I have a undergraduate student that has about a month do complete a research topic and she has decided that she would like to do something with 2D materials – probably the Jarvis_2D database. I am (relatively) new to matminer (and machine learning) myself and would like to offer her a tractable project. We have already gone through the 2020 matminer workshop together and she has a reasonable comprehension of machine learning. Can anyone offer any ideas for simple projects that would doable in a few weeks? I would like to find a topic that would both spur her interest and at the same time be tractable so she doesn’t get overwhemed as her background is more EE than physics (although she is a quick learner). Any suggestions would be gratefully received.
You could look at having her predict some interesting properties of the JDFT2D database either using the existing featurizers in matminer or by having her create her own (examining properties of symmetry, electronic configuration, local order, etc.).
AFAIK many of the featurizers in matminer have not been applied to 2D materials, yet they may be valid. The easiest/most readily available property for prediction would be exfoliation energy, though I’m sure there are others (e.g., in plane conductivity?) which might be interesting and available in JDFT as well.
If she does wind up using the JDFT2D dataset from matminer to predict exfoliation enthalpy, she might be interested in submitting to the matbench leaderboard for that task. Then she can compare against some of the leading (general purpose) algorithms and examine the properties that contribute to the property of interest.