Brad Theilman

Home

Blog

Projects

Notes

Links

Publications


Twitter

Github

Pushing 'trodes

29 Jan 2019

Today I’m jumping back and forth between working on my website and doing a recording from a starling’s brain. The kinds of experiments we do have lots of downtime during certain parts while you wait for some step to complete.

As I’m writing this, however, I’m about to enter the meat of the experiment where I won’t be able to focus on much else.

The experiment begins by opening a craniotomy over the brain nucleus of interest. Today, I’m aiming for a region called NCM. This nucleus is a secondary auditory region that does not receive direct auditory input from the brainstem or cochlea, but from “lower” regions in the pallium like Field L. So the information has already been “processed” for simpler features. NCM is a region that has been implicated in representing higher-order structures in the acoustic stimuli.

We think that the pattern of activity of the neurons in NCM, in response to a complex acoustic signal, yield “representations” of the information contained in that signal.

The real mystery, though, is how to map patterns of activity to “representations,” loosely defined.

This is where topology may be important. One of the key insights of Curto and Itskov, 2008 was that the only information accessible to neurons in a neural system is the activity of other neurons. This fact seems obvious but it has important implications.

Since neurons only have access to their own activity and its relationship to other neuron’s activity, then the representations we are interested in must be constructed solely from those relationships. The method invented in Curto and Itskov 2008 was precisely a method to map temporal relationships between the activity of different neurons to simplicial complexes. In this way, the topological structure of the simplicial complex reflects the relationships from which representations must be built. Therefore, analyzing the structure of the simplicial complex amounts to analyzing the neural representations. At least, that’s the idea.

Previous data we’ve collected indicates that neural activity in NCM produces “interesting” topological structure. But all of that data was collected using relatively short, 6-second-long stimuli. The purpose of the experiment today is to collect data from NCM neurons while the animal is exposed to longer, ~60 second long stimuli.

We need this data to understand how the topology of the neurally-derived simplicial complex evolves over longer time scales. We are curious about this question because in our first dataset, the topology did not appear to reach a “steady state” - and so we want to find out what happens next!