NamGyu Jung, SaeBom Lee, Chang Choi
ACM/SIGAPP Symposium on Applied Computing (SAC) 2024
We present a lightweight transformer model with shared embeddings between encoders and decoders, designed to enhance efficiency and address expression imbalance in construction safety prediction. The model reduces parameters by 48% compared to conventional transformers and improves performance by 4% over LSTM, enabling effective correlation analysis and deployment in edge computing environments.
NamGyu Jung, SaeBom Lee, Chang Choi
ACM/SIGAPP Symposium on Applied Computing (SAC) 2024
We present a lightweight transformer model with shared embeddings between encoders and decoders, designed to enhance efficiency and address expression imbalance in construction safety prediction. The model reduces parameters by 48% compared to conventional transformers and improves performance by 4% over LSTM, enabling effective correlation analysis and deployment in edge computing environments.
NamGyu Jung, Van Thuy Hoang, O-Joun Lee, Chang Choi
IEEE Internet of Things Journal 2024
We propose a context-aware hyper-personalized recommendation system for kiosk IoT devices, addressing data imbalance across domains with an efficient self-supervised learning method. The system demonstrated a 20% improvement in performance metrics and an additional 0.8% gain with self-supervised learning, ensuring high-quality recommendations and optimal resource usage.
NamGyu Jung, Van Thuy Hoang, O-Joun Lee, Chang Choi
IEEE Internet of Things Journal 2024
We propose a context-aware hyper-personalized recommendation system for kiosk IoT devices, addressing data imbalance across domains with an efficient self-supervised learning method. The system demonstrated a 20% improvement in performance metrics and an additional 0.8% gain with self-supervised learning, ensuring high-quality recommendations and optimal resource usage.
NamGyu Jung, JunHo Yoon, SaeBom Lee, PanKoo Kim, KiHo Lim, Chang Choi
International Conference on Research in Adaptive and Convergent System (RACS) 2023
We propose a multimodal fusion approach for news categorization, combining image and text data to enhance classification accuracy in digital journalism. Among the evaluated methods, early fusion achieved the best performance with 78.13% accuracy and an F1 score of 0.7810, demonstrating the effectiveness of integrating modalities.
NamGyu Jung, JunHo Yoon, SaeBom Lee, PanKoo Kim, KiHo Lim, Chang Choi
International Conference on Research in Adaptive and Convergent System (RACS) 2023
We propose a multimodal fusion approach for news categorization, combining image and text data to enhance classification accuracy in digital journalism. Among the evaluated methods, early fusion achieved the best performance with 78.13% accuracy and an F1 score of 0.7810, demonstrating the effectiveness of integrating modalities.