Settling in for another long night of recording...



(Dog Hates Me)

Possibly the coolest talk I heard at Cosyne, the annual computational and systems neuroscience conference, this year was by David Heeger, who's at NYU. Those of you who know my general distaste for functional MRI imaging -- or rather, the vast majority of experiments that make use of it -- will probably be somewhat surprised to hear this. The talk was on how different areas of the brain integrate information temporally. The expectation is that more peripheral areas, which are more directly connected to sensory inputs and motor outputs (e.g. primary visual cortex, primary auditory cortex, etc), will tend to reflect more of the moment-to-moment structure of the world, whereas higher processing centers have to integrate information over longer periods of time in order to make sense of what's going on. In the visual system, for example, the early stages of visual processing are responsible for the cinematography -- what's there, what direction is it moving -- while higher areas take care of the plot, the interactions between elements, and so forth.

A very elegant prediction of this hypothesis is that if you remove the longer-term temporal structure of a stimulus, lower areas should continue processing things as before, but the higher areas will fail to be engaged. Heeger and his colleagues tested this by measuring fMRI signals from subjects while they watched several versions of movies. One of the versions was played in the normal direction, one was played backwards, and several of the others were shuffled. The shuffled versions made use of the same cuts as the original version, but the order of the cuts was randomized, so in one scene you'd see Charlie Chaplin roller-skating in a department store and in the next he'd be working in a factory, and so forth. Heeger assessed how the brains of his subjects were processing this data by measuring the correlations between multiple presentations of the different movies. High degrees of correlation in a particular area indicate that the circuits in that area are processing the information the same way in each trial. In other words, the information in the stimulus is driving the activity in that area. A low degree of correlation, on the other hand, indicates that the region is not being driven by the stimulus. The results were more or less as expected. The normal movie produced activity that was highly correlated in almost all the areas involved in visual processing, whereas the time-reversed and shuffled movies produced low correlations in regions of the brain associated with processing higher levels of structure.

But the really interesting result came when they compared the responses across subjects. In other words, how similar is the activity in your brain when watching a movie that of the person sitting next to you? It turns out to depend, as in the previous result, on what areas of the brain you look at. But it also depends, to an almost shocking degree, on the movie. If you play subjects a video that doesn't have any plot -- a bunch of people walking through Central Park, for instance -- the degree of correlation is fairly low except at the earliest stages of processing. Some movies, like "The Good, the Bad, and the Ugly", produce much higher levels of correlation across a larger proportion of the brain. And some movies, like one directed by Alfred Hitchcock, induced high correlations across 70% of the brain.

It would be easy to overinterpret this finding, which Heeger doesn't do. Once it's published I expect the mainstream media will read that 70% figure as meaning that when you and your buddy watch Alfred Hitchcock your mental states are 70% the same. In order for that to be the case, the correlations would have to be perfect, which they're not. The low temporal and spatial resolution of fMRI imaging also means that the similarity is only on a fairly gross level. But even with those caveats, it's still kind of an incredible finding. The huge difference between the effects of an unstructured video and the effects of a Hollywood video indicate just how good the movie industry is at specifying the experience you're going to have when you go to a movie.

On my drive back from Salt Lake I happened to hear Bob Edwards interview Orhan Pahmuk, the Turkish author and Nobel Prize winner. Pahmuk made the observation that the rise of the novel coincides very nicely with the rise of nationalism. Novels are indeed highly cultural beasts. The best of them are highly idiomatic and difficult to translate. They are deeply embedded in the history of a place and a people. Movies are much more universal. Translation is much less of an issue, because so much of the content of a movie is in the cinematography, the acting, and the synergy of music and visual representation. Movies may be just as concerned with a time and place as a novel, but the language they use to discuss those concerns is entirely different. Pahmuk said that movies have a place in our new globalized culture analagous to that of the novel in the new nationalistic culture of the early nineteenth century. He seemed to see the ascendency of this art form as a good thing. I'm not so sure. I love movies, too, but there's something sad about the image of people around the world giving up music, verbal innovation, and the plastic arts to sit mutely in front of a DVD player, minds all running in parallel.

Update: The results have been published in the March 5 edition of the Journal of Neuroscience. Link here, although I don't think the full text is freely available yet.

Let me start by saying that although I dearly love philosophy, it is entirely as an amateur. I dabble in theory, I know its consolations, but ultimately am tied to the concrete realities of biology: the needs, natural history, and behavior of real animals. Perhaps the structure of these things reflects or embodies deeper realities -- Mind, Form, Being, what have you -- but I prefer to approach those realities through observation.

So it is a little strange to be up against a problem that has long been in the preserve of philosophers, theoreticians, and computer scientists. The question is, in empirical terms, how do animals recognize objects? I am using this term in what is essentially a grammatical sense: anything that could be the object of a verb. Animals have to recognize particular things (such as other individuals of one's species) as well as classes of things or phenomena that should be treated in the same way (a hawk's cry, a branch, a ripe fruit). We can leave aside the question of whether animals have concepts that correspond to these objects -- whether they ascribe intentionality to other animals, for example -- it is enough that they have distinct responses to particular things or types of things.

The problem is variation, which comes from many sources. The most obvious one is noise. The connection is bad or the TV is out of focus or the room is loud and dimly lit. But variation also comes from the nature of the object, as well. Faces appear from different angles, voices are modulated by emotion or employ different words, trees don't grow the same way, and pigeons come in different colors. It seems fairly obvious that pattern recognition is more than just matching sensory data to a stored picture or template. There have been many suggestions as to what it does involve, many of which turn out to work fairly well when implemented in software and given a relatively limited set of inputs over which to operate (1). But the question of how animals actually solve this task is still open. We know that there is a general increase in selectivity as sensory data is processed by the brain, and there are neurons in some areas of primate brains that seem to respond only to particular objects, but it is not know how that selectivity is built up from the interconnections of neurons, or how brains learn the features of a pattern in the first place.

That's what I'm working on, in case you were interested.

(1) I am not particularly interested, from a scientific standpoint, in whether such artificial systems will ever "outperform" biological ones. It is a little like asking whether one species of falcon outperforms another. Yes, a peregrine flies faster and takes larger prey than a merlin, but the two species occupy different niches. Performance is only an issue when animals are in competition. It is, perhaps, worth asking why so many people seem to think we are in competition with biological systems.

I'm in Vancouver, BC, for the big neuroethology conference. It's incredibly satisfying to hear people talking about real animal behaviors, especially when the animals are common model organisms. Yes, rats do interesting things, and place cells and barrel cortex are far more interesting than most people assume. My own behavior has been somewhat erratic as I managed to forget my poster in Chicago and had to make several trips to Kinko's to replace it.

Meanwhile I've gotten my first item of neuroscience spam. Just what would I do with a million neurons in some random tissue culture medium?

BrainBits LLC provides fresh (never frozen) brain tissue (Hippocampus or Cortex) from embryonic rats to you within twenty four hours at a cost of $99 for a million neurons. Broken down that is less then 10¢ per thousand neurons and this is guaranteed. The tissues you receive will come with instructions and everything you need to make cultures of primary neurons. How can we do this? The tissues are shipped in Hibernate E that will keep them alive for weeks under refrigeration. Hibernate A is also useful for the isolation of adult neurons or aid in keeping the neurons or brain tissue alive during electrophysiology or microscopy.
I believe BrainBits can help you in your research. If you are interested and want to find out more about other tissue offered or more information on our company click on to www.BrainBitsLLC.com