Carsen Stringer: Simplified models of high-dimensional visual cortical circuits

Janelia Research Campus, Howard Hughes Medical Institute

Visual sensory input is a complex spatiotemporal stream of information. How does the brain compress such high-dimensional visual input into learnable abstract features? We characterized the structure of neural responses to visual stimuli, in both mice and monkeys. We introduce a new class of simplified ANN models that can predict over 70% of the response variance of V1 neurons. We found that ANN models required only two convolutional layers for good performance, with a relatively small first layer. Models with this type of architecture performed well on texture and object classification tasks. We further found that we could make the second layer small without loss of performance, by fitting individual "minimodels" to each neuron. These minimodels can be used to gain insight into how stimulus invariance arises in biological neurons. In parallel, we quantified the multi-timescale dynamics of ongoing cortical activity. These dynamics could be explained by a linear dynamical system with random symmetric connectivity. This dynamical structure may provide a scaffold for integrating information from the sensory environment across time.

 

The talk will be streamed live into our lecture hall!

 

Guests are welcome!

 

Organized by

Torben Ott / Lisa Rosenblum