March 16, 2023: New Research from "Mathematics of Neural Networks" theme

Consider facial recognition technology.  Computers may be taught to "see" or identify faces, but to what extent can they understand what they are seeing? Western Fellow Dr. Marieke Mur and her co-authors investigate how and why AI technologies fail to replicate human systems of visual cognition.  As Mur, who is an assistant professor in Psychology and Computer Science, explains

We show that humans and deep neural networks, which are popular computational models of the visual system, differ in how they represent the outside visual world. Our results question the commonly held assumption that these networks are good models of human vision. The discrepancies that we discovered have implications for neuroscience - they help us better understand how the human brain processes visual information - as well as artificial intelligence - they provide guidelines on how to improve artificial systems by taking inspiration from the human brain.

Read about it in the Western News: https://news.westernu.ca/2023/03/deep-learning-human-vision/

Read the study here: https://www.jneurosci.org/content/43/10/1731

 

Image credit:  https://unsplash.com/photos/11KDtiUWRq4