Is uploading feasible?
erwin at trwacs.fp.trw.com
Fri Dec 17 15:43:43 EST 1993
In article <CI6s6D.6rE at carmen.logica.co.uk>,
Richard Wilson <WilsonR at LILHD.Logica.com> wrote:
>In article <2epn69$r4v at wp-sp.nba.TRW.COM>
>erwin at trwacs.fp.trw.com (Harry Erwin) writes:
>>Choice is a bit more complex than just non-determinism. There are actually
>Sure but does this mean that you agree it is non-deterministic?
Yes, a Nash equilibrium (non-zero-sum generalization of the saddle point
solution to a zero-sum game) requires 'die-rolling' processes. However, as
an open system, the brain can use external inputs as the source of the
necessary 'random numbers'. What gets interesting is that the consensus
strategy can be expected to evolve chaotically.
>>three behavioral options open during most choice processes: go right, go
>>with work I've been doing on mass action models (Freeman Nets) to define a
>Glad to be of service. What you say is of interest but seems to me to avoid
>or omit the act of choice. Yes there are several options and game and chaos
>theory are relevant to decision making but how does dynamic programming
>implement the act of choice?
Dynamic programming provides an optimal rule for making (or deferring) the
act of choice. Consider a quasiperiodic process with two associated
near-limit cycles. Cycle A implements a 'flight' plan. Cycle B implements a
'commit' plan. The system is loaded with two sets of patterns for
matching. If set A is matched, the dynamics drop into cycle A. If set B is
matched, they go into cycle B. If neither are matched, information
collection continues. In a game against nature, set A and set B can be
held constant. I've shown that in a game against an intelligent opponent,
the sets must be modified over the course of the engagement. Dynamic
programming explains how to define set A and set B. Note that defined
thus, this system can be implemented (conceptually) in three interacting
KII networks for games against nature. A game against an intelligent
opponent requires additional complexity in the network, including a source
of 'random numbers' to control the timing of set modifications and
additional KII networks to implement the set modifications. I don't know
(yet) whether I need recursive enumerability or whether context
sensitivity is enough...
Internet: herwin at cs.gmu.edu or erwin at trwacs.fp.trw.com
Working on Freeman nets....
More information about the Neur-sci