17 - A glance under the hood: visualizing what artificial intelligence algorithms are using to make their amazing (and sometimes not so amazing) species identification predictions.

Abstract: Mosquito species identification algorithms focused on images have proved to be extremely promising over recent years. And while the results are sometimes stunning, we often find ourselves wondering what the AI models are looking at to draw their conclusions. Here we take a peek under the hood at these powerful tools often described as ‘black box’ algorithms. Using combinations of GradCam, Guided back propagation, and other visualization techniques, we will provide visual representations of the model focus for its decisions, discussing both correct and incorrect predictions.