness. It turned out that the recoverable control point agent in our scheme successfully handled the events from the devices, regardless of one of recoverable control points being killed. We argue that our recovery scheme is fairly consistent, and the control point agents can recover from failures as quickly as possible. [ABSTRACT FROM AUTHOR]

Conventional adaptive systems have common well-known constraints when attempting to normalize environment. An adaptive system must contain a certain number of rules allowing such a system to adapt to specific situations. If there is an absence of a rule in a new situation, the system cannot take appropriate action. Building and managing such complex static adaptive systems places an enormous burden on system developers. In this paper, we propose a multi-agent based intelligent adaptive system with a self-growing engine. In this system, the inference agent evaluates input context with specific factors and analyzes the results. The decision agent selects the most appropriate action among alternatives available for a specific context and intelligently evolves and adapts by means of a self-growing engine (SGE). The SGE can evaluate actions and generate new rules by applying it to a practical situation using remote video conferencing with mobile devices such as PDAs, and PCs. [ABSTRACT FROM AUTHOR]

This research investigates transitive dependence relations, an extension of direct dependence relations, in multi-agent systems. In this paper, action dependence relations are employed to deduct transitive dependence relations from direct dependence relations. Transitive dependence is useful in representation, analysis, and social relations reasoning between agents, groups, organizations, etc. Furthermore, in this paper, dependence relations are differentiated by both dependence property and dependence degree, which is useful in quantitative social reasoning. [ABSTRACT FROM AUTHOR]

This paper highlights the Hybrid agent construction model being developed that allows the description and development of autonomous agents in SAGE (Scalable, fault Tolerant Agent Grooming Environment) - a second generation FIPA-Compliant Multi-Agent system. We aim to provide the programmer with a generic and well defined agent architecture enabling the development of sophisticated agents on SAGE, possessing the desired properties of autonomous agents - reactivity, pro-activity, social ability and knowledge based reasoning. Keywords: Autonomous agents, MAS, Intelligent Agent, Agent Architecture, Hybrid Agent [ABSTRACT FROM AUTHOR]

Recently, there is a massive use of the agent technology to develop the intelligent software and smart control module. In this paper, we propose a support tool for Multi-Agent Development, which enables the overall development time to be reduced. The proposed tool provides the general architecture constructed by components, interfaces, and functions of an agent an It also provides the extended Message Sequence Diagram for the communication of Multi-Agent. With the aid of proposed tool, developers who are inexperience in agent development can design and implement Multi-Agent rapidly. [ABSTRACT FROM AUTHOR]

In this paper, a new fold recognition model with mixed environment-specific substitution mapping (called MESSM) is proposed with three key features: 1) a structurally-derived substitution score is generated using neural networks; 2) a mixed environment-specific substitution mapping is developed by combing the structural-derived substitution score with sequence profile from well-developed sequence substitution matrices; 3) a support vector machine is employed to measure the significance of the sequence-structure alignment. Tested on two benchmark problems, the MESSM model shows comparable performance to those more computational intensive, energy potential based fold recognition models. The results also demonstrate that the new fold recognition model with mixed substitution mapping has a better performance than the one with either structure or sequence profile only. The MESSM model presents a new way to develop an efficient tool for protein fold recognition. [ABSTRACT FROM AUTHOR]

Prediction of peroxisomal matrix proteins generally depends on the presence of one of two distinct motifs at the end of the amino acid sequence. PTS1 peroxisomal proteins have a well conserved tripeptide at the C-terminal end. However, the preceding residues in the sequence arguably play a crucial role in targeting the protein to the peroxisome. Previous work in applying machine learning to the prediction of peroxisomal matrix proteins has failed to capitalize on the full extent of these dependencies. We benchmark a range of machine learning algorithms, and show that a classifier - based on the Support Vector Machine - produces more accurate results when dependencies between the conserved motif and the preceding section are exploited. We publish an updated and rigorously curated data set that results in increased prediction accuracy of most tested models. [ABSTRACT FROM AUTHOR]

Identifying bacterial promoters is the key to understanding gene expression. Promoters lie in tightly constrained positions relative to the transcription start site (TSS). Knowing the TSS position, one can predict promoter positions to within a few base pairs, and vice versa. As a route to promoter identification, we formally address the problem of TSS prediction, drawing on the RegulonDB database of known (mapped) Escherichia coli TSS locations. The accepted method of finding promoters (and therefore TSSs) is to use position weight matrices (PWMs). We use an alternative approach based on support vector machines (SVMs). In particular, we quantify performance of several SVM models versus a PWM approach, using area under the detection-error tradeoff (DET) curve as a performance metric. SVM models are shown to outperform the PWM at TSS prediction, and to substantially reduce numbers of false positives, which are the bane of this problem. [ABSTRACT FROM AUTHOR]

