Presentation Details

13:30~13:40   [개회사, 환영사]

• 개회사 (이지형 교수)

• 환영사 (TBD)

13:40~14:00  [성균관대 AI 성과 발표]

14:00~15:00  [초청강연]

• Towards Strong and Robust Deep Models 

• 윤상두 소장 (네이버 AI Lab)

• Abstract: This talk will present NAVER AI's recent research in developing strong and robust deep learning models. I will introduce cutting-edge techniques to enhance the model's generalization ability and out-of-domain robustness, focusing on innovative approaches in data processing, training methodologies, and model architecture. Through this talk, I expect to share meaningful insights and encourage discussions on the challenges and opportunities in the era of large-scale modeling. 

• Bio: Dr. Sangdoo Yun is a Research Director at Naver AI Lab. He received his Ph.D. and M.S. in Computer Vision from Seoul National University in 2017 and 2018, respectively. He is also an Adjunct Professor at the Seoul National University AI Institute. His research interests are in large-scale and multi-modal machine learning with a focus on real-world applications. His expertise spans both industry and academia, enabling him to contribute to the advancement of AI technology at the forefront of both sectors. 

15:15~16:30 [교수발표]

• SW & HW Optimization for Accurate and Efficient 3D Hand Pose Estimation 

• 고종환 교수

• 3D hand pose estimation, which determines the 3D coordinates of human hand joints, is a crucial task in various interaction applications such as AR glasses and robotic control. This presentation introduces various techniques proposed by our research lab, such as folding decoder, recurrent network, diffusion model, and graph transformer, to perform this task with high accuracy. Additionally, I present model compression and hardware/system design cases that enable efficient processing on resource-constrained devices. 

• Native DRAM Cache: Re-architecting DRAM as a Large-Scale Cache for Data Centers 

• 김정래 교수

• Contemporary data center CPUs are experiencing an unprecedented surge in core count. This trend necessitates scrutinized Last-Level Cache (LLC) strategies to accommodate increasing capacity demands. While DRAM offers significant capacity, using it as a cache poses challenges related to latency and energy. This talk introduces Native DRAM Cache (NDC), a novel DRAM architecture specifically designed to operate as a cache. NDC features innovative approaches, such as conducting tag matching and way selection within a DRAM subarray and repurposing existing precharge transistors for tag matching. These innovations facilitate Caching-In-Memory (CIM) and enable NDC to serve as a high-capacity LLC with high set-associativity, low-latency, high-throughput, and low-energy. 

• Robot, Embodied AI and Generative AI 

• 유현우 교수

• This seminar will explore the importance of Long-horizon Task Planning for mobile robots in industrial settings. We will discuss the ultimate goals of robotics, such as enhanced efficiency and task automation, and how generative AI is being applied to achieve these objectives, along with the essential technological components required for successful integration. Additionally, we will cover recent developments in generative model-based behavior learning and motion planning for heterogeneous robots, as well as general navigation strategies. Emphasis will be placed on the significance of cognitive generalization in robotics, allowing robots to adapt across different environments and tasks. Finally, we will examine current development trends and limitations using the latest technologies and publicly available datasets. 

16:30~18:00     [포스터 세션]

• Poster Session Details 참조