Designing the Color of Metal Thin Films with AI... Pukyong National University Develops Color Prediction Machine Learning Model

Professor Seunghoon Lee's Team from the Department of Physics
Proposes a New Direction for Designing Material Properties Using AI

A machine learning model that designs the color of metal thin films using AI has been developed, drawing significant attention from the academic community. This research is being evaluated as a study that presents a new direction for designing material properties using AI.

The research team led by Professor Seunghoon Lee from the Department of Physics at Pukyong National University (President Bae Sanghoon) has developed a new physics-based machine learning model that precisely predicts the color of metal oxide thin films using artificial intelligence (AI).

This study is noteworthy in that it simultaneously improves learning efficiency and prediction accuracy by directly incorporating electromagnetic principles into the training process of the machine learning algorithm through a "kernel trick" strategy.

The color of metal oxide thin films changes depending on the surface microstructure and the degree of oxidation, allowing for the realization of a variety of colors by controlling these factors. However, it has been a challenging task to quantitatively predict the nonlinear correlation between process conditions-such as oxidation time, temperature, and thickness-and the resulting color.

Research Overview. (Top) Overview of the study on developing a color prediction model according to oxidation process conditions, (Bottom left) Experimental data used for training, (Bottom right) Color chart predicted by the machine learning model according to oxidation process conditions.

To overcome these limitations, Professor Lee's team utilized machine learning technology to find a method for incorporating physical principles within the model. By directly designing the algorithm's kernel function based on the electromagnetic properties of the data, they proposed a strategy that enhances both the efficiency of learning and the performance of predictions.

Professor Lee stated, "This is a case that demonstrates how integrating physical understanding into machine learning can simultaneously improve learning efficiency and prediction accuracy," adding, "The concepts and concrete examples presented in this study will provide a foundation for making machine learning more accessible and applicable across various academic disciplines."

The results of this research were published online in the international journal Materials Research Letters (JCR ranking 7.2% in the field of metal materials engineering) under the title "Optimizing a machine-learning model for color design of metal oxides/metal multilayers with physics-guided kernel trick."

This research was conducted as a collaborative effort between master's student Lee Dongik (sole first author) from the Department of Physics at Pukyong National University and the research team of Professor Seyoung Jung at Pusan National University.

Professor Seunghoon Lee (corresponding author, left), Professor Seyoung Jung (co-corresponding author).

This content was produced with the assistance of AI translation services.