|
Abstract:
|
The robustness of machine learning (ML) algorithms can be affected by a wide range of variables including the composition of classes in a training dataset, a change in distribution between the training and test datasets, the architecture of the algorithm, interactions between the classes of data and their environments, etc. In this talk I will review how experimental design can be used to explore the robustness of a machine learning algorithm and make improvements to the design of the algorithm. We will cover applications that show how to select optimal training datasets for robustness and how to understand potential performance drops when the test environment differs from the training environment. Applications for defining algorithms best operating region will also be highlighted.
|
