School of Engineering Department of Computer Science and Engineering 125 Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: GONG Ruiliang / COMP Course: UROP 1100, Fall This report comprehensively introduces the learning outcomes and current progress of this semester in the Undergraduate Research Opportunity Program (UROP) project under the guidance of Professor Raymond Chi-Wing WONG. The main topic and the objective of the UROP project developing a Natural Language Processing (NLP) based system using the significant tool of PyTorch. The project involved self-learning knowledge of the algorithm, getting a comprehension of Natural Language Processing, and self-learning PyTorch through the official PyTorch documentation website (https://pytorch.org/tutorials/index.html). It mainly includes the background, objectives, methodology, and challenges encountered. The main idea is about my understanding of NLP concepts, basic algorithms, and the experience of PyTorch implementation I have gained during the project. Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: LIU Zihe / COMP Course: UROP 2100, Fall UROP 3200, Summer Question and Answer (QA) systems are integral to the process of knowledge acquisition, facilitating the extraction of insights from complex data. While much research has focused on QA in general-purpose programming languages, less attention has been given to declarative visual languages (DVLs), such as VegaLite, which are critical for data visualization tasks. FeVisQA, developed under the supervision of Prof. Raymond Chi-Wing Wong, represents the first neural network-based system designed for QA over data visualizations. This system is unique in its ability to predict answers to natural language questions by interpreting DVL-based visualization specifications. The FeVisQA system not only introduces a novel task but also employs a sophisticated model architecture, FeVisQANet, which includes a multi-modal encoder and an adaptive decoder. This paper extends the existing FeVisQA framework by introducing advanced implementation and improvement strategies, particularly focusing on hybrid models and attention mechanisms within Transformer architectures. Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: NIE Weihao / COMP Course: UROP 1100, Fall UROP 2100, Spring First of all, this report will introduce some basic knowledge about deep learning and the constructions of some basic models (CNN, RNN…). Subsequently, this report will introduce three innovative approaches to enhancing human-computer interaction through natural language processing and data visualization. It will discuss the development of systems that allow for natural language-driven database queries and the automatic generation of data visualizations, highlighting the importance of high-quality data synthesis for text-to-SQL parsing.
RkJQdWJsaXNoZXIy NDk5Njg=