DJ Strouse

Project Title: Reliable Brains from Unreliable Neurons – The Search for Synfire Chains in the Brain
Advisor: Professor Kwabena Boahen (Stanford University)
Collaborator(s): Peiran Gao (Bioengineering Graduate Program), Nick Steinmetz (Neuroscience Graduate Program)
Sponsorship: Stanford Amgen Scholars Program
Timeline: June 2010 to August 2010

For Grandma and Einstein*:
    Information in the brain is transmitted by cells called neurons via electrical and chemical signals. Like any machine, man-made or biological, neurons are not perfectly reliable. First, their responses to identical signals from nearby neurons presented at different times may yield subtly (and sometimes drastically) different responses. The sources of this “noise” in the response of a neuron are being studied but are not yet well understood. Second, neurons die throughout the life of an organism and are generally not replaced. The combined effect of these two sources of unreliability is that the brain must be organized in a way to perform reliable computation using unreliable components. Theoreticians (Abeles 1982) proposed a neural network structure called a synfire chain in order to solve this problem. A synfire chains is a fully-connected, feed-forward network of neurons. That is, a synfire chain consists of a sequence of groups of neurons in which each neuron in one group is connected to all of the neurons in the next group (see video, slides, and poster below for pictures). The highly dense and redundant connections in synfire chains enable them to reliably and synchronously propagate signals. There has been significant research about the computational benefits of synfire chains, but relatively little on how they might develop within real brains and whether they actually do.
    During the latter half of summer 2010, I worked with Kwabena Boahen’s Brains in Silicon group through the Stanford Amgen Scholars program. In order to guide the experimental search for synfire chains, I studied biologically realistic neural network models to determine suitable conditions for the existence of synfire chains. Our goal is to guide the experimental search for synfire chains by making predictions of their signature activity patterns as well as which regions of the brain are most likely to exhibit them. I sought mathematical and theoretical guidance from graduate student Peiran Gao and consulted frequently with experimental neuroscience student Nick Steinmetz to ensure the biological plausibility of my models. Through a combination of graph theory, probability theory, and an approach inspired by mean field theory (a technique used in statistical and solid state physics), I found that learning rules and past activity play a crucial role in the development of computationally useful synfire chains and presented my findings at the Stanford Amgen Scholars Symposium (video, slides, and poster available below). In future work, Peiran and I would like to investigate more realistic network connectivities and learning rules and compare the predictions of the model to recent experimental studies purporting to have discovered evidence of synfire activity (Ikegaya et. al. 2004).
    This work stressed to me the importance of collaboration with experimentalists, representing a stark difference between my quantum information and neuroscience projects. While the well-developed theoretical structure of modern physics enabled me to work independently of ongoing experiments and still make progress, neuroscience lacks such structure. Rather than an obstacle, however, I believe this dearth of theory represents an opportunity, one that I intend to spend at least the next several decades pursuing. Throughout graduate school and my career, however, I intend to emphasize theoretical and computational work done in tight collaboration with ongoing experiments.

Symposium talk video: available soon
Symposium talk slides (with speaker notes): available for download here
Symposium poster: available for download here