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: XU Yin Yui / COMP Course: UROP 1000, Summer We study the problem of efficiency of Graphics in predicting the embedding of new signal samples, which involved in the calculation of the entire dataset. We propose an averaging method, which learned features from high related data only, to reduce the computing and time cost when doing embedding prediction for new signal sample. The feasibility of this method is verified (> 90% in total accuracy) by using the dataset used in Graphics. In addition, we also proposed the idea of using pre-trained model to predict the embedding. Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: JU Jongho / COSC Course: UROP 2100, Fall Indoor Location-Based Services (LBS) play a pivotal role in smart cities, with a surge in demand for effective localization. Current approaches relying on Wi-Fi fingerprint and Bluetooth Beacon encounter challenges such as significant errors and high time overhead due to signal ambiguities. To address these issues, a novel framework employing multiple adaptive representations of signal sequences for localization is introduced. Utilizing geomagnetic signals as input, the approach extracts spatial and temporal representations, employing optimized neural networks for enhanced location clue extraction. The ensemble learning mechanism, coupled with a weighted k-NN-based algorithm, reinforces robustness. Among these representations, spatial features of the geometric sequences are tested. To implement feature extraction from the spatial sequence, several machine learning models are built to test the approximation from the spatial input data. Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: WANG Yumeng / COMP Course: UROP 1100, Spring This project applies Natural Language Processing (NLP) techniques, specifically Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Pre-trained Transformer (GPT) with multimodal prompt engineering, to the domains of indoor localization using magnetic data. By leveraging the power of NLP to analyze and interpret complex signal data, the project aims to enhance the accuracy and efficiency of indoor positioning systems. The methodology includes using CNNs for extracting features from raw signal data, RNNs for capturing temporal dependencies in sequential data, and GPT-4’s ability to discover multimodal inputs, thereby facilitating a deeper understanding and more precise localization within indoor environments. The project demonstrates the potential of integrating advanced NLP techniques with signal processing to unlock new possibilities in indoor localization.
RkJQdWJsaXNoZXIy NDk5Njg=