[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
computer-go: frame concept
Go is a good example of AI because of its complexity. We may have algorithm and
data structure for lot of local question, e.g. cut and connect, shapes, influence, life
and dead. The main question is how to use our resources, time and memory. (Of
course it is also important how to compare results of different functions.)
The main answer may be: we should use our resources to produce maximal
information. We should estimate produced information of every larger part of the
program. But it is not easy to define and measure information.
An example: let two alternatives A and B. We have two methods M1 and M2. First
is faster but second if better. Both able to evaluate both alternatives. Say
M1(A)=12.4 and M1(B)= 13.2. If we search for maximum, we may choice B in that
case. Or we may ask M2 to get more accurate result. If we get different numbers but
M2(A) < M2(B) then we don’t get information in first meaning because we take the
same decision. If M2(A)> M2(B) then it was useful to ask the more expensive M2.
And if we collect experiment, we may estimate the probability of "M2 says different
relation then M1". We may suppose that large changes of value is rare then small.
We may make a table: possibility of changes of relation depending on difference of
M1(A) and M1(B). We may define information production somehow in that
situation. And if we have M3 for the same function than we may use this concept
to optimize time using.
But we cannot compare immediately to another kind of situations. We may measure
information in another situation and use it locally. But if we have to compare
produced information of different methods then we should improve techniques for
that task. It may be as easy that we should multiply by 5 before compare, but may
be more difficult too.
Applying this concept to go is exceed my time so I trying to do something in
gomoku.
Best regards
Janos Wagner