UROP Proceedings 2022-23

School of Engineering Department of Computer Science and Engineering 91 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN, Gary Shueng Han / CSE Student: ZHANG, Liyu / COMP Course: UROP1100, Fall The dataset is some brain masks with tumors on them, offered by the Stanford Medical School. Our main goal is to recognize the tumors from these datasets. The dataset AI_SHC_met_156 contains two subsets: train dataset and test dataset. In the train dataset, there are 104 directories and in the test dataset there are 52 directories. In each directory, there are 5 .nii files, 4 of them are tumor images with float data, and the remaining is a ‘mask’ with bool data. We make some modifications to extract the data we want and make them compatible with the model. We read the files into tuples of (imgs, mask), where imgs has shape (4, H, W, S) and mask has shape (H, W, S). H == Height, W == Width, S == Slice, and the first dimension 4 implies that this dataset contains 4 sequences for each instance. Then we use the DataLoader in the PyTorch package to load the data for further use. After the .nii files have been converted into Python Numpy arrays, the shape of each array from one file is [160, 4, 512, 512]. 160 represents that there are 160 tumor image slides in each file, and the remaining dimensions mean that for each tumor image slide there are 4 channels of 512 * 512 pixels. Indoor Localization and Mobile Computing Supervisor: CHAN, Gary Shueng Han / CSE Student: CHEN, Siyu / CPEG Course: UROP3100, Spring This paper summarizes an indoor particle filter method designed for smart mobile device positioning under an indoor environment with floor constraints provided. The existing indoor positioning algorithm based on dead reckoning has the disadvantages of large cumulative error and low positioning accuracy, so a particle filter method is applied to try to solve this problem. This method uses the smart mobile device’s built-in sensor to determine user behavior and azimuth. And it also uses the known indoor floor constraints to control the weight of particles during the positioning process, thereby correcting the target position.