School of Engineering Department of Computer Science and Engineering 88 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN, Gary Shueng Han / CSE Student: HO, Ka Shuen / COMP Course: UROP1100, Fall With a more widespread of technology, digital healthcare has been a common place. Taking the advantages of large volume patients’ data available, a disease prediction system could be developed capable for covering a larger variety of diseases and handling more complex relations with more valid conditions of patients. A graph structure of machine learning is a prediction approach that is often being examined. Research on using different graph structures and refining the approaches are important to knowing the relations and improving the prediction system. Some other notable processes include data collection and preprocessing to enable a more relevant data input and to address concerns such as privacy. This project aims to develop a system for enhancing treatment on blood pressure. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN, Gary Shueng Han / CSE Student: JU, Jonghyeon / COSC Course: UROP1100, Fall Stroke is a significant cause of disease burden in the world, affecting 12.2 million people worldwide and 6.5 million died as a result. The recent medical technology is to find stroke lesions by multimodal imaging and the doctors use these found lesions for the reperfusion decision. Multimodal imaging technology nowadays uses the software RAPID, however, this software is very expensive and noise-prone. Hence, the AI model using deep-learning technologies (CNN and U-Nets) is used to predict the infarct core using MRI obtained from the patient and this research aims to obtain performance as good as RAPID. To achieve this, basic deeplearning technologies like PyTorch are studied and many related papers are studied to set the baseline of this project. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN, Gary Shueng Han / CSE Student: KANG, Zhaowei / SENG Course: UROP1000, Summer Wi-Fi fingerprint-based localization is an important research direction in indoor localization. Increasing the number of reference points (RPs) collected offline can improve its positioning accuracy. However, manual collection of fingerprints results in a high cost. Fingerprint augment is an effective solution to reduce the cost while ensuring localization accuracy. We proposed a super-resolution-based model to generate the augmented fingerprint image with a rough collection of signals at limited locations. The framework of the model is formulated, and the implementation of the model is given, including the fingerprint to image convention module, image extract module, and the super-resolution module. We evaluate the model on campus and measure the similarity between the model output and the contour of the floor plan.
RkJQdWJsaXNoZXIy NDk5Njg=