About me

I am a PhD student at the University of Maryland, Baltimore County, conducting research in machine learning, computer vision, and intelligent automation. I have over three years of combined academic and industry experience. Previously, I worked as a Graduate Research Assistant at PIC Lab, Kumoh National Institute of Technology, South Korea, where I contributed to research in industrial AI, multimodal learning, and warehouse automation. My research has been published in journals including IEEE Access, ICT Express, and IJSA, and conferences such as IEEE ICRA, IEEE ICMIC, IEEE ICC, and IEEE ICTC. Before transitioning to full-time research, I worked as a Software Engineer at Divine IT Ltd., where I developed enterprise software solutions.

Selected Publications

AI-Driven Digital Twin Framework for Warehouse Logistics Automation Optimization

IEEE ICRA Workshop 2026 [GitHub]

This paper introduces a dynamic warehouse slotting framework that integrates a Deep Q-Network (DQN) reinforcement learning agent into a FlexSim Digital Twin via AAS middleware. Powered by a hybrid CNN-BiLSTM-Attention architecture, the agent uses real-time operational data to optimize product placement, significantly outperforming static methods by reducing average picking time (29.66%) and travel distance (18.54%), while boosting total throughput (35.42%).

Diffusion-Augmented Industrial Anomaly Detection Leveraging Attention-Empowered Multi-CNN Transformer Fusion

IEEE Access [Paper] [GitHub]

The paper introduces MCTNet, a transformer-based fusion architecture for industrial anomaly detection. It extracts diverse feature representations using multiple attention-empowered CNN backbones, which are then integrated via a Transformer-based meta-learner. To address class imbalance and limited defect data, the framework employs a diffusion model for synthetic image generation combined with robust data augmentation techniques to enhance the model's overall generalization.

TMNet: Transformer-fused multimodal framework for emotion recognition via EEG and speech

ICT Express [SCIE, Q1, IF:4.6, AC: 26%] [Paper] [GitHub]

TMNet is an advanced multimodal emotion recognition system that combines speech and EEG brainwave signals. It uses a CNN-BiLSTM and BiGRU to extract distinct features from these inputs, and then employs a Transformer network with cross-modal attention to seamlessly fuse them, significantly improving the accuracy of emotion classification.

FMACNN: Federated multi-attention CNN framework for artificial image detection

Journal of Information Security and Applications [SCIE, Q1, IF:4.4] [Paper] [GitHub]

The paper introduces FMACNN, a federated multi-attention convolutional neural network framework designed to detect artificially generated images directly on edge devices. The architecture integrates multiple attention mechanisms to capture fine-grained synthetic artifacts while maintaining a lightweight structure. To ensure data privacy and enable decentralized scalability, the system is trained using a federated learning approach. Additionally, the framework is supported by the newly introduced RealAIGI dataset, which leverages a diffusion transformer to generate highly realistic synthetic images for robust model training.

BLIND-Twin: Blockchain-Assisted LLM-Based CDS for Digital Twin-Enhanced Industrial AIoT

IEEE ICC 2025 [Paper]

The BLIND-Twin framework secures Industrial AIoT networks by combining Digital Twins for threat simulation, an LSTM-AE and LLM pipeline for precise anomaly detection, and a permissioned blockchain for decentralized, automated responses. It delivers highly scalable intrusion detection, achieving up to 99.63% accuracy with minimal latency on real-world datasets.

Digital Twin-Driven Semi-Supervised Knowledge Distillation for HMC Monitoring on Edge Device

IEEE ICTC 2025 [GitHub]

The paper introduces a Digital Twin-Driven Semi-Supervised Knowledge Distillation (SSKD) framework designed for real-time monitoring of Human-Machine Collaboration (HMC) on edge devices. It utilizes an Unreal Engine 4 virtual environment coupled with ROS2-based communication to generate synthetic training data and enable semi-supervised pseudo-labeling via a teacher model. The knowledge is distilled into a lightweight student model, which is further optimized using post-training integer quantization to allow efficient deployment on resource-constrained embedded controllers.

Updates!

[June 2026] Paper accepted at the IEEE ICMIC 2026, Taiwan. [Best Paper]
[May 2026] Paper accepted at the IEEE ICRA 2026 Workshop, Vienna.
[April 2026] Journal Paper accepted in Journal of Information Security and Applications.
[February 2026] Graduated with Outstanding Research Achievement Award from Kumoh National Institute of Technology, Spring 2026.
[January 2026] Journal Paper accepted in IEEE Access.
[January 2026] Journal Paper accepted in ICT Express.
[December 2025] Paper accepted at IEEE ICAIIC 2026, Tokyo, Japan.
[November 2025] Paper accepted at IEEE ICCIT 2025, Cox's Bazar, Bangladesh.
[October 2025] Two papers accepted in KICS Fall 2025, Guyonju, South Korea.
[August 2025] Two papers accepted at IEEE ICTC 2025, Jeju Island, South Korea.
[June 2025] Two papers accepted at IEEE ICMIC 2025, Cebu, Philippines.
[May 2025] Three papers accepted in KICS Summer 2025, Jeju Island, South Korea. [Best Paper]
[April 2025] Journal Paper accepted in ICT Express.
[January 2025] Paper accepted at IEEE ICC 2025, Canada.
[December 2024] KICS Fall 2024 Gumi Industrial Complex Energy Self-Sufficiency Data Academic Paper Contest. [2x Best Paper]
[October 2024] Three papers accepted in KICS Fall 2024, Guyonju, South Korea.
[August 2024] Three papers accepted at IEEE ICTC 2024, Jeju Island, South Korea.
[May 2024] Four papers accepted in KICS Summer 2024, Jeju Island, South Korea.
[October 2023] Paper accepted at IEEE ICCIT 2023, Cox's Bazar, Bangladesh.
[December 2023] Reached 1600 Data Structure and Algorithm problem-solving milestones.
[October 2022] Paper accepted at IEEE ICCIT 2022, Cox's Bazar, Bangladesh.

Baltimore, MD, USA

Page updated

Google Sites

Report abuse