Seminars by Assistant Professor Jin-Sun Park and Professor Won Joo Hwang (Pusan National University)
Title: Multi-Modal Sensor Systems for Robust Visual Perception in Real-world Environments
Speaker: Assistant Professor Jin-Sun Park
Abstract:
This seminar introduces various multi-modal sensor systems for visual perception across diverse real-world applications. Those systems integrate multiple sensors including RGB, NIR, and thermal cameras, LiDARs, IMUs, GNSS sensors to operate reliably in dynamic and adverse environments.
Biography:
Jin-Sun Park received the B.S. (Hons.) in Department of Electronic Engineering from Hanyang University, Seoul, South Korea, in 2014 and M.S. degrees in School of Electrical Engineering, from KAIST, Daejeon, South Korea, in 2016, and the Ph.D. degree in School of Electrical Engineering, from KAIST, Daejeon, South Korea, in 2021. From 2019 to 2020, he was a Research Assistant at HikVision USA, CA, USA. In 2021, he was a Post-doctoral Researcher at the Information and Electronics Research Institute of KAIST, Daejeon, Korea. Since 2021, he has been an Assistant Professor at the School of Computer Science and Engineering of Pusan National University, Busan, Korea. His research interests include Computer Vision, Deep Learning, Multi-Modal Sensor Systems, Depth Completion / Depth Estimation, Sign Language Recognition. He received Best Demo Presentation Award, International Workshop on Frontiers of Computer Vision (FCV) in 2018, Academic Achievement Award, Hanyang University in 2013.
Title: Federated Domain Generalization with Data-free On-Server Machine Gradient
Speaker: Professor Won Joo Hwang
Abstract:
Domain Generalization (DG) aims to learn from multiple known source domains a model that can generalize well to unknown target domains. One of the key approaches in DG is training an encoder which generates domain-invariant representations. However, this approach is not applicable in Federated Domain General- ization (FDG), where data from various domains are distributed across different clients. In this paper, we introduce a novel approach, dubbed Federated Learning via On-server Matching Gradient (FedOMG), which can efficiently leverage domain information from distributed domains. Specifically, we utilize the local gradients as information about the distributed models to find an invariant gradient direction across all domains through gradient inner product maximization. The advantages are two-fold: 1) FedOMG can aggregate the characteristics of distributed models on the centralized server without incurring any additional communication cost, and 2) FedOMG is orthogonal to many existing FL/FDG methods, allowing for additional performance improvements by being seamlessly integrated with them. Extensive experimental evaluations on various settings demonstrate the robustness of FedOMG compared to other FL/FDG baselines. Our method outperforms recent SOTA baselines on four FL benchmark datasets (MNIST, EMNIST, CIFAR-10, and CIFAR-100), and three FDG benchmark datasets (PACS, VLCS, and OfficeHome). The reproducible code is publicly available.
Biography:
WON JOO HWANG received the B.S. (Hons.) at Dept. of Computer Engineering, Pusan National University in 1998 and M.S. degrees at Dept. of Computer Engineering, Pusan National University in 2000, and the Ph.D. degree at Dept. of Information Systems Engineering of Osaka University, JAPAN, in 2002. From 2002 to 2019, he was a Professor at Inje University. Since 2019, he has been a Professor at Pusan National University, Busan, Korea. He received Best Paper Award, MITA, Korea Multimedia Society (KMMS) in 2006, Best Paper Award, Biannual Conference of KMMS in 2007, 2008,2012, 2013, Academic (Best Researcher) Award, KMMS in 2008, Contribution Award, IEICE NOLTA in 2018, Best Researcher Award, Pusan National University in 2022.