Speaker: Farhan Samir
Supervisor: Marsha Chechik, Mentor: Alicia Grubb
Title: Empirical Research Project: Which Requirements Approach is Best?
Early-phase requirements engineering focuses on understanding the world in which a system exists. We have created a technique to help stakeholders choose between design alternatives when the satisfaction of each alternative changes over time. We have developed two web-based tools for drawing and simulating scenarios based on different underlying requirements modeling languages and technologies. In this talk, we will introduce our ongoing case study to determine the impact of each tool on modellers’ understanding and ability to reason about time-based design alternatives.
Speaker: Sicong Sheldon Huang
Supervisor: Professor Grosse
Title: Wasserstein Cycle Generative Adversarial Network for Music Emotion Transfer
Recent research in music emotion classification has shown promising results and led to the development of datasets containing the emotion annotation of music. However, there has been no success by way of emotion transfer of music. The goal of this project is to transfer emotions in music. We built the first Wasserstein Cycle Generative Adversarial Network(WCGAN), which is based on Wasserstein Generative Adversarial Network(WGAN) and Cycle Generative Adversarial Network(CycleGAN). CycleGAN is a new artificial deep neural network architecture that has shown excellent results in unpaired image translation. To stabilize the training, we modified the network by using Wasserstein loss in WGAN. The model will take in the raw waveform as the input, transfer the emotion in the music and output the music waveform with the desired emotion. Preliminary experiments showed some promising results: the CycleGAN can transfer audio sample with different timbre. A deep learning model that can transfer music across different emotions will have potential applications in the music industry and implications in music psychology research and cognitive science research.
Speaker: Silvia Sellá Gonzalez
Supervisor: Alec Jacobson, Mentor: Tim Jeruzalski
Title: Solving PDEs on Deconstructed Domains
Partial Differential Equations arise in many areas of science, from physics to biology to signal processing. When analytical solutions are too complex, it is necessary to approximate them via a discrete mesh of the domain. Sometimes a robust triangular mesh of the whole domain is a hard thing to accomplish, and in those cases we tend to break up the domain in various different subdomains with nontrivial intersection and solving the equation in each of them individually. Although this approach solves one problem, it creates another, i.e. how to combine the solutions on the intersection of these domains in a way that approaches the true solution. Our project seeks to solve that problem to some particular cases in a hopefully generalizable manner.