Self-driving materials science

A very quick overview of how machine learning can guide the search for new materials and accelerate their characterisation.

Transcript

1. SELF-DRIVING MATERIALS SCIENCE MACHINE LEARNING FOR CHEMISTRY
2. SELF-DRIVING MATERIALS SCIENCE HOW MANY MATERIALS ARE THERE? ▸ Naive estimates based on mixing elements: 4,139,315,402,300 ▸ Apply some chemical knowledge ▸ More reasonable, but un-reported combinations than there are stars in the universe CHEM, 2016, 1 , 617.
3. SELF-DRIVING MATERIALS SCIENCE MATERIALS DISCOVERY METHODS ▸ Materials development can be slow ▸ Trial and error screening is common ▸ Computational screening can reduce development costs
4. SELF-DRIVING MATERIALS SCIENCE COMPUTATIONAL MATERIALS DESIGN ▸ From quantum mechanics we can calculate the properties of a theoretical system with great accuracy ▸ BUT QM calculations are not cheap Η Ψ = Ε Ψ
5. SELF-DRIVING MATERIALS SCIENCE A JOB FOR MACHINE LEARNING Rapid machine learning Accurate Quantum Calcs ▸ ML calculations provide target compositions ▸ Screened 1,000,000+ oxide photocatalysts CHEM. SCI., 2018, 1, 1022
6. SELF-DRIVING MATERIALS SCIENCE ENHANCED CHARACTERISATION ▸ Operando X-ray tomogrpahy can follow chemistry in real time ▸ ML enabled image enhancement breaks the time vs space dilemma
7. SELF-DRIVING MATERIALS SCIENCE FIND OUT MORE NATURE, 2018, 559, 547. @keeeto2000 @ml_sci machinelearningmastery.com https://distill.pub/ linear-digressions data skeptic