Training AI using biological experts

 

Lead researchers: Dr Michael Pound, Dr Andrew French, Dr Rumiana Ray

Humans can learn to become experts at identifying crucial pieces of information on images and understanding their meaning. In plant phenotyping, scientists can easily identify disease symptoms on a picture of plant, and quantify the severity of infection. Artificial intelligence (AI) has the potential to revolutionise bioimage analysis, but training these methods on these images can be challenging if we can’t describe what the expert is looking for.

This project will develop a new way to train artificial intelligence methods by studying the way expert scientists read images. Using eye tracking technology to record their gaze, tracking where an expert looked on an image while examining for disease or other characteristics. The AI will learn both what to look for, and where in the image to look. By examining where experts look we hope to reduce the time required to annotate datasets, and will produce even more accurate AI results.

This project is funded by BBSRC.