Computer scientists from the Netherlands and Spain have determined how a deep learning system well suited for image recognition learns to recognize its surroundings. They were able to simplify the learning process by forcing the system’s focus toward secondary characteristics.
The researcher Estefania Talavera Martinez, lecturer and researcher at the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence of the University of Groningen in the Netherlands, wanted to understand why errors in AI classification arise. She studied the use case scenario of recognizing food encounters, and soon found out that the images were not scanned thoroughly for clues. Therefore, her team came up with a solution that distracts the CNN from its primary targets. After the first successful detection, the target part of the image is blurred and the system is retrained. The methodology is less time consuming, giving better classification results.
Source (University of Groningen. “Artificial Intelligence learns better when distracted.” ScienceDaily. ScienceDaily, 29 July 2021.)
Original paper: Morales, D., Talavera, E. and Remeseiro, B., 2021. Playing to distraction: towards a robust training of CNN classifiers through visual explanation techniques. Neural Computing and Applications, pp.1-13.