제목

Generation Research Breakthroughs and Highlights (Jin-Hwa Kim)

작성자

첨단컴퓨팅학부

작성일

2025.01.13

최종수정일

2025.01.13

분류

세미나

링크URL

게시글 내용

일시: 2025. 1. 16.(목) 11:00

장소: 제4공학관 D408

Title: Generation Research Breakthroughs and Highlights

Presenter: Jin-Hwa Kim 

Leader of Generation Research and Neural 3D TF at NAVER AI Lab

Guest Assistant Professor at the Artificial Intelligence Institute of Seoul National University (SNU AIIS)

Abstract: In this talk, I will present recent works from the Generation Research Team at NAVER Cloud AI Lab. Our team aims on advancing 3D representation learning, multimodal diffusion models, and visual generation using large language models (LLMs), with a strong emphasis on ensuring safe generation practices. I have selected three key highlights from our achievements over the past year. First, "Direct Unlearning Optimization for Robust and Safe Text-to-Image Models (Park et al., NeurIPS 2024)" introduces an innovative approach for unlearning unsafe generations by leveraging SDEdit’s controlled generation framework alongside direct preference optimization (Rafailov et al., 2023). Second, "Effective Rank Analysis and Regularization for Enhanced 3D Gaussian Splatting (Hyung et al., NeurIPS 2024)" demonstrates that effective rank (Roy and Vetterli, 2007) is a powerful metric for analyzing the shape characteristics of Gaussian primitives. Our proposed effective rank regularization elegantly mitigates needle-like primitives and significantly improves the quality of mesh extraction. Finally, "Polyhedral Complex Derivation from Piecewise Trilinear Networks (Kim et al., NeurIPS 2024)" explores a novel approach for analytically extracting meshes directly from the pretrained parameters of NeRFs, inspired by geometric insights into neural networks. Remarkably, this method outperforms the conventional Marching Cubes algorithm under the eikonal constraint in SDF-learned networks. Through this talk, I hope to convey our passion for pursuing cutting-edge research in visual generation and inspire opportunities for diverse research collaborations.

Bio: Jin-Hwa Kim has been the Leader of Generation Research and Neural 3D TF at NAVER AI Lab, working since August 2021, and Guest Assistant Professor at the Artificial Intelligence Institute of Seoul National University (SNU AIIS) since August 2022. He has studied multimodal deep learning, multimodal generation, ethical and safe AI, and other related topics. In 2018, he received a Ph.D. from Seoul National University under the supervision of Professor Byoung-Tak Zhang for the work on “Multimodal Deep Learning for Visually-grounded Reasoning.” In September 2017, he received 2017 Google Ph.D. Fellowship in Machine Learning, Ph.D. Completion Scholarship by Seoul National University, and the VQA Challenge 2018 runners-up at the CVPR 2018 VQA Challenge and Visual Dialog Workshop. He was Research Intern at Facebook AI Research (Menlo Park, CA) mentored by Yuandong Tian, Devi Parikh, and Dhruv Batra, from January to May in 2017. He worked for SK Telecom (August 2018 to July 2021) and SK Communications (January 2011 to October 2012).