Quantifying Uncertainty in Ensemble Deep Learning

Neural networks are an emerging topic in the data science industry due to their high versatility and efficiency with large data sets. The purpose of this modern machine learning technique is to recognize relationships and patterns in vast amounts of data that would not be explored otherwise. Past research has utilized machine learning on experimental data in the material sciences and chemistry field to predict properties of metal oxides. Neural networks can determine underlying optical properties in complex images of metal oxides and capture essential features which are unrecognizable by observation. However, neural networks are often referred to as a “black box algorithm” due to the underlying process during the training of the model. The explanation for a prediction is unable to be traced, therefore poses a concern on how robust and reliable the prediction model actually is. Building ensemble neural networks allows for the analysis of the error bars of the prediction model. The project’s objective is to determine the comparative differences between the predictive ability of each individual convolutional neural network versus the ensemble neural network. Additionally, the paper explores how to use the ensemble model as a method of uncertainty quantification. Overall, ensemble neural networks outperform singular networks and demonstrate areas of uncertainty and robustness in the model