A review of the hybrid paradigm in Artificial Intelligence based on the conceptualization, design and development of new intelligent systems with several inference algorithms: applications and usefulness for decision support in the Engineering field
DATE:
2023-10-03
UNIVERSAL IDENTIFIER: http://hdl.handle.net/11093/5219
UNESCO SUBJECT: 1209.04 Teoría y Proceso de decisión ; 1203.04 Inteligencia Artificial ; 3310 Tecnología Industrial
DOCUMENT TYPE: doctoralThesis
ABSTRACT
Engineering is a multidisciplinary, adaptive, and dynamic discipline that could be characterised by endeavours in which ongoing decision-making processes take place. Each of these decisions is framed within a changing environment, in the permanent presence of uncertainty, and the success or failure of the actions taken brings together unique difficulties that affect the whole project. It is therefore essential to provide stakeholders with appropriate tools to support them in their search for optimal choices, so that these can be based mostly on objective aspects. Decision Support Systems have emerged in response to this need, and from simple models that compare and select options based on certain criteria to the current intelligent models, they all share the same objective: to make the best and most rational decisions that is possible. This objective, as simple to understand as it is ambitious in its purpose, constitutes the theoretical core from which this doctoral dissertation has emerged. It will explore, through a series of papers, the evolution from the common Decision Support Systems to those that now incorporate the concept of Artificial Intelligence in their definition and implementation. The theoretical and practical understanding of these Intelligent Decision Support Systems will be the common thread that will unite the different proposals that the progressive development of Artificial Intelligence has defined. Therefore, this work will start with an overview of models based on knowledge representation. These models aim to capture the knowledge of a decision domain by representing it in a formal and diversi&ied way. To do this, they use ontologies and logical languages that allow the definition of ordered, structured, and permanent knowledge bases on which reasoning can be carried out using symbolic inference structures. On the other hand, models based on Machine Learning propose, as an alternative to knowledge, the identification of relationships between data, so that, after a process of statistical inference, they work to find hypotheses in different models that would allow, in a general way, to relate these data to a set of real outcomes, their labels. Both models, with a common objective but different conceptualisation, are in fact attempts to characterise reasoning based on the duality between knowledge and data, that is, between the permanent and the ephemeral. While knowledge representation is a solid, transferable, and robust decision principle, statistical learning is a predictive tool that is agile, versatile, and adaptable to domains where knowledge is difficult to express. However, despite these differences, both models have inescapable advantages, and it is in this sense that the concept of hybrid intelligent systems has emerged, as systems that intend to combine models based on knowledge and models based on statistical learning. Their proposals are diverse, and cover aspects such as the automatic generation of procedural inference rules or the formalisation of datasets with semantic potential in a symbolic language. This doctoral dissertation, designed as a compendium of publications, collects, groups, and organises a set of 11 scientific papers published in high-impact journals, internationally recognised, and indexed in the Journal Citation Reports (JCR). These papers explore the design and development of Intelligent Decision Support Systems through the diverse integration of different inference processes, in a sequential, concurrent or hybrid way, forming a relevant and categorising corpus of the incorporation of Artificial Intelligence in decision support. Therefore, the ambitious objective of improving engineering decisions has been developed through each of the 11 papers, which are only representatives of a line of research of which this doctoral dissertation is witness and proof, and whose limits are unfathomable at this moment. The exploration and future evolution of these systems is the next challenge. The approach of new data-driven knowledge formalisation structures opens the possibility of hybridising systems from the formation of their initial information sets, understanding hybridisation as the construction of a single intelligent model with inference capabilities that maximises the advantages of both the symbolic and the statistical approaches while minimising the loss of a consolidated explanation of reasoning. This dissertation is, thus, just one step in a long journey of research and innovation, and is therefore not an end, but the extraordinary opportunity of a beginning.