UROP Proceedings 2022-23

School of Engineering Department of Computer Science and Engineering 131 Large-Scale Spatiotemporal Data Analytics and Learning Supervisor: ZHOU, Xiaofang / CSE Student: YIN, Zhuohao / DSCT Course: UROP1100, Fall While conventional data science approaches have been applied on various domains to extract insights from large quantities of data, spatiotemporal data, due to its structural difference, have been poorly explored and exploited. While some classic machine learning algorithms are proven to work well with spatiotemporal data, such as K-means, K-Medoids, and EM algorithm, these methods can only handle limited data formats and lack interpretability. A remedy for low interpretability is through data visualizations. To this end, this UROP project seeks to develop a user-friendly API, which allows users to upload spatiotemporal data and produce interactive and informative visualizations. Large-Scale Spatiotemporal Data Analytics and Learning Supervisor: ZHOU, Xiaofang / CSE Student: ZHANG, Zhanhua / DSCT Course: UROP1100, Spring This report is to update on the progress of my research on the latest AI techniques. In this semester, I have been exploring various AI paradigms, familiarizing with their major principles, applications, potential, and limitations. I have also been combing and categorizing them in different areas based on their principles and purposes. I hope this report can demonstrate my understanding of relevant knowledge. In this report, I will discuss the development of AGI and the use of artificial neural networks to build computer systems that can match human cognitive abilities, as well as the recent advances in NLP models, such as SOTA NLP, ChatGPT, and Stable Diffusion. Towards Efficient Queries on Graph Databases Supervisor: ZHOU, Xiaofang / CSE Student: LI, Yunqi / CSE Course: UROP3200, Summer In this report, we present a novel framework called FRESH which could handle multiple types of queries in an outsourced graph. The framework harmonize client side and cloud side processing together to achieve efficiency and privacy protection during answering queries. On client, we develop a privacy-preserving contraction method to reduce the graph’s size as well as protect structural privacy during outsourcing. On cloud, as a proof of concept, we use three different types of queries to show FRESH could smoothly deal with queries on compacted graph. Finally, we demonstrate the experiment result at last to compare our framework against traditional frameworks.