Nicola Di Mauro

Avoiding order effects in incremental learning

This paper addresses the problem of mitigating the order effects in incremental learning, a phenomenon observed when different ordered sequences of observations lead to different results. A modification of an ILP incremental learning system, with the aim of making it order-independent, is presented. A backtracking strategy on theories is incorporated in its refinement operators, which causes a change of its refinement strategy and reflects the human behavior during the learning process. A modality to restore a previous theory, in order to backtrack on a previous knowledge level, is presented. Experiments validate the approach in terms of computational cost and predictive accuracy.

Incremental multistrategy learning for document processing

This work presents the application of a multistrategy approach to some document processing tasks. The application is implemented in an enhanced version of the incremental learning system INTHELEX. This learning module has been embedded as a learning component in the system architecture of the EU project COLLATE, which deals with the annotation of cultural heritage documents. Indeed, the complex shape of the material handled in the project has suggested that the addition of multistrategy capabilities is needed to improve effectiveness and efficiency of the learning process. Results proving the benefits of these strategies in specific classfication tasks are reported in the experimentation presented in this work.

Machine Learning for Digital Document Processing: from Layout Analysis to Metadata Extraction

In the last years, the spread of computers and the Internet caused a significant amount of documents to be available in digital format. Collecting them in digital repositories raised problems that go beyond simple acquisition issues, and cause the need to organize and classify them in order to improve the effectiveness and efficiency of the retrieval procedure. The success of such a process is tightly related to the ability of understanding the semantics of the document components and content. Since the obvious solution of manually creating and maintaining an updated index is clearly infeasible, due to the huge amount of data under consideration, there is a strong interest in methods that can provide solutions for automatically acquiring such a knowledge. This work presents a framework that intensively exploits intelligent techniques to support different tasks of automatic document processing from acquisition to indexing, from categorization to storing and retrieval. The prototypical version of the system DOMINUS is presented, whose main characteristic is the use of a Machine Learning Server, a suite of different inductive learning methods and systems, among which the more suitable for each specific document processing phase is chosen and applied. The core system is the incremental first-order logic learner INTHELEX. Thanks to incrementality, it can continuously update and refine the learned theories, dynamically extending its knowledge to handle even completely new classes of documents. Since DOMINUS is general and flexible, it can be embedded as a document management engine into many different Digital Library systems. Experiments in a real-world domain scenario, scientific conference management, confirmed the good performance of the proposed prototype.

GRAPE: an expert review assignment component for scientific conference management systems

This paper describes GRAPE, an expert component for a scientific Conference Management System (CMS), to automatically assign reviewers to papers, one of the most difficult processes of conference management. In the current practice, this is typically done by a manual and time-consuming procedure, with a risk of bad quality results due to the many aspects and parameters to be taken into account, and on their interrelationships and (often contrasting) requirements. The proposed rule-based system was evaluated on real conference datasets obtaining good results when compared to the handmade ones, both in terms of quality of the assignments, and of reduction in execution time.

Applying the information bottleneck to statistical relational learning

In this paper we propose to apply the Information Bottleneck (IB) approach to the sub-class of Statistical Relational Learning (SRL) languages that are reducible to Bayesian networks. When the resulting networks involve hidden variables, learning these languages requires the use of techniques for learning from incomplete data such as the Expectation Maximization (EM) algorithm. Recently, the IB approach was shown to be able to avoid some of the local maxima in which EM can get trapped when learning with hidden variables. Here we present the algorithm Relational Information Bottleneck (RIB) that learns the parameters of SRL languages reducible to Bayesian Networks. In particular, we present the specialization of RIB to a language belonging to the family of languages based on the distribution semantics, Logic Programs with Annotated Disjunction (LPADs). This language is prototypical for such a family and its equivalent Bayesian networks contain hidden variables. RIB is evaluated on the IMDB, Cora and artificial datasets and compared with LeProbLog, EM, Alchemy and PRISM. The experimental results show that RIB has good performances especially when some logical atoms are unobserved. Moreover, it is particularly suitable when learning from interpretations that share the same Herbrand base.

Machine learning approaches for inducing student models

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.

Coalition structure generation with GRASP

The coalition structure generation problem represents an active research area in multi-agent systems. A coalition structure is defined as a partition of the agents involved in a system into disjoint coalitions. The problem of finding the optimal coalition structure is NP-complete. In order to find the optimal solution in a combinatorial optimization problem it is theoretically possible to enumerate the solutions and evaluate each. But this approach is infeasible since the number of solutions often grows exponentially with the size of the problem. In this paper we present a greedy adaptive search procedure (GRASP) to efficiently search the space of coalition structures in order to find an optimal one.

An Exhaustive Matching Procedure for the Improvement of Learning Efficiency

Efficiency of the first-order logic proof procedure is a major issue when deduction systems are to be used in real environments, both on their own and as a component of larger systems (e.g., learning systems). Hence, the need of techniques that can speed up such a process. This paper proposes a new algorithm for matching first-order logic descriptions under θ-subsumption that is able to return the set of all substitutions by which such a relation holds between two clauses, and shows experimental results in support of its performance.

Relational Temporal Data Mining for Wireless Sensor Networks

Wireless sensor networks (WSNs) represent a typical domain where there are complex temporal sequences of events. In this paper we propose a relational framework to model and analyse the data observed by sensor nodes of a wireless sensor network. In particular, we extend a general purpose relational sequence mining algorithm to take into account temporal interval-based relations. Real-valued time series are discretized into similar subsequences and described by using a relational language. Preliminary experimental results prove the applicability of the relational learning framework to complex real world temporal data.

Probabilistic Inference over Image Networks

Digital Libraries contain collections of multimedia objects providing services for the management, sharing and retrieval. Involved objects have two levels of complexity: the former refers to the inner object complexity while the latter takes into account the implicit/explicit relationships among objects. Traditional machine learning classifiers do not consider the relationships among objects assuming them independent and identically distributed. Recently, link-based classification methods have been proposed, that try to classify objects exploiting their relationships (links). In this paper, we deal with objects corresponding to digital images, even if the proposed approach can be naturally applied to different kind of multimedia objects. Relationships can be expressed among the features of the same image or among features belonging to different images. The aim of this work is to verify whether a link-based classifier based on a Statistical Relational Learning (SRL) language can improve the accuracy of a classical k-nearest neighbour approach. Experiments will show that the modelling of the relationships in a real-word dataset using a SRL model reduces the classification error.