Neural logic reasoning and applications

Abstract

Recent years have witnessed the success of deep neural networks in many research areas. The fundamental idea behind the design of most neural networks is to learn similarity patterns from data for prediction and inference, which lacks the ability of cognitive reasoning. However, the concrete ability of reasoning is critical to many theoretical and practical problems. On the other hand, traditional symbolic reasoning methods do well in making logical inferences, but they are mostly hard rule-based reasoning, which limits their generalization ability to different tasks since different tasks may require different rules. Both reasoning and generalization ability are important for prediction tasks such as recommender systems, where reasoning provides a strong connection between user history and target items for accurate prediction, and generalization helps the model to draw a robust user portrait over noisy inputs. In this dissertation, we propose a Neural Logical Reasoning (NLR) framework to integrate the power of deep learning and logical reasoning. NLR is a dynamic modularized neural architecture that learns basic logical operations such as AND, OR, NOT as neural modules, and conducts propositional logical reasoning through the logical structured network for inference. In order to examine the effectiveness of the neural logical reasoning concept, we do extensive studies including (i) Conducting experiments on theoretical tasks to verify that our proposed framework has the ability to conduct logical inference; (ii) Applying the neural logical reasoning framework to recommendation task to explore its ability of solving real-word problems; (iii) Extending the application to more general graph-related tasks such as knowledge graph completion. Experiments show that our approach achieves state-of-the-art performance in various application scenarios. Moreover, we utilize the neural architecture search strategy to allow the model to learn the adaptive logical neural architectures automatically which brings flexibility to our framework.