UROP Proceedings 2020-21

School of Engineering Department of Computer Science and Engineering 112 Practical ML-based Mobile Applications Supervisor: CHATZOPOULOS Dimitrios / CSE Student: THIEN Zhong Vei / SENG Course: UROP1100, Summer The idea I have in mind for the UROP project is to build an app for users to order different types of coffee using a mobile device. A pop-up message containing information about which type of coffee is most suited for the user is display as soon as the mobile application is run. Nevertheless, users can still choose coffee of their liking despite the suggested selection. In this context, the prediction of the best-suited coffee would be decided by the current time and weather. Machine learning on the other hand is being implemented to predict the best type of coffee based on all the previous selection of the users of the app. Whenever a user selects a coffee, a dataset will be created and saved to the database to improve the accuracy of the regression analysis. Practical ML-based Mobile Applications Supervisor: CHATZOPOULOS Dimitrios / CSE Student: ZHENG Tianshi / COGBM Course: UROP1100, Summer In this project, I aimed to build a task management app in React Native. The app collects task finished information and data from the user and generate a report based on it. It is a minimum viable product of a mature full-stack task-management app. It includes a front-end UI, backend model and API built by flask RESTful, and it also applies Artificial Intelligence techniques such as Natural Language Processing, to analyze the tasks by category, etc. Data-Efficient, Domain Generalizable and Interpretable Deep Learning Supervisor: CHEN Hao / CSE Student: HUANG Junkai / COSC Course: UROP1100, Summer This experiment is inspired by the Mitosis Domain Generalization (MIDOG) Challenge 2021. Mitotic figure density is an important criterion for tumor diagnosis and prognosis. With the advancement of deep learning, contemporary deep learning-based algorithms for mitotic figure identification can now match the performance of human specialists. However, there is a bottleneck: domain-shift impairs the performance of deep learning models. We undertake a comparison experiment to assess the degree of loss in mitotic figure identification ability using state-of-the-art deep learning models to better demonstrate this concern. In this experiment, we put YOLOv3, Faster R-CNN, and Cascade R-CNN to the test. Then examine the results of the tests. The codes are available at: https://drive.google.com/drive/folders/1zvDqtN6Zgx9UcYVYFSI12_QjFLoozKG?usp=sharing

RkJQdWJsaXNoZXIy NDk5Njg=