Random projections and kernelised leave one cluster out cross validation: universal baselines and evaluation tools for supervised machine learning of material properties
Samantha Durdy, Michael W. Gaultois, Vladimir V. Gusev, Danushka Bollegala, Matthew J. Rosseinsky
Abstract
Kernelised LOCO-CV can measure the extrapolatory power of an algorithm. Random projections are a versatile benchmark for composition featurisation.
Topics & Concepts
Machine learningArtificial intelligenceComputer scienceMetric (unit)Kernel (algebra)Predictive powerProcess (computing)Cluster (spacecraft)SkewFunction (biology)Range (aeronautics)Representation (politics)Data miningMathematicsTelecommunicationsEconomicsOperating systemEpistemologyMaterials scienceProgramming languageCombinatoricsPoliticsPolitical scienceBiologyComposite materialOperations managementLawEvolutionary biologyPhilosophyMachine Learning in Materials ScienceX-ray Diffraction in CrystallographyCrystallization and Solubility Studies