View unanswered posts | View active topics It is currently 23 Aug 2017 09:44



Reply to topic  [ 4 posts ] 
New approach 
Author Message
Rear Admiral
Rear Admiral

Joined: 31 Dec 2008 20:59
Posts: 3362
I read in the library yesterday an article in [i]Discover Magazine/i] it is on neural network computers and the latest approaches. It of course got some thoughts started.

Energy budgets, is the first thing that comes to mind for computers in this line of thought. This determines two things what you can do, and how much heat is produced per unit of computer power. So lets say that you have a fixed budget of 2,000 watts for a given ship. That is all. It will of course produce a given amount of heat. This heat in space won't be easy to get rid of, so you are limited here in what you can do...

Also lets keep this energy budget for decades, maybe centuries to come. Why? Because it is a good figure, as a starting point.

Now about what the article said. Brains are very remarkable in their energy budgets, the human brain consumes only about twenty watts of power to do what it does, and that the computer being designed, by the scientist, will consume about one hundred times more than what the human brain does. This is far more efficient than the way we do things now. Each of the chips has 256 neurons, and 256 by 1024 synapses... This is the expected low density for test articles. So using SFC, once again, an elementary program independent thought computer, would have more capability than required, but the article states that they realized that, not all neurons had to be turned on all the time, greatly reducing the overall power requirements, thus greatly improving overall performance. This 'elementary' system would be more in the way of a prototype, rather than a full up operational system. That is, it would operate on minimal data inputs.

What is of value according to the article the ability to generalize, from the data sets introduced during training, this to a certain degree is old news. But for this "prototype", it should be the equivalent maybe a very fast 16 bit computer... In other words the jobs assigned to it, wouldn't be too difficult, ship control, and the such like...

But where it starts to get interesting, is with the next generation, the 'intermediate'. It is described as having ten times the Independent thought circuitry as the previous generation. But this new approach suggests what memory scan may be. Consider, you have an energy budget, and you have to work around this budget, so you do the old thing - page memory. This would great expand the functionality of the computer. So think in terms of the IBM PC versus the IBM PC AT... In other words the difference between a 64k computer, and a 640k computer. In other words a vast difference.

The next generation after this, the computer would be able to pass the Turning test - which is the ability to fool a human into thinking they are talking 70% of the time for five minutes...

But there is one fly in ointment. You guessed it St: Voyager. They are 300 + years too long of the mark...


14 May 2013 09:20
Profile
Commodore
Commodore
User avatar

Joined: 20 Dec 2008 18:39
Posts: 2377
Location: toltec sector
Quote:
But there is one fly in ointment. You guessed it St: Voyager. They are 300 + years too long of the mark...


well... to be fair... Tuvok could probably pass a Turing Test... ; )

_________________

All matter originates and exists only by virtue of a force...
We must assume behind this force the existence of a conscious and intelligent Mind.
This Mind is the matrix of all matter.
Max Planck


14 May 2013 16:54
Profile
Rear Admiral
Rear Admiral

Joined: 31 Dec 2008 20:59
Posts: 3362
:lol: :lol: :lol: :lol: :lol: :lol: :lol: :lol: :lol: :lol: :lol:


14 May 2013 20:32
Profile
Rear Admiral
Rear Admiral

Joined: 31 Dec 2008 20:59
Posts: 3362
Now getting back to this idea...

I want you to keep in mind that this is general, not specific to SFC. That is, if I analyze this correctly, then I should have something that drops into FRS with out a ripple, as well as other idea...

The first next thing to know, is that one doesn't have to teach from scratch, a neural network. That is, translations programs exist to translate one type of programming from one nature to another. That is, one can write a program, that is hard coded for a traditional CPU, and have rewrite that program into the new neural network format. Such that, if you have a program that is designed to behave as a human would, in a given system, on a standard CPU, such as a NPC, it can be translated to be entirely neural network.

What this means is that a Neural Network's job can be specialized according to need. What one has to know, is the inputs. This will determine a great many things down the road. To do this one must know the objective. What is the goal that you are hoping to achieve?

For example, in SFC, there is made mention of the DY-500, having an Elementary Program Independent Thought Digital Computer. What would a DY-500 need this for? What advantages does it give to the ship, and above all else what advantage does it give to the crew??

In theory, it should enable the DY-500 to operate with out a crew. In theory. So where does it fail? Or does it??

To answer these questions, one must do a systems analysis upon the DY-500. What is its mission? What capabilities does Earth have vis-à-vis space travel? In other words what limitations does Earth have, and how do they play out for the DY-500?

So what is so special about neural networks? It is their ability to generalize for specific data. That is, to think things through from a specific event, that may be ongoing, and come up with a solution that is not preplanned out, that is unique to the situation, that works. Now keep in mind that in the wild predators are only 50% successful in getting they dinner per strike. So what is going on with the DY-500? Orbital adjustments around a planet; Orbital insertion into solar orbit( which may or may not be to solar escape); orbital insertion to another world's orbit.

Then you have science data. How should the DY-500 react to a unique event? Should it? What limitations on the ship may cause it to back off?

What changes are discovered to be necessary between 2018 and 2028? Why the Intermediate Program Independent Thought memory Scan Computer?


15 May 2013 12:28
Profile
Display posts from previous:  Sort by  
Reply to topic   [ 4 posts ] 

Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum

Search for:
Jump to:  
cron

Forum hosting by ProphpBB | Software by phpBB | Report Abuse | Privacy