From neurons to brain waves

I was prompted by an article in ScienceNews to take a look at some of the work of Earl Miller, Mikael Lundqvist and their various colleagues; most recently that reported at reference 2 earlier this year but going back to reference 3 and beyond, a span of well over a decade. A little out of my comfort zone, but I was tempted to attempt some report.

I have been impressed by the way that it now appears to be possible to model the activity a patch of cortex and its neurons in some considerable detail and then to project that activity into the sort of electrical activity you might get at the surface, the sort of thing you can pick up with an EEG machine.

The detail includes the sort of physiological stuff illustrated above – which certainly impresses me – and is paid for by the large amounts of computing power needed to drive the model. So back in 2010 we are told that: ‘simulations were typically performed on 128 nodes of the Blue Gene/L computer at the Center for Parallel Computers at KTH. It took 81 seconds to simulate one second of network activity’. Where KTH is the KTH Royal Institute of Technology, Stockholm, Sweden (reference 7). And Blue Gene/L is a very big computer (reference 8).

The model includes neurons assembled into mini-columns and mini-columns assembled into maxi-columns, with patterns defined in term of the mini-columns and with this biologically plausible modular structure able to exhibit the sort of attractor behaviour with regard to these patterns that is thought to underpin working memory. With attractors on the back of a postage stamp at reference 5 and on the back of an envelope at reference 6.

The idea seems to be that you define a pattern by taking one mini-column from each maxi-column. Patterns are more or less mutually exclusive and they are sometimes orthogonal although I have yet to find out what that means. In this work, there are not that many of these patterns, say less than 10.

Then when things are quiet, none of the patterns are active. But when the system is stimulated in the right way, there is a bit of a tussle, a winner-take-all competition, after which just one pattern is completed and makes it to the top of the heap, is active. And might generate the sort of brain waves that an EEG machine might recognise as bursts of gamma activity against the background beta activity.

I see an analogy with more complex behaviour in the bush. One sees a fragment of something moving in the bushes, and one rapidly decides what animal the fragment belongs to, one rapidly completes the pattern. And given the need for speed in such circumstances in the past, is it any wonder that in our more civilised world we still tend to make decisions very quickly, on the basis of fragments, even though now we would often do better to take a little more time and trouble about it?

We find it hard to stay in unstable equilibrium, perched between two attractors. For earlier musings on this front, in connection with Chief Inspector Morse, see reference 9.

But I am attempting to run before Miller and his colleagues have learned to walk. So far, they have only modelled a small bit of the layer 2/3 cortex (out of the usual six) of a rat. Some way from a human brain.

Other points

The SPLIT simulator from KTH used for some of this work looks sophisticated enough to me. But there are plenty of other simulators out there, including the NEURON simulator from Yale (references 10 and 11), which looks even more sophisticated, has a fancy website and a big Google footprint. But I have no real idea how the two compare.

I took a look at all this from a slightly different point of view, back in the middle of 2017, at reference 12. I think that a point I made then, that these models do not do synaptic plasticity was wrong, or at the very least is wrong now. Some of these models certainly do learn by means of synaptic plasticity.

Trivia: the habit of removing spaces from but retaining the internal capitals in phrases like ‘ScienceNews’ seems to be spreading, having started, as far as I am concerned, with OneDrive and OneNote from Microsoft.

References

Reference 1: https://www.sciencenews.org/.

Reference 2: Gamma and beta bursts during working memory readout suggest roles in its volitional control - Mikael Lundqvist, Pawel Herman, Melissa R. Warden, Scott L. Brincat & Earl K. Miller – 2018.

Reference 3: Attractor dynamics in a modular network model of the neocortex - Lundqvist M, Rehn M, Djurfeldt M, Lansner A – 2006.

Reference 4: http://ekmillerlab.mit.edu/.

Reference 5: https://en.wikipedia.org/wiki/Attractor_network.

Reference 6: http://scholarpedia.org/article/Attractor.

Reference 7: https://www.kth.se/en. The KTH Royal Institute of Technology in Sweden.

Reference 8: http://www-03.ibm.com/ibm/history/ibm100/us/en/icons/bluegene/.

Reference 9: http://psmv3.blogspot.co.uk/2016/10/a-choice.html.

Reference 10: https://www.neuron.yale.edu/neuron/.

Reference 11: Systematic generation of biophysically detailed models for diverse cortical neuron types - Nathan W. Gouwens, Jim Berg, David Feng, Staci A. Sorensen, Hongkui Zeng, Michael J. Hawrylycz, Christof Koch & Anton Arkhipov – 2018.

Reference 12: http://psmv3.blogspot.co.uk/2017/08/big-brains.html.