Fig. 14: Advantage of Bayesian Active Learning over random selection of compositions.
From: Electronic structure prediction of medium and high entropy alloys across composition space
This figure compares the maximum NRMSE across the composition space of the CrFeCoNi quaternary system using two different sampling strategies. The first bar shows the result from a model trained with 20 compositions selected via Bayesian Active Learning. The second bar corresponds to one of three models trained on 20 randomly selected compositions; the error bars indicate the range of maximum NRMSE values observed across the three models. All models were trained using the same number of data points, demonstrating the improved accuracy achieved through Bayesian Active Learning.
