h a l f b a k e r yNot from concentrate.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
|
The famous SGI (Silicon Graphics) workstations are examples of machines optimized for a specific kind of CPU-intensive task, in this case 3D rendering. I propose a workstation optimized for AI. Much as the SGIs contain hardware that accelerates ray-tracing and similar operations, the AI-Station would
be equipped with silicon to aid in the simulation of neural networks (commercial neural-simulator ICs are available), as well as hardware to accelerate computations involving genetic algorithms and similar obsessions of modern AI. It could be used with a UNIX variant optimized for such work, much as SGI's IRIX is optimized for multimedia processing.
Woikstation with a lisp
http://www.ipsj.or....302/article007.html
[thumbwax, Oct 04 2004]
[link]
|
| |
Alas, before you can optimise a task you need to have a clue how on earth you plan to do it. AI research is largely still at the stage of "hey, what does this button do?" |
|
| |
sadie is right. we just aren't quite ready for this yet. altho, it could be a better idea to create a clustered hardware platform where different nodes are geared a different task, yet all compatible. |
|
| |
LISP machines, optimized for "symbolic processing", were as close as you're ever likely to see. See [thumbwax]'s link. |
|
| |
I think I more likely format would be a plugin card with a specialised "neural net" or "genetic algorithm" processor on it. These kinds of systems have already been implimented in FPGAs, so there is no reason why you couldn't build an equivalent chip in 'normal' silicon if there was sufficient demand. |
|
| |
As ironfroggy hinted, I think it would be worthwhile having multiple processors for different tasks. Just as we currently have 3D graphics chips and audio DSPs in many of our PCs to take the strain off the main CPU, perhaps we will soon have a specialised chip for speeding up neural network processing. |
|
| |
It would perhaps be fair to say that military simulations and subsequently computer games have driven the development of the first two, perhaps it will also drive the third? |
|
| |
"...the AI-Station would be equipped with silicon to aid in the simulation of neural networks (commercial neural-simulator ICs are available), as well as hardware to accelerate computations involving genetic algorithms and similar obsessions of modern AI..." |
|
| |
Sounds like Danny Hillis' "Connection Machine"...? |
|
| |
Assuming that we are talking about existing A.I. technologies (neural networks, fuzzy logic), then I think it would be easier to buy off-the-shelf hardware. Companies have made (and eventually abandonded) fuzzy logic- and neural network-specific hardware. Yes, you can still find some, but most of these devices are in the graveyard. The plain fact is that most of this stuff is math-oriented and conventional hardware (which is cheap-cheap-super-cheap because it is a commodity) catches up to specialized stuff relatively quickly. My desktop machine would have been considered a supercomputer perhaps 5 to 8 years ago. |
|
| |