Research

Research Statement

Our mission is to advance Generative AI (Gen AI) to enhance productivity and benefit as many people as possible across various sectors of society and industry. We believe that improving Gen AI is not only about improving image generation quality; rather it’s about understanding how people interact with AI and how their actual work and life. By focusing on creating intuitive, efficient, and user-friendly Gen AI models that accurately reflect users’ intentions, we aim to drive productivity innovation on a broad scale. Upon this, our lab revolves around four main research themes toward this goal: Gen AI + X, Creative Gen AI, Ethical Gen AI, and Evaluation for Gen AI.

Gen AI + X We aim to enhance productivity across various industries by applying Gen AI technologies. Our research focuses on empowering domain experts with specialized workflows to effectively utilize Gen AI models in their respective fields. For this research theme, this involves improving the controllability of Gen AI systems and ensuring they provide reliable and trustworthy outputs. By addressing these challenges, we can tailor Gen AI solutions to diverse domains such as manufacturing, healthcare, and more.

Creative Gen AI We also want to revolutionize the creative process in art and media through the integration of Gen AI. Beyond simple image generation, we strive to enable creators to incorporate Gen AI into their workflows so that productivity is boosted and they can focus more on innovation and artistic expression. We are also dedicated to enhancing the controllability of Gen AI models. With these, we believe that artists and media professionals can intuitively and effectively harness these tools to expand creative boundaries.

Example research areas include 1) developing artistic image synthesis with advanced style transfer methods, 2) creating Gen AI tools for innovative content generation in media production, 3) enhancing user control in Gen AI models for more precise artistic outputs, and 4) integrating Gen AI into traditional creative workflows to augment productivity.

Ethical Gen AI While Gen AI has driven advancements across various fields, it has also introduced significant social and ethical challenges. Issues such as sensitive data leakage and copyright infringement of creators have become increasingly concerning. Our research focuses on addressing these challenges by developing methods to protect creators’ rights, enabling pre-trained Gen AI models to forget sensitive knowledge, and creating mechanisms to identify AI-generated content. These efforts aim to promote trustworthy and secure Gen AI solutions.

Example research areas include 1) protecting intellectual property rights of creators in Gen AI outputs (as shown in above figure), 2) developing machine unlearning to enhance data privacy, 3) designing systems for detecting AI-generated images and content, and 4) creating ethical guidelines and frameworks for responsible Gen AI use.

Evaluation for Gen AI To ensure the responsible and effective advancement of Gen AIs, establishing robust evaluation protocols is essential. Our research emphasizes the development and refinement of evaluation metrics that assess Gen AI models on effectiveness, reliability, and ethical considerations. By implementing comprehensive evaluation frameworks, we can guide Gen AI development in the right direction. To do so, Gen AI not only meet performance objectives but also adhere to ethical guidelines and foster user trust.

Example research areas include 1) developing metrics for assessing the quality and authenticity of Gen AI outputs, 2) evaluating the reliability and robustness of Gen AI models in various applications, 3) assessing ethical implications and biases in Gen AI-generated content, and 4) creating standardized evaluation protocols for Gen AI across different domains.