An Open Combinatorial Diffraction Dataset Including Consensus Human and Machine Learning Labels with Quantified Uncertainty for Training New Machine Learning Models

Modern machine learning and autonomous experimentation schemes in materials science rely on accurate analysis of the data ingested by these models. Unfortunately, underlying can be difficult, even for domain experts, complicating training models intended to drive experiments. This is especially true when goal identify presence weak signatures diffraction or spectroscopic datasets. In this work, we examine a set as-obtained that track phase transition from monoclinic tetragonal Nb-doped VO2 film as function temperature dopant concentration. We then task experts with identifying which present each pattern manually algorithmically, respectively; both cases, labels vary dramatically, at boundaries. use mode Shannon entropy method capture, preserve propagate consensus their variance. Further expert benchmark demonstrate weighted scoring test performance generated labels. Finally, propose material challenge centered around generating improved labeling algorithms. real-world dataset curated act bed new The raw data, annotations code used study are all available online data.gov interested reader encouraged replicate improve existing