Abstract
The best performing computer vision systems are based on deep neural networks (DNNs). A study in this issue of PLOS Biology shows that DNNs trained on noisy stimuli are better than standard DNNs at mirroring both human behavioral and neural visual responses.
Highlights
The best performing computer vision systems are based on deep neural networks (DNNs), which often achieve or even surpass human performance on object recognition tasks
DNNs are currently championed as models of the neural processing underlying human object recognition, based on an observed correspondence between patterns of activity in DNNs and neural activity throughout the ventral visual stream [2]
The signal-tosignal-plus-noise ratio (SSNR) threshold for the noise-trained DNN reported by Jang and colleagues was slightly lower than that of the human viewers [3]
Summary
DNNs are currently championed as models of the neural processing underlying human object recognition, based on an observed correspondence between patterns of activity in DNNs and neural activity throughout the ventral visual stream [2]. In this issue of PLOS Biology, Jang and colleagues [3] explore one such divergence: noise robustness. These controversial stimuli, which confuse standard trained DNNs, are used to arbitrate among candidate computational models and to probe the mechanisms underlying noise robustness in human vision.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have