School of Engineering Department of Computer Science and Engineering 119 Graph Machine Learning for Logical Reasoning Supervisor: SONG, Yangqiu / CSE Student: TSANG, Hong Ting / COMP Course: UROP1000, Summer This project endeavours to merge logical reasoning and machine learning by constructing a framework that leverages graph machine learning to enable reasoning capabilities. Its primary objective is to devise a novel approach for tackling intricate logical problems and potentially pave the way for the development of General AI. This report provides an overview of current research in the field of logical reasoning, encompassing both traditional algorithms and cutting-edge deep learning techniques. By delving into these studies, we aim to gain valuable insights and identify potential avenues for advancing the integration of logical reasoning and machine learning. Graph Machine Learning for Logical Reasoning Supervisor: SONG, Yangqiu / CSE Student: WANG, Yicheng / COSC Course: UROP4100, Summer Abductive Reasoning on a knowledge graph involves the challenging task of discovering the most suitable first-order logical query that closely matches a given answer set. Although numerous studies have focused on the inverse problem, we embark on exploring innovative approaches, leveraging existing techniques like the encoder-decoder transformer T5, to address this novel challenge. Our project introduces a more efficient representation for first-order logical queries and devises a novel decoding strategy. Drawing inspiration from the Abstract Meaning Representation field, we design an evaluation metric to assess the performance of the proposed model. Through experimentation, comparison, and thorough analysis on various datasets, we aim to showcase the effectiveness and potential of our approach. Graph Machine Learning for Logical Reasoning Supervisor: SONG, Yangqiu / CSE Student: ZHANG, Yujun / COSC Course: UROP1000, Summer Intelligence involves both reasoning and learning. This UROP project focuses on exploring the integration of logical reasoning with machine learning. The goal is to develop a framework for learning algorithms that can learn reasoning abilities autonomously, creating a novel approach to solving complex problems that require logical reasoning, and eventually building AI that’re capable of human-level reasoning. In the following, I will first share the framework of a field made up by me named AI4Reasoning. This framework is drafted based on my own understanding gained through this process. Then, the foundational concepts that underlie modern symbolic reasoning engines (e.g. SAT, SMT) is introduced. Lastly, I will share my survey of recent research that uses machine learning to enhance reasoning systems, as well as the reverse.
RkJQdWJsaXNoZXIy NDk5Njg=