Search Papers | Poster Sessions | All Posters
Poster B9 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
RTNet: An image computable model of human choice, response time, and confidence
Medha Shekhar1 (), Farshad Rafiei1, Dobromir Rahnev1; 1Georgia Institute of Technology
Convolutional neural networks show promise as models of biological vision. However, unlike humans, they are deterministic and use equal number of computations for easy and difficult stimuli, which limits their applicability as models of human behavior. Here we develop a new neural network, RTNet, that generates stochastic decisions and human-like response time (RT) distributions. Through comprehensive tests, we show that RTNet reproduces all foundational features of human accuracy, RT, and confidence and does so better than all current alternative models. We further test RTNet’s ability to predict human behavior on novel images by collecting accuracy, RT, and confidence data from 60 human subjects performing a digit discrimination task. The responses produced by RTNet for individual novel images correlated with the same quantities produced by human subjects and these correlations were higher than those produced by all competing models. Overall, RTNet is a promising model of human response times that exhibits the critical signatures of perceptual decision making.
Keywords: Deep neural networks reaction time sequential sampling perceptual decision making