brain vs artificial net and human brain power (was: confused)

Bruce Raoul Parnas brp at muttley.berkeley.edu
Tue Aug 6 12:04:52 EST 1991


In article <MORE.91Aug5183430 at tk3.oulu.fi> more at stekt.oulu.fi writes:
>slehar at park.bu.edu (Steve Lehar) writes:
>
>No. The reason we abandoned analog computing was the evolution of
>digital computers. With digital simulations of analogical signals we
>could work faster and more accurately (at least in usual cases).

No, actually, analog computation is faster than digital computation.  An
analog computation system computes one second of real time in...one
second.  Digital systems are certainly more flexible- it's much easier
to change code than resistor values- and this, I believe, accounts for
the switch from analog to digital.  Since neural systems are inherently
analog and massively parallel, I feel that we will have to revert (or
RE-progress) to analog, parallel computation in order to get a handle
on it.  Analog representations of neurons are being built in arrays, and
these will be the future of neural computation, IMHO.

Analog systems are also inherently more accurate, if characterized 
properly at the outset.  The nonlinear properties of the devices are
well-suited for simulating the nonlinear processes in neural systems.
There is no quantization error to speak of, no floats, ints or
doubles.

>Actually neuron cell potentials are quantized in ion level. There is
>abt 10^6 ions (abt 20 bits/cell) in one neural cell.

I guess I don't follow the conversion between ions/cell to bits/cell.

>Like you said earlier; the brain never converges.

Brains do eventually converge to a final solution.  It's called death.

>3. Dynamic resource allocation (dynamic topology)
>   Neural structures are constructed during learning.
>   --> Neural environment will organize to do lazy computation.
>   This means the best result with the least computation.
>   Unnecessary neurons are killed.

Neurons unnecessary for a particular calculation should not be removed
from the model.  They should be reassigned to another task.  If all
neurons not now necessary are killed there is no room for learning
new information without degrading old.  This doesn't make for a realistic
model of brain activity.

>more at stekt.oulu.fi - Jyrki Alakuijala - Atraintie 6, 90550 Oulu, Finland
>+358-81542334, male, 20 years, University of Oulu, Neurosurgical Research Unit


bruce
(brp at bandit.berkeley.edu)



More information about the Neur-sci mailing list