Developing new materials is an expensive and time-consuming process. AI can help to speed it up through powerful optimization methods and machine learning models that approximate the expensive processes involved. This tutorial will give an overview of applications of AI in materials science, highlighting challenges and opportunities for AI research.
Advanced materials and manufacturing methods are important drivers of today’s economy. They have made consumer devices like TVs and computers ubiquitous by enabling cheap mass production of reliable products and enabled industrial applications on a scale not seen before. There is a need for new materials in high-performance environments, for example in optoelectronics for computer and cellular technology, photovoltaics and consumer electronics, as well as for extreme environments in space and aviation applications. However, designing new materials is a slow and laborious process, requiring significant investments in capital and labor. Materials synthesis, development, and characterization are time-intensive and expensive processes; in particular as a large number of different alternatives have to be evaluated and optimized over. Often, there is little guidance from theoretical models, and synthesizing new materials to be tested is only possible on a small scale.
While models based on a small number of empirical observations have been used to speed up this process for decades, recent advances in AI and machine learning have ushered in a new era of effective and efficient modeling and exploration of large spaces of possible designs with complex performance landscapes. Powerful optimization techniques that make no assumptions about properties of the underlying process are able to quickly narrow in on promising designs. Process models induced by machine learning allow to approximate complex relationships reliably and predict properties of hypothetical materials. Despite a growing number of successful applications and large interest from domain scientists, the AI community has paid little interest to these important applications. This tutorial will provide an introduction and overview of this application area, and highlight challenges and opportunities for AI research.
The Optimization Simulators require R and the mlr, mlrMBO, smoof, rgenoud, DiceKriging, and randomForest packages. Please install these requirements before the tutorial if you want to follow along.