School of Engineering Department of Computer Science and Engineering 102 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: SZE Kai Tik / COGBM Course: UROP3100, Summer This paper reports the development of a Guided Wire Recognition System for Tseung Kwan O Hospital (TKOH). In this project, the challenges and concerning solutions in the training process of the YOLO v5 model will be highlighted, and the current development process will be updated, especially in the stage of user acceptance test, we understand more regarding the customer needs and re-customize our model training based on the hook recognition. In addition, the paper will raise some reflections on the current stage and put forward ideas about follow-up research. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: PENG Zhuoxuan / COSC DAI Tianyuan / COSC Course: UROP4100, Summer UROP1100, Summer Crowd counting is to approximate the number of objects in an image or a video of unconstrained scenes. Current state-of-art method utilized in crowd counting is based on convolutional neural networks (CNNs). Despite satisfactory performance, unreality of exact counting of objects is still the fact due to large variation in object scales and isolated small clusters of objects. To preserve long-range context information when generating density maps, multi-scale architectures are embraced by several researchers. However, long training time, non-effective branch structure and scant of dots in density maps also bring about great challenges. Recently, some researchers argue using the Transformer architecture in vision tasks, mirroring its success in the field of nature language processing. Unfortunately, high resolution of images and largescale variation are obstacles waiting to be solved. In this paper, we compare currently prevailing models in crowd counting by setting our own pipeline and metrics on different datasets. Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: LIU Dingdong / COMP Course: UROP1100, Fall In UROP1100 - Indoor Localization and Mobile Computing project, I learned about the common practices in Indoor localization. I analyzed an existing project, which is the final year project done by other students. Based on the analysis of this previous final year project and another set of data collected by the Microsoft team, we discussed the feasibility of a new indoor localization approach, which uses map information to calibrate the sensor data. Finally, we concentrated on an existing indoor localization problem, namely the floor classifying problem, and my mentor provided a novel solution to it. In this report, I will introduce the work I have done and elaborate on my experiment's analysis. This project serves as an extension of my summer COMP4900 course. In this project, I continued to learn about crowdsourcing methods and have an experiment on them. Hopefully, in the continuing sections of UROP, I can dive into the cutting edge and raise my ideas.