English
Menu

Computational materials science

In materials informatics, when a machine learning model is built from data on existing materials to predict the properties of new materials, which caution is correct?

  1. AThe smaller the prediction error on the training data, the more accurate the predictions for new materials are sure to be, so minimizing the training error should be the only goal
  2. BPerformance should be judged by the error on data not used for training (e.g. by cross-validation), and predictions far outside the training data's range are unreliable
  3. CEven with little data, adding more and more descriptors (features) adds information, so the predictions for new materials are sure to become more accurate
Show answer and explanation

Answer: B. Performance should be judged by the error on data not used for training (e.g. by cross-validation), and predictions far outside the training data's range are unreliable

A machine learning model learns a function that predicts a target property from descriptors (features) that express composition, crystal structure and so on as numbers. If it is fitted too closely to the training data (overfitting), the error on that data is small but predictions for new data are poor. The data are therefore split into training and validation sets, or cross-validation is used, and performance is evaluated by the error on data not used for training. Adding too many descriptors when data are scarce makes overfitting more likely. Models are also good at interpolating within the range of the training data but poor at extrapolating far outside it. For this reason, data are increased efficiently by methods such as Bayesian optimization, which uses both the predicted value and its uncertainty to choose the next candidate to try.

Report an error

Practice this topic Back to topic

More Computational materials science questions

  1. Which pairing of a materials computation method with the range of space and time it can usually handle is most appropriate?
  2. Which statement about the ideas behind density functional theory (DFT) is correct?
  3. Which is a correct tendency of the results of standard DFT calculations using LDA or GGA exchange-correlation functionals?
  4. A 108-atom supercell (with periodic boundary conditions) of a metal calculated with DFT has a total energy of −405.00 eV. Removing one atom to create one vacancy and relaxing the structure gives −400.10 eV. Which is closest to the vacancy formation energy calculated as E_v = E(N−1) − [(N−1)/N]E(N)?
  5. In molecular dynamics, you want to integrate the equations of motion with a time step of 2 fs to calculate atomic motion over 1 ns. How many time steps are needed?