Connectionists: [Bmi] The Brain Principles Manifesto: with links

[Juyang Weng]

On 2/22/15 6:54 AM, Jont Allen wrote to BMI mailing list:
> 100% yes, the brain operates on classical principles. Does anyone
> disagree?
> If this were not the case, then there would be no point of
> neuro-science studies of the brain.
>
> Perhaps we have not yet discovered the basic principle of how the
> brain operates.
>
> Here is my proposal: Each neuron is a low-pass filter, due to the
> membrane's RC properties.
>
> If you want a reference to this fact, read George Campbell's famous
> papers on wave filters:
> ,title={Physical Theory of the Electric Wave Filter}
> ,journal=BSTJ ,year=1922 ,month=Nov ,volume={1} ,number={2} ,pages={1-32}
>
> The inputs are, of course, large numbers of spikes at the synaptic
> inputs. Thus
> the low-pass filtered sum of the inputs leads to an output spike, when
> the lowpass voltage
> crosses a threshold.
>
> One key to this computation is the incredibly low noise of the
> neuron's low-pass filter, which
> is given by sqrt{kT/C} where C is the membrane capacitance.
>
> This very low noise results in reliable computation: Same input, same
> output.
>
> Jont

Jont, your proposal looks like what an electrical engineering student
would see in a neural net course.
Prof. Shun-Ichi Amari at RIKEN in Japan and Prof.  Jun Zhang at
University of Michigan at Ann Arbor seem to take a similar physics based
approach.   But I think the approach is way too limited.

Many respected neural net modelers had PhD degrees in physics which give
them an intuition in modeling events of physics.   However, I think that
this physics background is also a trap, in the sense that the
researchers trapped into a wish that physics-based mechanisms plus a
weak mechanism (e.g., information theory) are sufficient for them to
model brain-like neural networks.   This wish is an illusion.

The brain operates on MUCH MORE PRINCIPLES THAN classical principles.

If they have taken a course in automata theory, a course in Artificial
Intelligence, and the BMI 871 course on brain-mind modeling, they will
see how limited their tools are.  They should also take courses in
biology, neuroscience, and psychology that BMI 2015 plan to offer.

In particular, a low-pass filter is not sufficient to characterize what
each neuron does.  For example, where does each neuron get inputs from?
The automata brain model tells you where:  Not just sensory source, but
also motor source.  It is also very different from classical control
theory too, not just feedback control.

-John