Department of Neurobiology
Insights into the single neuron in the age of generative networks
Visual cortex neurons are optimized to respond to natural images, realizing our ability to understand the world. What is the nature of this optimization? We are pursuing two overarching hypotheses about how neurons respond to natural images. One hypothesis is that in anterior visual areas, neurons encode complex features, such as objects or places. A second hypothesis is that these neurons encode niche features: attributes that are simpler than objects, just rarer and cleverly “chosen” to correlate with some object types. Do neurons have tuning for complex or niche features? In this talk, we will discuss experiments in monkey V1, V4, inferotemporal- and prefrontal cortex, using deep generative networks to identify each neuron’s critical features. We will explore similarities and differences between the visual system and artificial neural networks designed for visual tasks. Overall, our results suggest that much of our visual recognition relies on simpler features than previously believed.
A pizza lunch willl be served.