| Paper: | RS-2B.1 |
| Session: | Late Breaking Research 2B |
| Location: | Late-Breaking Research Area |
| Session Time: | Friday, September 7, 19:30 - 21:30 |
| Presentation Time: | Friday, September 7, 19:30 - 21:30 |
| Presentation: |
Poster
|
| Publication: |
2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania |
| Paper Title: |
Visual search behavior of neural networks |
| Authors: |
David Nicholson, Emory University, United States |
| Abstract: |
The visual search paradigm, in which subjects search for a target among distractors, forms the basis of most theoretical models of visuospatial attention. Many consider convolutional neural networks to be useful models of the visual system, which raises the question of how they behave in the visual search paradigm, especially in comparison to humans. I reproduce and extend results from https://arxiv.org/abs/1707.09775, measuring the accuracy of AlexNet in the visual search paradigm. I then propose a method to measure the reaction times of spiking convolutional nets, by adding a winner-take-all decision module (implemented with https://www.nengo.ai/nengo-dl/). Initial results using MNIST as a benchmark suggest this method will work. I will present experiments in progress on the reaction times of this spiking network in the visual search paradigm, and compare with human subject data (e.g. http://search.bwh.harvard.edu/new/data_set_files.html). Finally I will discuss how the results align with current theoretical models of visuospatial attention. |
