Técnicas de Inteligencia Artificial basadas en una integración de la lógica simbólica y no-simbólica

Permanent URI for this collection

https://repositorio.uai.edu.ar/handle/123456789/301

Browse

Browsing Técnicas de Inteligencia Artificial basadas en una integración de la lógica simbólica y no-simbólica by Subject "artificial intelligence"
  • Item
    Artificial Intelligence techniques based on the integration of symbolic logic and deep neural networks : a systematic review of the literature
    (Iberoamerican Society of Artificial Intelligence (IBERAMIA), 2022-6) Negro, Pablo Ariel ; Pons, Claudia Fabiana
    Artificial Intelligence is tackled from two predominant but very different approaches: symbolic Artificial Intelligence, which is inspired by mathematical logic and is based on the manipulation of abstract linguistic representations, and non-symbolic Artificial Intelligence, which focuses on the construction of predictive mathematical models from large sample data sets. Significantly, the shortcomings of each of these approaches align with the strengths of the other, suggesting that an integration between them would be beneficial. A successful synthesis of symbolic and non-symbolic artificial intelligence would give us the advantages of both worlds. This work aims to identify and classify solutions and architectures that use applied Artificial Intelligence techniques, based on the integration of symbolic and non-symbolic logic (particularly machine learning with artificial neural networks), to provide a comprehensive, exhaustive and organized vision of the solutions available in the literature, making them the subject of a carefully designed and implemented SLR (Systematic Literature Review). The resulting technologies are discussed and evaluated from both perspectives: symbolic and non-symbolic Artificial Intelligence. The PICOC method (Population, Intervention, Comparison, Outputs, Context) plus Limits, which determine the scope of the search, has been used to define the research questions and analyze the results. From a total of 65 candidate studies found, 24 articles (37%) relevant to this study were selected. Each study also focuses on different application domains such as intelligent agents, image classification, theorem provers, cyber-security, image interpretation, mathematics, medicine, robotics and general application. Through the analysis of the selected works, it was possible to classify, organize and explain the different ways in which the deficiencies of non-symbolic Artificial Intelligence are addressed by proposals based on symbolic logic. The study also determined in which stages of the development process said proposals are applied. In addition, the study made it possible to determine which are the logic tools that are preferably applied, for each area and each domain. Although no clear architectural pattern has been found, efforts to find a general-purpose model that combines both worlds are driving trends and research efforts.
  • Item
    Rule Extraction in Trained Feedforward Deep Neural Networks - Integrating Cosine Similarity and Logic for Explainability
    (Universidad Abierta Interamericana. Facultad de Tecnología Informática, 2024-12-30) Negro, Pablo Ariel ; Pons, Claudia Fabiana
    Explainability is a fundamental aspect in the field of machine learning, particularly in ensuring transparency and trust in decision-making processes. As the complexity of machine learning models increases, the integration of neural and symbolic approaches has emerged as a promising solution to the explainability problem. In this context, the utilization of search methods for rule extraction in trained deep neural networks has been proven effective. This involves the examination of weight and bias values generated by the network, typically through calculating the correlation between weight vectors and outputs. The hypothesis developed in this article states that by incorporating cosine similarity in this process, the search space can be efficiently narrowed down to identify the critical path connecting inputs to results. Furthermore, to provide a more comprehensive and interpretable understanding of the decision making process, this article proposes the integration of first-order logic (FOL) in the rule extraction process. By leveraging cosine similarity and FOL, a groundbreaking algorithm that is capable of extracting and explaining the rule patterns learned by a feedforward trained neural network was designed and implemented. The algorithm was tested in three use cases showing effectiveness in providing insights into the model’s behavior.