ABSTRACT
Machine learning is being used in various scientific and technological aspects which demand high reliability and precision from the models that are employed. These models, however, are accompanied by an inseparable component, i.e., uncertainty. Uncertainty can be defined as something that is not certain or known and is inversely proportional to the precision of the system and depends on many factors such as inherent randomness in the data distribution, hyperparameters in neural networks, sampling errors, model misspecification, insufficient input data, and much more. Presence of high uncertainty lowers the confidence of the model over its predictions and further raises precision issues in its applications. Therefore, machine learning experts aim to reduce total uncertainty and balance it with precision. To achieve that, uncertainty has to be accurately estimated and discovered. Then, optimization under uncertainty can achieve required precision by using specific mathematical programming techniques and optimization algorithms. In this chapter, we aim to define uncertainty and highlight the progress that has been made in the optimization and quantification of uncertainty.
