Gong, Taesik
공태식
Department of Computer Science and Engineering(컴퓨터공학과)
- Ubiquitous AI Lab
- Website
- Lab Description
- Our lab’s research is at the intersection of cutting-edge AI and its seamless integration into real-world applications, focusing on enhancing user experiences through intelligent, adaptive, and efficient on-device AI systems. We currently explore three primary research areas (but not limited to) that collectively aim to push the boundaries of AI technology, making it more human-centered, personalized, and resource-efficient.
1. Human-Centered AI Applications“How can we enrich users’ daily lives with on-device AI?”Our research in human-centered AI is driven by the goal of embedding AI into everyday devices to enhance user experiences while ensuring privacy and security. As AI continues to make strides in tasks like image classification and natural language processing, its deployment on devices with various sensing capabilities (e.g., vision, audio, and motion sensors) creates new opportunities for intelligent applications. These applications process sensitive user data locally, reducing privacy concerns associated with cloud-based solutions. Our work is pioneering innovative ways to leverage on-device AI to enrich daily life, focusing on how these technologies can operate seamlessly and beneficially within the user’s environment.Our recent projects like Knocker (UbiComp ’19), MyDJ (CHI ’22), and MIRROR (UbiComp ’24) exemplify our efforts to bring AI closer to users in a manner that is both practical and privacy-conscious.
2. Adaptive and Personalized AI“How can we adapt AI to different individuals and environments?”Adapting AI to diverse users and environments is critical to ensuring its broader applicability and effectiveness. Users differ in their physical conditions, behaviors, and lifestyles, and devices vary in their technical capabilities. These variations lead to significant challenges when deploying AI models trained on specific data sets to new environments. Our research addresses these challenges by developing frameworks that allow AI systems to adapt with minimal user intervention, ensuring that performance remains consistent even when conditions change.We explore techniques such as test-time adaptation, few-shot learning, and self-supervised learning to create AI systems that can personalize themselves to individual users and their unique environments. Projects like NOTE (NeurIPS ’22), SoTTA (NeurIPS ’23), and AETTA (CVPR ’24) highlight our contributions to making AI more adaptive and personalized, reducing the barriers to its widespread adoption.
3. Efficient On-Device AI Systems“How can we support AI in a resource-efficient manner?”Deploying AI on devices with limited resources presents a formidable challenge. Unlike cloud-based systems, which have vast computational resources, on-device AI must operate under strict constraints, such as limited processing power, memory, and battery life. Our research focuses on creating AI systems that are not only powerful but also resource-efficient, enabling them to perform advanced computations on small devices without compromising speed or functionality.We delve into areas such as collaborative learning, split learning, and the development of tiny AI accelerators, aiming to optimize the use of available resources while maintaining high performance. Our projects, including MetaSense (SenSys ’19), DAPPER (UbiComp ’23), and Synergy (arXiv ’24), demonstrate our commitment to making on-device AI systems more efficient and accessible.In summary, our lab’s research is dedicated to advancing the field of AI by focusing on human-centered applications, adaptive and personalized AI, and efficient on-device AI systems. Through our work, we aim to bridge the gap between AI’s potential and its practical, real-world application, ensuring that AI technologies are not only intelligent but also user-friendly, adaptable, and resource-conscious.
There are no entries in the repository.