Feature Visualization in Artificial and Biological Neural Networks

The use of neural networks in neuroscience is often met with two criticisms: that neural networks are 'black boxes' that defy interpretation, and that they have not proven useful for studying their biological counterparts. An emerging technique to understand networks and apply them to neuroscience is image synthesis. In this session, we will discuss and play with gradient-based methods for synthesizing images that maximally drive neurons, both in artificial neural networks and in the primate brain. Graduate students and postdocs are welcome to attend. Presented by Eshed Margalit.