Finding motifs that can elucidate rules that govern peptide binding to medically important receptors is important for screening targets for drugs and vaccines. This paper focuses on elucidation of peptide binding to I-Ag7 molecule of the non-obese diabetic (NOD) mouse - an animal model for insulin-dependent diabetes mellitus (IDDM). A number of proposed motifs that describe peptide binding to I-Ag7 have been proposed. These motifs results from independent experimental studies carried out on small data sets. Testing with multiple data sets showed that each of the motifs at best describes only a subset of the solution space, and these motifs therefore lack generalization ability. This study focuses on seeking a motif with higher generalization ability so that it can predict binders in all Ag7data sets with high accuracy. A binding score matrix representing peptide binding motif to Ag7was derived using genetic algorithm (GA). The evolved score matrix significantly outperformed previously reported motifs. [ABSTRACT FROM AUTHOR]

Due to the large number of genes measured in a typical microarray dataset, feature selection plays an essential role in tumor classification. In turn, relevance and redundancy are key components in determining the optimal predictor set. However, a third component - the relative weights given to the first two also assumes an equal, if not greater importance in feature selection. Based on this third component, we developed two novel feature selection methods capable of producing high, unbiased classification accuracy in multiclass microarray dataset. In an in-depth analysis comparing the two methods, the optimal values of the relative weights are also estimated. Keywords: Feature selection, microarray, tumor classification, redundancy [ABSTRACT FROM AUTHOR]

An important and common problem in microarray experiments is the detection of genes that are differentially expressed in a given number of classes. As this problem concerns the selection of significant genes from a large pool of candidate genes, it needs to be carried out within the framework of multiple hypothesis testing. In this paper, we focus on the use of mixture models to handle the multiplicity issue. With this approach, a measure of the local FDR (false discovery rate) is provided for each gene. An attractive feature of the mixture model approach is that it provides a framework for the estimation of the prior probability that a gene is not differentially expressed, and this probability can subsequently be used in forming a decision rule. The rule can also be formed to take the false negative rate into account. We apply this approach to a well-known publicly available data set on breast cancer, and discuss our findings with reference to other approaches. [ABSTRACT FROM AUTHOR]

Normally the microarray data contain a large number of genes (usually more than 1000) and a relatively small number of samples (usually fewer than 100). This makes the discriminant analysis of DNA microarray data hard to handle. Selecting important genes to the discriminant problem is hence of much practically significance in microarray data analysis. If put in the context of pattern classification, gene selection can be casted as a feature selection problem. Feature selection approaches are broadly grouped into filter and wrapper methods. The wrapper method outperforms the filter method in general. However the accuracy of wrapper methods is coupled with intensive computations. In present study, we proposed a wrapper-based gene selection algorithm by employing the Regularization Network as the classifier. Compared with classical wrapper method, the computational costs in our gene selection algorithm is significantly reduced, because the evaluation criterion we used does not demand repeated trainings in the leave-one-out procedure. [ABSTRACT FROM AUTHOR]

This paper proposes a multi-population χ2 test method for informative gene selection of a tumor from microarray data based on the statistical multi-population χ2 test with the sample data being grouped evenly. To test the effectiveness of the multi-population χ2 test method, we use the support vector machine (SVM) to construct a tumor diagnosis system (i.e., a binary classifier) based on the identified informative genes on the colon and leukemia data. It is shown by the experiments that the constructed diagnosis system with the multi-population χ2 test method can 100% correctness rate of diagnosis on colon dataset and 97.1% correctness rate of diagnosis on leukemia dataset, respectively. [ABSTRACT FROM AUTHOR]

Biological research requires information from multiple data sources that use a variety of database-specific formats. Manual gathering of information is time consuming and error-prone, making automated data aggregation a compelling option for large studies. We describe a method for extracting information from diverse sources that involves structural rules specified by example. We developed a system for aggregation of biological knowledge (ABK) and used it to conduct an epidemiological study of dengue virus (DENV) sequences. Additional information on geographical origin and isolation date is critical for understanding evolutionary relationships, but this data is inconsistently structured in database entries. Using three public databases, we found that structural rules can be used successfully even when applied on inconsistently structured data that is distributed across multiple fields. High reusability, combined with the ability to integrate analysis tools, make this method suitable for a wide variety of large-scale studies involving viral sequences. [ABSTRACT FROM AUTHOR]