School of Engineering Department of Computer Science and Engineering 123 Acceleration Techniques for Real-time Rendering Supervisor: SANDER Pedro / CSE Student: HU Benran / DSCT Course: UROP3100, Fall Real-time high-quality rendering in virtual reality is essential for immersive experiences but remains to be a challenging area due to the performance restrictions of untethered head-mounted displays as well as the need for higher framerates and resolutions. On the other hand, perception-based rendering acceleration methods have long been adopted in monoscopic rendering to improve the performance while not sacrificing the perceivable quality. Given that stereo rendering enables us to exploit more properties of the human visual system, it is then a natural choice to boost the performance of VR applications with perception-based rendering methods. In this report, we will examine characteristics of the human visual system that are useful for efficient rendering and elaborate on the design of perception experiments to be performed for verifying and evaluating the feasibility of utilizing these characteristics. Commonsense Reasoning with Knowledge Graphs Supervisor: SONG Yangqiu / CSE Student: CHOI Sehyun / COMP Course: UROP2100, Fall Commonsense Knowledge Base (CSKB) Population is a task where it aims to automatically introduce new commonsense entities and relations into CSKB. In our previous work, it was found that incorporation of graph structure into the rich semantics of pre-trained language models led to a promising results. Improving upon this finding, a new graph neural network framework that can effectively capture a larger context of supporting graph information for each candidate knowledge and different pre-trained language model configurations were experimented. The empirical evaluation support the new model framework’s ability to generalize better to unseen examples, while eliciting the existence of spurious bias in the training dataset.