Facing the Music

OK, this post is just an addendum to Elias's report, live from the scene at ARO. I should add here that me and Elias have long discussed the 'special' nature of music and speech. I'd like to believe that audition evolved first as a distant-stimulus sensor and communication sounds evolved later to exploit this. This fact may have profound influences on the way we think about auditory memory, for example. Mort Mishkin gave a talk at APAN last year (my pre-blog days, also known as the Dark Ages) where he put forward a hypothesis that nobody really liked. From lesion experiments of multiple areas in monkey (including the medial temporal lobe, entorhinal cortex and the amygdala), Mishkin thought that maybe auditory long term memory did not exist in the same functional form as visual long term memory (in the sense that a temporal sequence of activity being stored to remember sounds). Mishkin did not offer alternatives, but I think it is clear this suggests some associative form.

Our intuition here is that speech and music may share common features as both involve explicitly remembering prosody - longer temporal features. I'm a musician (sort of), and I tend to remember Gilmour's searing solos in a lexical way that's more like remembering speech (actual speech itself, not an abstracted version of the content/meaning of speech). I think this may be a somewhat general principle... that speech and music are remembered by the way they are produced.

On the other hand, when I try and remember my auditory environment, I tend to describe it in terms of objects present and the sounds they generated... like a clock ticking or a ventilation duct's noisy humming. In this case the memory is more abstract - the actual temporal sequence of sounds are associated with the source and the source and it's location are remembered. It seems to be a more efficient way of doing things, and may explain what Mishkin saw. It may be relatively easy to do an fMRI study to test this, if done right, it could go a long way.

While the front-end mechanisms for speech/music/sounds may be the same, there may be important differences over how these stimuli are treated at higher stages of processing, and so the transitive logic Elias was talking about may not hold at several levels. While I'm not against studying the "representation" of music in its own right (and it should be studied), I do come out strongly against suggestions that it is a way to study auditory processing - see Elias's report on Griffith's talk, for example.

But the bottom line is: audition, independent of speech and music does exist... and needs to be studied too.
[+/-] show/hide

The Myth of Chamber 3 dispelled

Well, Chamber 3 (code-name for the new experimental set-up I was working on) is finally operational. Here are the first spikes we digitized... using a friend's implanted animal. Yeah, signal-to-noise is a bit lower than usual, that's because the electrodes are a lower impedance - the chronically implanted type. With our usual high-impedance setup, we should be in for some big spikes.



No setup story is of course complete without a narration of voodoo-ing away the 60 Hz. This time, it was holding up an unconnected alligator clip in the air and chanting 'I think we should connect this to...'. I swear it had no rational explanation.
[+/-] show/hide

Let's get specific

OK then, if you've been reading our blog so far, you're probably wondering what the heck we're talking about. You've either read recent papers coming out of auditory cortex that suggest A1 is 'cracked', or you're sick of the cutthroat competition in vision and the grass looks greener over here. We've kept the issues pretty general so far, but I can't take any more ambiguity: so I decided (after winning Elias's vote) that I'd post two slides I used in my thesis committee meeting.


(click on figure for a bigger version)

This slide pretty much sums it up. After 26 years of research, the only thing we know for certain about the primary auditory cortex is that it is tonotopically organized. It is either testament to the fact that 'audition is more difficult than vision', or, what lack of theory can do to you.

[+/-] show/hide

Let's get specific 2

Here's the next slide:



This slide is meant to illustrate all the progress we've been making in the past few years. I believe these data are a fair portrayal of the current state of the field. Data includes stuff from my own lab, and other leading labs in the field.

These data illustrate: synchronizing and non-synchronizing responses, low- and high-contrast responses, temporal and rate coding of sAM frequency, selectivity to ramped/damped stimulus waveform, presence of multi-peaked tuning, the integration time constant of neurons indicating their 'role in object processing', responses to low-probability sounds, synchronized onset responses, STRF-type mehtods providing descriptions of linear receptive fields and the mapping of arbitrary model parameters on the cortical surface.

It is becoming increasingly difficult to synthesize all these data (which on their own, are pretty well-done studies) into a cohesive picture. This lack of such a unifying picture is a matter of concern in certain circles in the field.

To a new student/uninitiated theorist, or even a relatively experienced person like myself, a read of current literature is very confusing, and any theorist attempting to understand and synthesize all the above data is bound to quickly depart in favor of greener pastures in vision and computer science.

The next slide in my talk to the thesis committee was a couple of quotes:

"The key observation is that neurophysiology and psychophysics have as their business to describe the behavior of cells or of subjects but not to explain such behavior."
- Marr, Vision, 1982

"If one just wants lots of varied evidence, one can assemble an encyclopaedia of facts, but the hypothesis that is their conjunction will be incredibly complex because the facts are so heterogenous."
- Lipton, Science, 2005

I think they hit the nail on the head (the aforementioned nail being the last one on the coffin of auditory cortex research, if we were to continue this way). All of the above results are important in themselves, they are results of testing hypotheses of themselves. Me and Elias call these 'point theories', and we believe it is misleading to go this route. But the picture is still hopeful: as long as the aspiring phenomenologist even thinks about linking up his contribution to atleast one stage below his level of description, we would have a much more cohesive picture.

Note: All the figures are taken from papers published in peer-reviewed journals. They represent contributions of many different labs. I'm not citing specific references because I feel that it's irrelevant to the discussion at hand: what's more important is the data that's out there, not who is doing what.



