BPJ0%LEHIGH.BITNET@ibm1.cc.lehigh.edu (Bin) (02/24/89)
I would like to hear about any interesting usage of the Amiga related to artificial intelligence, something which wouldbe hard for a normal pc to do. Use of products like Magellan, Mandala or authoring systems like Pilot etc.... Trying to Amigatize the AI lab, Bin.
edwin@hcr.UUCP (Edwin Hoogerbeets) (02/28/89)
In article <9472@louie.udel.EDU> mmdf@udel.UUCP writes: >I would like to hear about any interesting usage of the Amiga related >to artificial intelligence, something which wouldbe hard for a normal >pc to do. I always thought the Amiga would be great for simulating neural nets, because of the multi-tasking OS. Each node would have its own process. Each node could be connected to others via [fast] links that are, in reality, messages between MsgPort's. This makes it easy to simulate any kind of perverted topology imaginable! Shared memory is easy, because any address valid for one process is valid for the other processes, and semaphores are already implement in the OS. The Amiga is a virtual playground for such programming. [pun intended, so there.] Seriously, I think the Amiga is uniquely suited to this sort of programming. Try that on a single-tasking Pissy^H^H^H^HC or Atari, [insert suitable rhyming first name here]. This is something I would have done by now, but I don't really know enough about neural nets. Edwin
w-colinp@microsoft.UUCP (Colin Plumb) (03/02/89)
edwin@hcrvax.UUCP (Edwin Hoogerbeets) wrote: > In article <9472@louie.udel.EDU> mmdf@udel.UUCP writes: >> I would like to hear about any interesting usage of the Amiga related >> to artificial intelligence, something which would be hard for a normal >> pc to do. > I always thought the Amiga would be great for simulating neural nets, > because of the multi-tasking OS. > Each node would have its own process. Each node could be connected to > others via [fast] links that are, in reality, messages between > MsgPort's. This makes it easy to simulate any kind of perverted > topology imaginable! Shared memory is easy, because any address valid > for one process is valid for the other processes, and semaphores are > already implemented in the OS. Edwin... a cell in a neural net has about 3 words of context. Adding Tasks, MsgPorts, and Messages flying around would kill it in overhead, both in space and time. Sorry, but... not this time. Well, what about AI is hard for an IBM-style PC to do? Mostly, AI stuffe needs a large address space. Even Amy's 9.5 Meg (without an '020) might not be big enough. -- -Colin (uunet!microsoft!w-colinp) "Don't listen to me. I never do." - The Doctor