Charlie Fieseler: How to build a brain

A fundamental problem in neuroscience is how to design a brain to quickly achieve robust behavior while processing chaotic sensory input. Fortunately, neuroscience is undergoing a revolution due to large-scale high-resolution recording technologies. Recent research on such datasets suggests a general design principle: a distributed representation of instantaneous behavior across a large population of neurons, producing a low-dimensional manifold. Studying such principles in small C. elegans nematodes has unique benefits, because we can record the neuronal activity of almost every single neuron during natural behaviors.

I will review recent work on large-scale brain recordings and manifolds, with particular reference to brain-computer interfaces. I will next discuss the historical cross-pollination of the neuroscience and AI fields, and then discuss lessons learned from reading the mind of this animal. We have observed the manifold concept to be extremely important even in this minimal system, and it may have particular relevance for engineers designing embedded systems that act in real time. 

In addition, I will introduce the neural networks and techniques used to transform our 3d microscopy videos into clean time series. I will end with some directions of future work modeling the dynamics of such neuronal systems, combining a sprinkle of control theory with neural networks.

Charles Fieseler is a postdoc in the Zimmer lab at the University of Vienna. He did his Ph.D. in physics at the University of Washington, focusing on modeling neuronal activity using data-driven control theory. He is interested in developing practical pipelines for analyzing terabyte scale microscopy data, and in analyzing minimal brains and understanding their design principles.