Catálogo de publicaciones - libros

Although computational intelligence (CI) covers a vast variety of different methods it still lacks an integrative theory. Several proposals for CI foundations are discussed: computing and cognition as compression, meta-learning as search in the space of data models, (dis)similarity based methods providing a framework for such meta-learning, and a more general approach based on chains of transformations. Many useful transformations that extract information from features are discussed. Heterogeneous adaptive systems are presented as particular example of transformation-based systems, and the goal of learning is redefined to facilitate creation of simpler data models. The need to understand data structures leads to techniques for logical and prototype-based rule extraction, and to generation of multiple alternative models, while the need to increase predictive power of adaptive models leads to committees of competent models. Learning from partial observations is a natural extension towards reasoning based on perceptions, and an approach to intuitive solving of such problems is presented. Throughout the paper neurocognitive inspirations are frequently used and are especially important in modeling of the higher cognitive functions. Promising directions such as liquid and laminar computing are identified and many open problems presented.

Clustering is commonly regarded as a synonym of unsupervised learning aimed at the discovery of structure in highly dimensional data. With the evident plethora of existing algorithms, the area offers an outstanding diversity of possible approaches along with their underlying features and potential applications. With the inclusion of fuzzy sets, fuzzy clustering became an integral component of Computational Intelligence (CI) and is now broadly exploited in fuzzy modeling, fuzzy control, pattern recognition, and exploratory data analysis. A lot of pursuits of CI are human-centric in the sense they are either initiated or driven by some domain knowledge or the results generated by the CI constructs are made easily interpretable. In this sense, to follow the tendency of human-centricity so profoundly visible in the CI domain, the very concept of fuzzy clustering needs to be carefully revisited. We propose a certain paradigm shift that brings us to the idea of knowledge -based clustering in which the development of information granules – fuzzy sets is governed by the use of data as well as domain knowledge supplied through an interaction with the developers, users and experts. In this study, we elaborate on the concepts and algorithms of knowledge-based clustering by considering the well known scheme of Fuzzy C-Means (FCM) and viewing it as an operational model using which a number of essential developments could be easily explained. The fundamental concepts discussed here involve clustering with domain knowledge articulated through partial supervision and proximity-based knowledge hints.

Capability of generalization in learning from examples can be modeled using regularization, which has been developed as a tool for improving stability of solutions of inverse problems. Theory of inverse problems has been developed to solve various tasks in applied science such as acoustics, geophysics and computerized tomography. Such problems are typically described by integral operators. It is shown that learning from examples can be reformulated as an inverse problem defined by an evaluation operator. This reformulation allows one to characterize optimal solutions of learning tasks and design learning algorithms based on numerical solutions of systems of linear equations.

In this chapter, advances on regularization and model selection in statistical learning have been summarized, and a trend has been discussed from a Bayesian Ying Yang learning perspective. After briefly introducing Bayesian Ying- Yang system and best harmony learning, not only its advantages of automatic model selection and of integrating regularization and model selection have been addressed, but also its differences and relations to several existing typical learning methods have been discussed and elaborated. Taking the tasks of Gaussian mixture, local subspaces, local factor analysis as examples, not only detailed model selection criteria are given, but also a general learning procedure is provided, which unifies those automatic model selection featured adaptive algorithms for these tasks. Finally, a trend of studies on model selection (i.e., automatic model selection during parametric learning), has been further elaborated. Moreover, several theoretical issues in a large sample size and a number of challenges in a small sample size have been presented.

The chapter considers recent achievements and perspectives of Computational Intelligence (CI) applied to mind games. Several notable examples of unguided, autonomous CI learning systems are presented and discussed. Based on advantages and limitations of existing approaches a list of challenging issues and open problems in the area of intelligent game playing is proposed and motivated. It is generally concluded in the paper that the ultimate goal of CI in mind game research is the ability to mimic human approach to game playing in all its major aspects including learning methods (learning from scratch, multitask learning, unsupervised learning, pattern-based knowledge acquisition) as well as reasoning and decision making (efficient position estimation, abstraction and generalization of game features, autonomous development of evaluation functions, effective preordering of moves and selective, contextual search).

The oriental game of Go is among the most tantalizing unconquered challenges in artificial intelligence after IBM's DEEP BLUE beat the world Chess champion in 1997. Its high branching factor prevents the conventional tree search approach, and long-range spatiotemporal interactions make position evaluation extremely difficult. Thus, Go attracts researchers from diverse fields who are attempting to understand how computers can represent human playing and win the game against humans. Numerous publications already exist on this topic with different motivations and a variety of application contexts. This chapter surveys methods and some related works used in computer Go published from 1970 until now, and offers a basic overview for future study. We also present our attempts and simulation results in building a non-knowledge game engine, using a novel hybrid evolutionary computation algorithm, for the Capture Go game.

In this Chapter, we review the virtues and limitations of the Hopfield neural network for tackling NP-hard combinatorial optimization problems (COPs). Then we discuss two new neural network models based on the noisy chaotic neural network, and applied the two methods to solving two different NP-hard COPs in communication networks. The simulation results show that our methods are superior to previous methods in solution quality. We also point out several future challenges and possible directions in this domain.