By Vignesh Gokul.
Computer Science [cs]. University of California San Diego, 2024. English.
Abstract: With the development of intelligent chatbots, humans have found a method to communicate with artificial digital assistants. However, human beings are able to communicate an enormous amount of information without ever saying a word, eg gestures and music. My research objective is to enable non-verbal communication between humans and artificial agents, a problem I call co-creativity. My work explores controlling generative models that can intelligently interact with data such as gestures (videos) and music (audio). Another area of my research focuses on ethical issues stemming from using data to train such machine learning models. We investigate ways to protect the privacy of the data owner and prevent unauthorized usage/ leakage of private information during training machine learning models. We then propose a method to create unlearnable audio datasets to prevent unauthorized usage of data for model training.
