Date published: 1/10/2019

A study by Professors Philip Kellman and Hongjing Lu, graduate student Nicholas Baker, and postdoc Gennady Erlikhman, published in the journal PLOS Computational Biology, is being prominently featured on the UCLA Newsroom.

The researchers explain how artificial intelligence systems can be easily fooled. The team looked into the best object recognition systems in artificial intelligence, known as deep learning computer networks, and asked how closely these machines resemble the human brain. From a series of experiments, the team concluded that humans recognize objects based on their shapes, while deep learning networks respond to fragments of objects, and surface textures, but do not have any access to overall shape. To learn more, the UCLA Newsroom article is available at http://newsroom.ucla.edu/releases/can-artificial-intelligence-tell-a-polar-bear-from-a-can-opener, and the PLOS Computational Biology article is available at https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006613.

The research has also received attention from other websites (listed below):

Categories: 
Faculty