 |
|
|
 |
|
|
|
|
AIstudent_msc |
| | | | | | posted 8/7/2012 06:47 |  | | | | | | | | | I've searched the web to find a reference comparing AI algorithms like Genetic Algorithms, Hill Climbing , Tab search, Simulated Annealing, Particle Swarm optimization. Like which has been proven to be good in which kind of problems. I know for example that Genetic Algorithms are good in global search but Hill climbing are good for local search.. stuff like that. Any idea about a good reference for this ?
I'm trying to combine (GA + Something) to solve a specific problem.
Thanks !
| | |
| | |
AiHasBeenSolved |
| | | | | | posted 8/7/2012 15:26 |  | | | | | | | | | | |
AIstudent_msc wrote @ 8/7/2012 6:47:00 AM:
[...] I know for example that Genetic Algorithms are good in global search but Hill climbing are good for local search.. stuff like that. Any idea about a good reference for this ? [...]
| |
If the question is specificially about AI algorithms for search, such as might be used in future vesions of a Web search engine like Google or a winner-take-all successful competitor to Google, here is today's AiHasBeenSolved message about using MindForth AI algorithms for the optimization of search engines:
Tues.7.AUG.2012 -- Testing for Self-Referential Thought
Testing the AI for self-referential thought involves the behavior of
the AI when responding to queries and when left to its own devices.
The AI needs to accept voluminous input about itself and it must then
be able to respond correctly to any mix of questions with the following
four formats: Who are you?; What are you?; What do you do?; and
What do you [Verb]?, such as "What do you make?" or "What do you need?".
The AI needs a gradual increase in its "CNS" memory size so as to
be able to retain all the knowledge necessary for answering queries.
When left to its own devices without human input for a period of time,
the AI needs to follow chains of thought during which its attention
wanders from topic to topic and during which it generates its own
queries for further infomation when a chain of thought peters out.
-----------------------------------------------------------------
Transcript of Forthmind 20120803 interview at 4 17 30 o'clock on 7 August 2012.
Human: you make trouble
Robot: I HELP KIDS
Human: you need energy
Robot: I MAKE TROUBLE
Human: what are you
Robot: I AM PERSON
Human: who are you
Robot: I AM ROBOT
Human: what do you do
Robot: I NEED ENERGY
Human: what do you make
Robot: I HELP KIDS
Human: what do you need
Robot: I MAKE TROUBLE
Human:
Robot: WHAT AM ENERGY
Human:
User Command: halt
-----------------------------------------------------------------
The above transcript shows areas in need of troubleshooting,
such as getting the AGI to answer queries more immediately,
instead of letting some currently active thought intervene.
But the whole issue of an AGI responding to queries emboldens
us to make a suggestion for future search engines.
Suppose that a search engine like Google or Altavista or Bing
developed an AGI way of organizing the content of the Web.
Instead of holding a massively flat "cache" of the Web-at-large,
suppose that the search engine took a version of MindForth or
AGI-what-have-you and dedicated a large number of servers so
that X (large number of) servers each individually became
dedicated to maintaining knowledge about X (large number of)
topics empirically selected as being the X (large number of)
most frequent Web queries made by human beings.
Then Google or what-have-you would continue in operation
with its traditional Larry-Page-rank algorithms, except for
when any of the X (large number of) specific topics came
into play, such as, say, the topic of robots. (I was
going to suggest "unicorns", but not enough is known about
the topic of unicorns.) The process of answering the query
would get shunted or assigned to the AGI-server in charge
of knowing everything pertinent to the topic of "robots".
Then the AGI-server would respond to the query not with
Larrys and Larrys (Pages) full of random cache-retrievals,
but with one _thought_ after another about robots and
in discussion of robots. The response would be knowledge,
not a data-flood. The human user would be in a dialog
with the search engine, and not in a flash-flood of hits.
| | |  Free AI source code of MindForth artificial general intelligence (AGI) with alforithms for knowlefge-based search | | |
|
|
|
|
|
