| Paper: | RS-2A.3 |
| Session: | Late Breaking Research 2A |
| Location: | Late-Breaking Research Area |
| Session Time: | Friday, September 7, 17:15 - 19:15 |
| Presentation Time: | Friday, September 7, 17:15 - 19:15 |
| Presentation: |
Poster
|
| Publication: |
2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania |
| Paper Title: |
Combining convolutional neural networks with a model of the dynamics of visual cortex |
| Authors: |
Grace Lindsay, Ken Miller, Columbia University, United States |
| Abstract: |
Convolutional neural networks can capture features of the representation of visual information that emerges in ventral visual stream. However these networks tend to lack certain biological details such as excitatory and inhibitory cells, recurrent connections, and computations implemented via dynamics. Here, we take an established model of visual cortical areas that includes these details, the stabilized supralinear network, and incorporate it into the convolutional layers of a convolutional neural network. The existence of recurrent horizontal connections in this network introduces dependencies between different feature maps at a given layer and can influence the development of feature selectivity. The model also offers a way to explore the dynamics of inter-area and communication and the extent to which dynamics in earlier areas impact classification at the final layer. |
