Perception Science in the Age of Deep Neural Networks

“…Furthermore, and in direct connection with the present study, previous authors have argued that deep neural networks should prove useful for the study of perceptual inversion effects ( VanRullen, 2017 ). In our study, perhaps the most pertinent demonstration of the profitability afforded by this computational tool is the stratification of relevant effects across layers ( Figure 6 a); indeed, it is difficult to imagine how this type of analysis would have been possible using more conventional modeling tools.…”

“…Only by focusing on physiological knowledge will it be possible to keep the DNN technology in the research area of the biological brain. The present research result suggests that neuroscientists and psychologists should not underestimate the value of the method (VanRullen, 2017 ) to call “Reverse Psychology.” Various types of sensory illusions represented by visual illusions, at least, could be the lodestar that supports the validity of DNNs as a tool for studies of the brain.…”

“…While exploratory, our results also suggest that infants saw a greater proportion of hands around naming events, hinting that children may have been orienting towards their caregiver when they heard labels for objects. While this effect was not present for faces, other work (Yoshida & Smith, 2008;Yu & Smith, 2013, 2017, including Franchak et al (2018), has found that infants spend much more time looking at the toys vs. their caregiver's faces during these play sessions, and highlighted the importance of hand-following as a component of joint attention . However, given that there were only two possible referents in the room at a time-and one of them was always a familiar category-this particular play session did not present many opportunities where children would need to use gaze cues to disambiguate referents.…”