Catálogo de publicaciones - libros

The focus of this paper is on the practical aspects of efficiently resolving inconsistencies when merging probabilistic rule sets. We consider the problem of prioritized merging, when one core knowledge base is to be used without modifications and to be extended by information from other sources. This problem is addressed by our flexible system that aims at restoring consistency by finding those parts of the additional rule bases which are compatible with the core base and are considered most valuable by the user. We give an overview on the methodological framework of the system and describe some details of its main techniques. In particular, offers a convenient interface to inductive probabilistic reasoning on maximum entropy. An example from the domain of auditing illustrates the problem and the practical applicability of our framework.

Markov logic, as a highly expressive representation formalism that essentially combines the semantics of probabilistic graphical models with the full power of first-order logic, is one of the most intriguing representations in the field of probabilistic logical modelling. However, as we will show, models in Markov logic often fail to generalize because the parameters they contain are highly domain-specific. We take the perspective of generative stochastic processes in order to describe probability distributions in relational domains and illustrate the problem in this context by means of simple examples.

We propose an extension of the language that involves the specification of a priori independent attributes and that furthermore introduces a dynamic parameter adjustment whenever a model in Markov logic is instantiated for a certain domain (set of objects). Our extension removes the corresponding restrictions on processes for which models can be learned using standard methods and thus enables Markov logic networks to be practically applied to a far greater class of generative stochastic processes.

This work presents , a system that searches for answers to definition questions in several languages on web snippets. For this purpose, biases the search engine in favour of some syntactic structures that often convey definitions. Once descriptive sentences are identified, clusters them by and presents the most relevant phrases of each to the user. The approach was assessed with TREC and CLEF data. As a result, was able to extract descriptive information for all definition questions in the TREC 2001 and 2003 data-sets.

SPARQL is the upcoming W3C standard query language for RDF data in the semantic web. In this paper we propose a formal semantics for SPARQL based on datalog. A mapping of SPARQL to datalog allows to easily reuse existing results from logics for analysis and extensions of SPARQL. Using this semantics we analyse the complexity of query answering in SPAQRL and propose two useful extensions to SPARQL, namely binding of variables to results of filter expressions and views on RDF graphs as datasets for queries. We show that these extensions to not add to the overall complexity of SPARQL.

In recent years a lot of research has been done in order to determine factors of complexity in spatial relational reasoning, like the number of models, the wording of conclusion or the influence of relational complexity. But research so far focused on affirmative statements only, i. e. negated expressions have not yet been investigated. In spatial reasoning and in human machine interaction, however, negation plays a fundamental role. Central questions are: How are negated statements represented? What happens in multiple-model cases? Which effects have different reference frames? We conducted three experiments to show that humans (i) negate a relation by using the opposite relation, (ii) construct preferred mental models and use an economic principle, and (iii) have more difficulties in reasoning with negated relations. The goal is to extend our cognitive and computational model – the SRM.

We introduce relational variants of neural gas, a very efficient and powerful neural clustering algorithm, which allow a clustering and mining of data given in terms of a pairwise similarity or dissimilarity matrix. It is assumed that this matrix stems from Euclidean distance or dot product, respectively, however, the underlying embedding of points is unknown. One can equivalently formulate batch optimization in terms of the given similarities or dissimilarities, thus providing a way to transfer batch optimization to relational data. For this procedure, convergence is guaranteed and extensions such as the integration of label information can readily be transferred to this framework.

In this paper we present a novel general framework for encoding and evolving networks called Common Genetic Encoding (CGE) that can be applied to both direct and indirect encoding methods. The encoding has important properties that makes it suitable for evolving neural networks: (1) It is in that it is able to represent all types of valid phenotype networks. (2) It is , i. e. every valid genotype represents a valid phenotype. Similarly, the encoding is such as structural mutation and crossover that act upon the genotype. Moreover, the encoding’s genotype can be seen as a composition of several subgenomes, which makes it to inherently support the evolution of modular networks in both direct and indirect encoding cases. To demonstrate our encoding, we present an experiment where direct encoding is used to learn the dynamic model of a two-link arm robot. We also provide an illustration of how the indirect-encoding features of CGE can be used in the area of artificial embryogeny.

In this paper, we show how reinforcement learning can be applied to real robots to achieve optimal robot behavior. As example, we enable an autonomous soccer robot to learn intercepting a rolling ball. Main focus is on how to adapt the Q-learning algorithm to the needs of learning strategies for real robots and how to transfer strategies learned in simulation onto real robots.

Recently, the aspect of visual perception has been explored in the context of Gibson’s concept of affordances [1] in various ways. We focus in this work on the importance of developmental learning and the perceptual cueing for an agent’s anticipation of opportunities for interaction, in extension to functional views on visual feature representations. The concept for the incremental learning of abstract from basic affordances is presented in relation to learning of complex affordance features. In addition, the work proposes that the originally defined representational concept for the perception of affordances - in terms of using either motion or 3D cues - should be generalized towards using arbitrary visual feature representations. We demonstrate the learning of causal relations between visual cues and associated anticipated interactions by reinforcement learning of predictive perceptual states. We pursue a recently presented framework for cueing and recognition of affordance-based visual entities that obviously plays an important role in robot control architectures, in analogy to human perception. We experimentally verify the concept within a real world robot scenario by learning predictive visual cues using reinforcement signals, proving that features were selected for their relevance in predicting opportunities for interaction.

Simulation results of bistable perception due to ambiguous visual stimuli are presented which are obtained with a nonlinear dynamics model using perception–attention–memory coupling. Percept reversals are induced by attention fatigue and noise, with an attention bias which balances the relative percept duration. The dynamics of the attention parameter exhibits qualitative agreement with the eye blink rate variation [4]. Coupling of an attention bias to the perception state introduces memory effects leading to significant long range correlations of perceptual duration times as quantified by the Hurst parameter (H > 0.5). This prediction is in agreement with recent experimental results [1]. Deviations of the reversal time statistics from the -distribution increase with decreasing memory time constant and attention noise. Mean perceptual duration times of 2 – 5 s are predicted in agreement with experimental results [7] if a feedback delay of ca. 40 ms is assumed which is typical for cortical reentrant loops.