Equal predictions with less data
Research areas
External Member
Description
Abstract: We are in the process of building a neural network to class crop leaves based on key microscopic traits, a process which is very important to Cotton breeders for example. However, microscopic data is costly and rarely available in large numbers. The aim of this project will be to identify the data which contributes most to the predictions of this model, and to utilise this information, together with a suite of data augmentation techniques, to build a model which is less data-hungry but has similar predictive power so that it can be more broadly applied.
Contact:Vivien Rolland