09/18(금) 나노융복합산업기술 특강은 Virtual Lab 박민규 박사님을 모시고 진행합니다.

관심있는 분들의 많은 참석 부탁드립니다.

ㅁ주제:  The Role of Materials Data in Accelerating AI-Based Scientific Discovery

ㅁ일시: 09/18(금) 15:00

ㅁ장소: 제2종합연구동 83188호

ㅁ약력: 

 교육 : 

• 2012 – 2019 UST 물리학 박사 (2019.02), 지도교수 : 김용성 

            - 최종학위논문: “Computational Study on Thermal Properties of Nano Materials” 

• 2004 – 2012 서울시립대학교 신소재공학사 (2012.02) 

  연구 경력: 

• 2020 – 현재 버추얼랩 (전략/과학 최고 책임자, 부사장) 
• 2019 – 2020 KIST 계산과학연구센터 (박사후연구원) 
• 2012 – 2019 한국표준과학연구원 (Research Scientist) 

  연구 분야: 

  1. First-principles calculations of phonon and thermal properties of 2D materials 

  2. Molecular dynamics calculation on lattice thermal conductivity 

  3. Development of machine learning (deep learning) potential  

ㅁ초록: 

The advancement of AI in materials science fundamentally depends on the availability and quality of ground truth materials data. While modern machine-learning models promise accelerated prediction and design, their reliability remains limited without experimentally validated, well-structured data. This work emphasizes that robust AI-driven discovery requires systematic collection of high-fidelity materials data spanning synthesis conditions, characterization results, and performance metrics. Traditional materials research often suffers from incomplete, inconsistent, and non standardized datasets, which restrict the effectiveness of simulation-based models and hinder reproducibility. By contrast, establishing a coherent data infrastructure—supported by ontology-based organization and rigorous data provenance—enables accurate model training, trustworthy property prediction, and meaningful inverse design. We highlight the role of active learning, surrogate modeling, and machine-learning interatomic potentials as methods that benefit significantly from high-quality ground truth data. Additionally, large language models contribute by extracting structured knowledge from literature, yet still rely on experimental validation to ensure scientific rigor. As a practical realization of these principles, D³Square is presented as an integrated platform designed to operationalize ground truth–centered research workflows. The system enables AI ready data generation, experiment-to-database automation, no-code model development, and active-learning–based experimental design. Through structured data pipelines and materials ontology integration, D³Square provides a scalable foundation for reliable AI applications in materials science.