김동준 박사 (Stanford University)
Title: Recent Trends of Distillation Methods in Diffusion Models
Image synthesis with diffusion models has been highlighted in computer vision and graphics for its remarkable ability to various tasks, such as text-to-image generation or image editing. However, diffusion models require an iterative denoising process for sample generation, which prevents diffusion models from being widely used for practical real-time industrial applications. In this talk, I will cover the forefront of research aimed at reducing the computational demands of diffusion models, making them more suitable for use in real-world industrial scenarios.
Dongjun Kim is currently a PostDoc advised by Professor Stefano Ermon, at Stanford University. He received M.S in the department of mathematical science at KAIST in 2016 and Ph.D. in the department of industrial and systems engineering at KAIST in 2023, advised by Professor Il-Chul Moon. His research primarily focuses on generative modeling and its downstream real-world applications.
주재걸 교수 (KAIST)
Title: 3D-aware Image Generation and Editing Approaches and Applications
Recently, 3D-aware image generation and editing techniques has been actively studied, significantly advancing image generation quality and speed. In this talk, I will cover various different problem settings and their relevant approaches, along with real-world applications, such as NeRF, pi-GAN, EG3D, Instant-NGP, Nerfies, and 3D Gaussian Splatting.
Mar. 2020 - Present, Associate Professor, KimJaechul Graduate School of Artificial Intelligence, KAIST
Mar. 2019 - Feb. 2020, Associate Professor, Dept. of Artificial Intelligence, Korea University
Mar. 2015 - Aug. 2019, Assistant Professor, Dept. of Computer Science and Engineering, Korea University
송경우 교수 (연세대)
Title: Fairness in Machine Learning: Challenges, Analysis, and Solutions
Recent breakthroughs in machine learning have ushered in a new phase of real-world AI applications. However, many case studies show that even large-scale multimodal models often fail to achieve fairness. This tutorial will introduce key concepts of fairness and propose strategies to address fairness-related risks. Specifically, the tutorial will highlight the fairness challenges inherent in current machine learning models, offer a formal definition of fairness, and conclude with a comprehensive review of principles and recent trends for mitigating fairness-related risks.
Kyungwoo Song is an Assistant Professor at the Department of Applied Statistics and the Department of Statistics and Data Science at Yonsei University. He is also a visiting researcher at the Seoul Institute of Technology. His research focuses on multimodal learning, generative models, and causality to make machine learning algorithms more trustworthy. He received his Ph.D. in Industrial & Systems Engineering and his B.S. in Mathematical Sciences from KAIST.
임재환 교수 (KAIST)
Title: Robot Learning
Robotics is one of the most exciting and diverse applications for machine learning. In this tutorial, I will introduce the recent advances in the field of robot learning, with the focus on perception, manipulation, and reasoning. The tutorial will cover broad range of topics, such as reinforcement learning, visual perception, multimodal learning, and human-robot interaction. Additionally, I will also discuss recent trends of combining large language models and world models with robotics.
Joseph Lim is an Associate Professor in the Kim Jaechul School of Artificial Intelligence at Korea Advanced Institute of Science and Technology (KAIST). Previously, he was an assistant professor at the University of Southern California. Before that, he completed PhD at Massachusetts Institute of Technology under the guidance of Professor Antonio Torralba, and also had a half-year long postdoc under Professor William Freeman and a year long postdoc under Professor Fei-Fei Li at Stanford University. He received his bachelor degree at the University of California - Berkeley, where he worked in the Computer Vision lab under the guidance of Professor Jitendra Malik. He received the best presentation paper award at CoRL2020, and the best system paper award at RSS2023.
이재윤 교수 (서울대)
Title: Injecting output constraints and dependencies into neural models
Injecting human prior to a machine learning model has been a long-lasting topic in the research community. With the advancement of neural black-box models such as ChatGPT and its silly mistake despite its remarkable performance, the topic of injecting constraints is gaining ever more interest. In this two-part tutorial, I will provide an overview of constraint injection methods as well as introduce my recent work on reflecting latent constraints. First, I will give an overview of constraint injection methods centered around symbolic output constraints on neural natural language processing models. Second, in contrast to the first part, I will discuss how the model could capture and reflect constraints when they are not known a priori. In this second part, I will introduce my recent work on structured energy networks and box embeddings.
Jay-Yoon Lee is an assistant professor in the Graduate School of Data Science at Seoul National University (SNU). His research interest primarily lies in injecting knowledge, and constraints into machine learning models using the tools of structured prediction, reinforcement learning, and multi-task learning. He has worked on injecting hard constraints and logical rules into neural NLP models during his Ph.D., and now he is expanding his research area towards automatically capturing constraints, human-interactive models, and science problems such as protein interaction. Prior to joining SNU, he conducted his postdoctoral research in the College of Information & Computer Sciences at UMass Amherst with Professor Andrew McCallum. Jay-Yoon received his Ph.D. in Computer Science in 2020 from Carnegie Mellon University where he was advised by Professor Jaime Carbonell and received his B.S. from KAIST in electrical engineering.
김승룡 교수 (고려대)
Title: Recent Advances in Language-driven Computer Vision Tasks
Providing user intentions for various computer vision tasks can be achieved through a variety of input conditions. Recently, leveraging language-based user prompting has gained widespread adoption due to its efficiency and effectiveness. Nonetheless, since vision data, including image, video, or 3D, and language data fundamentally differ in their form, representation, and semantics, careful design is essential to effectively apply them to a range of computer vision tasks. In this tutorial, we will explore recent advancements in language-driven computer vision tasks. Specifically, we will examine current progress in tasks such as open-vocabulary segmentation, text-to-image generation, text-to-3D generation, and text-to-video generation. Furthermore, we will engage in discussions regarding the future directions in these areas.
Seungryong Kim is an assistant professor in Department of Computer Science and Engineering, Korea University, Korea. Before joining Korea University, he was Post-Doctoral Researcher in Yonsei University, Korea (from 2018-2019) and EPFL, Switzerland (from 2019-2020). His research interests lie in computer vision, machine learning, and computational photography. He actively serves as program committee in prominent computer vision and artificial intelligence conferences such as CVPR, ICCV, ECCV, NeurIPS, etc.