School of Engineering Department of Electronic and Computer Engineering 137 Artificial Intelligence Methods for Medical Videos Supervisor: LI Xiaomeng / ECE Student: HUANG Junkai / COSC Course: UROP1100, Fall Phase recognition of surgical video is of particular interest to computer assisted surgery systems. Current fully supervised methods require the training data to be manually labeled by human experts, which is a tedious and time-consuming work. In this project, we aim to develop a surgical video phase recognition method that can be trained under a weakly-supervised setting, in which only the order of phases appears in the video is provided during training. We have conducted experiments on state-of-the-art video phase segmentation pipelines and methods including MSTCN, Timestamp supervision, and Cross Pseudo Supervision. All the experiments are performed using Cholec80 dataset. Then we analyze the results and look for possible chances for improvements. Artificial Intelligence Methods for Medical Videos Supervisor: LI Xiaomeng / ECE Student: LIU Yunfei / CPEG Course: UROP1100, Spring UROP2100, Summer An adversarial attack is a technique to create adversarial models. Thus, an adversarial model is a contribution to an AI model that is deliberately intended to make a model commit an error in its expectations notwithstanding looking like a substantial contribution to a human. Thusly, to stay away from such bothers, adversarial training is especially significant, which is a savage power regulated learning technique where however many adversarial models as could be expected under the circumstances are taken care of into the model and expressly marked as compromising. This is a similar methodology the normal antivirus programming utilized on PCs utilizes, with different updates consistently. In this article, I will present a report about adversarial training that I have perused and comprehend in UROP2100.