School of Engineering Department of Computer Science and Engineering 86 AI Meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: JU Jonghyeon / COSC Course: UROP 2100, Fall With recent advancements in AI, there has been significant advancement in the deep learning models. As the performance of the model gets stronger, enormous amounts of computation and memory are required as a tradeoff. To resolve these issues, emergent technology is network pruning, which is the process of making the model to be lightweight. However, it is often time-consuming as it requires multiple iterations of the training. Hence, the method that uses single-pass iteration for training by gradually adding the important channels is researched. To achieve this, many basic deep-learning technologies and many papers regarding network pruning are studied to get an idea of the proposed solution. Then, the baseline of this project is set by training the ResNet-18 model. AI Meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: KANG Zhaowei / COSC Course: UROP 1100, Fall UROP 2100, Spring UROP 3100, Summer Wi-Fi fingerprint augmentation is a technique that utilizes a machine learning model to generate virtual WiFi fingerprints, thereby enhancing the precision of Wi-Fi fingerprint localization. However, current Wi-Fi fingerprint augmentation models require retraining when applied to a new site, which is time-consuming. To address this issue, we propose a multi-site fingerprint augmentation model that do the augmentation by two steps: the patch-level step and the site-level step. And by incorporating topographic maps and position embeddings, our model can perform Wi-Fi fingerprint augmentation on a new site without retraining. We also evaluate the model’s performance on several real-world datasets. AI Meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: POON Kwan Hei Anson / COMP Course: UROP 1100, Fall UROP 2100, Spring Advancements in computer vision raise privacy concerns due to the extensive personal information captured by camera-based human action recognition systems. To address this, researchers are exploring millimeterwave (mmWave) technology, used in 5G, which offers privacy-friendly, non-intrusive spatial data capture. This paper evaluates mmWave-based models for human action recognition, crucial for elderly care, where active systems like Personal Emergency Link Services fall short. MmWave sensors enable continuous, private monitoring in sensitive areas. Our evaluation of the MiliPoint mmWave model shows accurate capture of larger body movements but difficulties with finer motions due to sparse point cloud data. These findings highlight both the potential and limitations of mmWave technology for privacy-sensitive applications.
RkJQdWJsaXNoZXIy NDk5Njg=