DESCRIPTION

In recent years, many fields of research have been revolutionized by a set of approaches and statistical methods commonly referred to as machine learning (ML) or artificial intelligence (AI).
In computational material science, AI approaches have been actively studied as promising candidates to obtain accurate predictions for large scale systems at an affordable computational cost. At this early stage of development, most efforts are in the direction of understanding how AI might be exploited, by investigating different approaches and providing benchmarks on known systems for known properties. [1-3]
At present, few AI packages are readily available and only a fraction of these is based on Neural Networks (NN). [4,5]
The know-how and the tools are centered around very few research groups and taking advantage of these resources is not straightforward for PhD students and researchers alike.
This event will be devoted to the exchange of know-how and ideas on AI techniques applied to material science, in a bilateral Italy-France perspective and in a friendly environment.
In a first moment the foundations of AI will be analyzed and establish and common language between participants will be established. The mathematical tools and ML approaches used to
create atomic potentials will be described. Particular attention will be given to NN approaches and the variety of geometrical and topological descriptors proposed for the description of atomic environments.
Later, the participants will attend a hands-on tutorial on PANNA (Properties from Artificial Neural Network Architectures). PANNA is a free and open source software developed at SISSA and CNR-IOM, based on the cutting edge TensorFlow framework, implementing several NN approaches applied to material modeling.[6]
A round table will close the study day. The participants will be invited to discuss how AI could benefit their research and how their research could stimulate new developments, in order to open and/or strengthen the collaborations between partner institutions and countries.

REFERENCES

  1. 10.1039/c1cp21668f
  2. 10.1103/PhysRevB.87.184115
  3. 10.1063/1.5019779
  4. 10.1038/s41586-018-0337-2
  5. 10.1557/mrc.2019.95
  6. 10.1016/j.cpc.2020.107402

8 December 2020

SISSA - Trieste, Italy