Machine learning for the discovery of extreme thermal materials

Play Video
Entry for:
Image1485080361?1485080361  2019 Pitch it Clever





1. Summary

We need materials that can protect us from very hot things, such as nuclear power plants, or materials that are able to detect very mild changes in temperature, such as very sensitive thermometers. Those thermal materials have interesting response to changes in temperature, which engineers measure in terms of what’s called the “specific heat capacity”. Material scientists, like myself, can accurately predict the specific heat capacity of a material by using the laws of quantum mechanics. For a given material, the calculation is complex, and would take days or even weeks if it is to be performed on a personal computer. With the availability of enormous material databases, one can - in principle - predict the specific heat capacity for millions of materials. With the available computational resources, this is a formidable task. Therefore, I resorted to machine learning to predict this quantity for any given material. Using only about 1,500 materials, I trained several machine learning models to predict this value. Surprisingly, the machines excelled in their prediction when tested, and they did so in less than a second. I am now predicting the specific heat for 29,000 other materials, and I will submit this work soon.



No discussion yet, be the first one to comment


I am passionate about the physics and chemistry of 2D materials, and how they stretch the boundaries of our understanding of materials and interfaces. I use various flavours of quantum mechanics, p...

Recent Voters