I'm a Research Scientist in Computer Science specializing in Document AI and Machine Learning at various research institutions. I previously did research work presented at major conferences including EMNLP, ICDAR, BMVC, and ICPR, collaborating with researchers from multiple institutions to advance the field of document intelligence.
My work focuses on developing novel deep learning architectures for information extraction from business documents, with particular emphasis on handling visually-rich documents like forms, invoices, and receipts. I've published several influential papers on end-to-end information extraction methods, including the Multi-Stage Attentional U-Net architecture which achieved state-of-the-art results while using 40% fewer parameters than previous approaches.
Recently, I've been working on safety-critical AI applications, developing the IncidentAI dataset for preventing industrial failures through better natural language processing of incident reports. My research also explores confidence measurement in AI systems, hierarchical relation extraction, and improving document understanding with limited training data through reinforcement learning techniques. I am additionally interested in large language models, agent systems, and applying them to real-world business problems.
My work focuses on developing novel deep learning architectures for information extraction from business documents, with particular emphasis on handling visually-rich documents like forms, invoices, and receipts. I've published several influential papers on end-to-end information extraction methods, including the Multi-Stage Attentional U-Net architecture which achieved state-of-the-art results while using 40% fewer parameters than previous approaches.
Recently, I've been working on safety-critical AI applications, developing the IncidentAI dataset for preventing industrial failures through better natural language processing of incident reports. My research also explores confidence measurement in AI systems, hierarchical relation extraction, and improving document understanding with limited training data through reinforcement learning techniques. I am additionally interested in large language models, agent systems, and applying them to real-world business problems.
Publications
PCA-based 3D Facial Reenactment From Single Image
N. T. Dat, Tuan-Anh Dang Nguyen, D. V. Sang
International Conference on Multimedia Analysis and Pattern Recognition 2020