[Comp-neuro] From Socrates to Ptolemy
todd.troyer at utsa.edu
Mon Aug 11 04:31:20 CEST 2008
I would like to highlight the 'perhaps most important' part of Jim's
definition of 'realistic' models, namely the concept of being 'idea
neutral.' Most neural models take the form of 'this hypothesized mechanism
is sufficient to account for that set of experimental data.' The dominant
effort is to bolster or clarify a previously existing idea. Without
constant vigilance, these models can flow right into Jim's 'tyranny of
I would posit that the tyranny stems in part from an over-reliance on a
cartoon version of the scientific method. Contrary to popular dogma, science
doesn't begin with a testable hypothesis. Such hypotheses have to come from
somewhere and finding them is just as much a part of science as is the
testing. One place hypotheses come from is from observation, including
using systematic 'exploration' and careful measurement to 'characterize' a
system. Although the quoted words will still kill most grant applications,
the ability to gather, store and process huge amounts of data is currently
pushing the envelope of the scientific method, with data-mining approaches
to genomic analysis being the most prominent example.
Closer to home, modern neurophysiology labs are generating huge data sets,
and these are often much richer than the tightly honed hypotheses that the
original experiments were meant to 'test.' Computational neuroscience has
much to contribute here, but we need to start thinking about how to exploit
more exploratory 'idea neutral' approaches to complement more traditional
'hypothesis-driven' approaches. I think Jim's arguments about large-scale
realistic modeling are closely related to this broader issue, and agree with
much of what he says.
Of course 'abstract' models have a crucial role to play as well. Returning
to the question of where hypotheses come from, these might also arise from
the field of 'theoretical neuroscience' (quotes used to avoid having to
clearly define what I mean). Certainly one role of theory in science is to
clearly define questions and concepts, and to attempt to systematize
knowledge. Simplified models can play a critical role in these efforts.
Unfortunately such work is relatively thankless, and there is more
professional payoff for explaining/refining the latest hot set of
Given that physics and biology are such different beasts, I'm often
unconvinced that drawing parallels between them is all that useful. But I
certainly agree with Jim that there is a lot of conceptual work left to be
done in neuroscience. We should put more effort into thinking about our
methods, and we should do a lot more observational science.
As for the more mundane questions of learning and noise, I think the
argument for learning in the brain actually follows Jim's example about
stimuli and outcomes fairly closely, with one important difference.
Certainly, having different responses to different inputs doesn't imply
learning. But learning might reasonably be defined as the process of
creating different responses to the SAME input presented over time,
presumably related in some way to the experience of the animal during the
intervening periods. There's plenty of room to argue over more precise
definitions and the nature of the experimental evidence, but in the broadest
sense the evidence for learning comes from demonstrations of
experience-dependent changes (improvements) in the function of the system.
A similar operational approach can be used to define noise as the
variability in response to an identical stimulus. Much of the discussion of
noise in the preceding thread focused on the origin of this variability. (I
wouldn't say that it was firmly established that true 'thermal' noise is
COMPLETELY negligible.) But whatever you think of the origin of this
variability, it's important to remember that the brain must be organized so
that this variability does not disrupt the mappings between a stimuli and
the corresponding 'percepts' (in a sensory example), and this correspondence
should be stable over time and maintained as the system changes over time.
UT San Antonio, Biology Dept.
One UTSA Circle
San Antonio TX 78249
Todd.troyer at utsa.edu
W: 210-458-5487, FAX: 210-458-5658
More information about the Comp-neuro