Professor Kang Suk-ju’s Research Team from the Department of Electronic Engineering Wins Haedong Best Paper Award from the Korean Journal IEIE Transactions on Smart Processing & Computing
The research team led by Professor Kang Suk-ju (including Sogang University master’s program graduates Kim Hyun-sung, Kim Ki-hoon, and James Lee) in the Department of Electronic Engineering, in collaboration with NAVER CLOUD and Pusan National University, has received the Haedong Best Paper Award for their paper published in IEIE Transactions on Smart Processing & Computing. This award is given out only to one best paper and three excellent papers annually, selected through a rigorous review of all papers published over the past year.
The paper, titled “Enhanced Control of Human Motion Generation using Action-conditioned Transformer VAE with Low-rank Factorization”, suggests a methodology that uses a Variational Autoencoder (VAE) to generate human motion of a specific class while modifying only the desired body parts. The study applied Low-Rank Factorization, previously used in GANs, to a Transformer-based VAE, and addressed the issue of posterior collapse by introducing a sigmoid annealing scheduling technique. The research used a data augmentation method that randomly increases the range of motion in the input human motion data to amplify the degree of variation in generated motions.
Master’s program graduate Kim Hyun-sung, who participated in the research, commented, “It is an honor to have our long-standing research recognized by IEIE SPC and to receive the Best Paper Award. This achievement was possible thanks to Professor Kang’s meticulous guidance and the passion of my colleagues.”
▶Paper title: “Enhanced Control of Human Motion Generation using Action-conditioned Transformer VAE with Low-rank Factorization”
▶Journal title: IEIE Transactions on Smart Processing & Computing
▶Award information: Haedong Best Paper Award
▶Authors: Kim Hyun-sung (First Author, Sogang University), Kong Hyeong-bo (Co-First Author, Pusan National University), Kim Ki-hoon (Co-First Author, Sogang University), James Lee (Co-First Author, Sogang University), Cha Geon-ho (NAVER CLOUD), Jang Ho-duk (NAVER CLOUD), Wee Dong-yoon (NAVER CLOUD), Kang Suk-ju (Corresponding Author, Sogang University)
The paper, titled “Enhanced Control of Human Motion Generation using Action-conditioned Transformer VAE with Low-rank Factorization”, suggests a methodology that uses a Variational Autoencoder (VAE) to generate human motion of a specific class while modifying only the desired body parts. The study applied Low-Rank Factorization, previously used in GANs, to a Transformer-based VAE, and addressed the issue of posterior collapse by introducing a sigmoid annealing scheduling technique. The research used a data augmentation method that randomly increases the range of motion in the input human motion data to amplify the degree of variation in generated motions.
Master’s program graduate Kim Hyun-sung, who participated in the research, commented, “It is an honor to have our long-standing research recognized by IEIE SPC and to receive the Best Paper Award. This achievement was possible thanks to Professor Kang’s meticulous guidance and the passion of my colleagues.”
▶Paper title: “Enhanced Control of Human Motion Generation using Action-conditioned Transformer VAE with Low-rank Factorization”
▶Journal title: IEIE Transactions on Smart Processing & Computing
▶Award information: Haedong Best Paper Award
▶Authors: Kim Hyun-sung (First Author, Sogang University), Kong Hyeong-bo (Co-First Author, Pusan National University), Kim Ki-hoon (Co-First Author, Sogang University), James Lee (Co-First Author, Sogang University), Cha Geon-ho (NAVER CLOUD), Jang Ho-duk (NAVER CLOUD), Wee Dong-yoon (NAVER CLOUD), Kang Suk-ju (Corresponding Author, Sogang University)