[+/-] show/hide

JC 2: Some fine detective work

Ref.: Livingstone MS and Conway BR, Was Rembrandt Stereoblind? N Engl. J. Med. 351: 1264-1265


Abstract:
Stereopsis is an important cue for depth perception, yet it can be a hindrance to an artist trying to depict a three-dimensional scene on a flat surface. Art teachers often instruct students to close one eye in order to flatten what they see. Therefore, stereoblindness might not be a handicap — and might even be an asset — for some artists. Stereopsis requires precise alignment of the two eyes. We examined a number of self-portraits of Rembrandt,
an artist known for his astute powers of observation, and noticed that many of them show his eyes as exotropic, some to a degree that would be incompatible with normal stereopsis. We wondered whether the gaze angle of the eyes in Rembrandt’s self-portraits was random or whether the gaze deviation was systematic, as it would be if he were accurately portraying a feature of his physiognomy.

Summary:
One of Livingstone's interests is the interaction between Vision and Art. In this paper, she analyzes the self-portraits of Rembrandt in an effort to understand if an observed lateral deviation of his left eye in these paintings was intentional portrayal of a real strabismus. They analyzed high resolution images in 24 oil paintings and 12 etchings where they could quantify the position of the pupil (or center of the iris) and the extent of its deviaiton from the center. In 23 of the 24 paintings, the right eye looks toward the viewer and the left eye looks outward. In all 12 etchings, this trend is reversed, because of the method of etching: the artist first draws on a metal plate and tranfers it to paper, creating a mirror inversion. The one painting that did not show the effect, the authors guess, could either be a copy of an etching or painted by a student. Since lack of stereo-vision is an advantage to artists (as they have to paint 3D scenes on a flat surface, Rembrandt's stereo-blindness could have contributed to his greatness.

Comments:
The purpose of the JC series was also to discuss what we thought were cool papers, so I guess I'll vaguely justify this post by saying it was cool.


This is not an article in the class of Lipton, but it still shows how far some logical thinking goes. An observation of an apparent strabismus, a logical experiment to generalize the observation (to make sure it is not a proof-by-example type effort) together with a theory of how it may have helped the great Master. Again, the perfect interaction of Thinking and Experiment. Plus, I think it's really cool: like performing a forensic exam from a painting.

[+/-] show/hide

Vatsun meets the Committee

Well, I'm not sure why you'd be interested in how it went, but it went well... considering my incendiary slides (which included a graphic timeline of how much the world has moved on between 1975 and 2001, while A1 still obsessed over tonotopy). Let's call the committee members A - E (for obvious reasons, plus you'll have fun guessing who they are). 'A' and 'B' kinda looked upset over my trashing of A1 research, not to mention the statement I made about vocalizations being only a subset of natural stimuli. 'C' was mostly concerned that I was a 4th year student still making too specific hypotheses. 'D' was insightful but silent mostly. 'E' was the only person interested in my two-tonepip idea.

'A' graciously defended me citing all the effort I had put into my erstwhile doomed project, and gave me credit for building a new setup etc. Other than that, I guess we'd already spoken about most other things. B's reaction was more interesting. Initially he was sceptical, perhaps even a tad outraged (especially when I commented on unsubstantiated theories of an ex-postdoc of his), but eventually, conceded that he didn't have a better idea of studying the system either, so he might as well support me. Most of my argument with him centered on the dimensionality of audition: B made the commonly made point that the 'linear' STRF is actually the second-order Weiner kernel and therefore non-sync. neurons were possibly 3rd or 4th order, and I countered with my intuition for sound which is that once frequency channels are separated, it is essentially a first -order amplitude-time system.

'C' was mostly concerned that I would not be able to graduate. He wanted me to get some data, both from my old and new projects and describe something to get out. His concern about my hypotheses being too specific was valid. And, 'C' made the very insightful suggestion that perhaps I should incorporate natural statistics into my two tone-pip stimuli.

Not much else to say about 'D' and 'E'. I should talk to them more though, they probably have more ideas than I do on these matters. We'll see how it goes.




[+/-] show/hide

But Mooooooomm... there's so much variability in Biology....

You got to say the title in the voice of Eric Cartman (go see South Park).

But something struck me. The Brain isn't stochastic. Behavior isn't stochastic (not unless you believe in Free Will, and I very much doubt that). If a lion stalks a zebra and the zebra notices, barring other complications (such as maybe altruistic sheilding of a baby zebra, whatever you call those), 10 times out of 10, the zebra will run. It's a certainty, not a probability. Of course, it wouldn't work if you studied a zebra in a lab while simulating a stuffed tiger attack, that isn't variability: that's stupidity on the experimenter's part.

Sure, it depends on what level you look at. Spike trains from individual cortical neurons do appear to be stochastic. But experiments suggest that by looking over the population and measuring the mean rate, or (not to offend anyone), measuring synchrony, this stochasticity is taken care of. When you talk about behavior, it is predictable. Look at the stuff coming out of Newsome's lab these days for proof. There is one thing the brain must obey: causality.

For all the talk about physicists having it really easy (all they have to do is explain the universe, no big deal really), aren't they the guys who invented uncertainty? They found a way to deal with it. Yes, it required a paradigm shift, but you don't hear physicists whining: 'But Mooooomm... Heisenberg ruined my career...'
[+/-] show/hide

Meet the Committee

No, not the sequel to Meet the Fokkers, although I'm bound to be as
hapless as Ben Stiller. I'm talking about my upcoming thesis committee
meeting where I'll renounce all material and scientific pursuits,
convert to Zen Buddhism and retire to a lonely peak of the Himalayas.

Seriously, what do I tell them? I plan to start by telling them how
after recording from over two hundred neurons, I managed to uncover 4
synchronizers and maybe 10 non synchronizers. Not the promised kind
either. Then I meticulously take down the auditory system, the cause
of all my misery. Maybe I'll spend time on the evils of phenomenology.
Then my plans for the future; more on that later.

If this blog sounds a bit out of place, don't bother. It's just a test
of the blog-by-e-mail feature.

[+/-] show/hide