See for more.

I think the birds singing thing is pretty exciting; there's apparently only a couple of nuclei involved in singing the song. I think it may be the first place where they figure out the actual low-level representation of a memory of a sequence (although not for many years).

more on my research interests:

"good enough" models. When you model something, you have to decide the granularity of the model; e.g. if you are modeling the brain, do you model each molecule? each cellular channel? each dendritic spine? each cell? each nucleus?

e.g. if you model evolution in an ecosystem, do you model each atom? each base of DNA? each gene? each animal?

Any description of a system is a kind of model. By choosing a granularity, you are choosing to ignore events beneath that level of description. This goes against the grain of some people, but it cannot be gotten around. You cannot describe a system without purposefully ignoring things.

Neuroscience is a field where the appropriate granularity for various models is a contentious issue. The problem is that we don't know enough about computation in the brain to know what phenomena we can ignore.

I am interested in determining what one can ignore in neuroscience models. I am interested both in how one should decide what to ignore, and also in actually constructing models which ignore the right things,

A good example of how to know when something can be ignored is Abarbanel's lab's lobster stomatogastric ganglion project. The Abarbanel lab created a model of this CPG. Then, to test their model, they constructed robotic cells which realized the model. They killed some cells in the real CPG and replaced them with robotic mimics. Then they verified that the hybrid CPG still functioned normally. This allows them to infer that, yes, their model captured all of the "important" parts of the real system, and that everything that they ignored while constructing their model was not crucial to the description of the system.


more on models:

here's a research programme for modeling the brain. we need to find out what level of granularity to model the brain at. Here are some criteria:

We want to find a level of modeling description such that:

interested in spike timing dependent plasticity

--- put up CV