School of Engineering Department of Computer Science and Engineering 126 Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: WANG Shaoqi / COMP Course: UROP 2100, Fall The objective of this report is to detail the work I completed as part of the Knowledge Discovery Over Database project, which involved the research: Learned Data-aware Image Representations of Line Charts for Similarity Search. Throughout the project, I accumulated knowledge about the latest techniques in Data Mining, especially the key method in this research, Vision Transformer-based Triplet Autoencoder. I also managed to run and debug the code of this research, and learned a lot of techniques in code running and debugging. In this report, I will explain the important knowledge I learned in this research. It also concludes with a summary of the insights I gained from the project and my perspectives on the future of this project. Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: WANG Yuning / DA Course: UROP 1100, Summer The research community and related industry have been paying increasing attention to efforts for building effective speech-based interfaces to query structured data in relational databases. One existing study demonstrates an improved Speech-to-SQL system, which is the first attempt to directly synthesize SQL based on speech rather than text. Consequently, this system has advanced behavior in terms of efficiency and accuracy. This report aims to introduce my gaining from studying the paper. By having deeper researchbased insights into the classical machine learning models involved, such as Convolutional Neural Network (CNN) and Graph Neural Network (GNN), I obtained a much more comprehensive understanding for the ensemble of these models. Knowledge Discovery Over Database Supervisor: WONG Raymond Chi Wing / CSE Student: YE Yilin / DSCT Course: UROP 1100, Spring This report focuses on the realization and application of recommendation models using the Neural Attentive Session-based Recommendation Model (NARM) in recommendation system. The NARM model makes precise predictions about the user's upcoming activities by utilizing attention mechanisms to extract pertinent information from the session history and return the result. Compared to other existing recommendation model, NARM is more efficient and accurate. As a session-based recommendation model, it responds perfectly to the concerns on security issue from customers. To test its performance on various datasets, we build up the model and record three evaluation metrics and mean loss along training process and it turns out reaching our expectation and it applicable to some basic recommendation tasks such as nest product prediction and location prediction.
RkJQdWJsaXNoZXIy NDk5Njg=