AI Designs Quantum Physics Experiments Beyond What Any Human Has Conceived

So then again it is the humans setting it all up. The final data accumulation and decisions based on that is done by the computer, but completely under human established routines and parameters.
Humans have to do all the organization because we are the ones who define success.
Of course it's true that humans design the system, but after the design they must be out of the loop for it to be useful.

The whole idea of industrial AI is for unskilled workers to handle the system. That has been going on for decades. Some AI systems are self-correcting. If it detects that the recognition process is getting out of tolerance, it reprograms itself. That can happen if the chip changes appearance, because of a new batch from a different manufacturer.
 
In a chip location program there is no decision tree. The network is programmed to handle any chip. No further programing is necessary. In the factory a worker simply presses a button so the same algorithm learns the pattern (changes the synaptic weights) for recognition of the new chip.

I have never heard of a "chip location program".
But there is no way to program a network.
You can only program individual computers with what will be executed on that computer.
And I see no way one could program for any chip because new chips are constantly being created, so will need to be have new resources and definitions added for new chips constantly.
If a factory worker presses a button, it will only cause what a programmer created, to be executed.
And if it has an algorithm, it does not learn anything.
It just executes.
If it then accumulates data, that is not learning really.
If a new chip causes different data, the algorithm can't change any heuristic decision weights on its own.
It can only change weights if the programmer has properly caused the right code path to be taken for that chip.
It is all very static and must be, otherwise you could never verify the operation of the neural net.
 
Why is it on 'Star Trek', every time they go where no man has gone before, there is always some there waiting for them?
Fair question.

Perhaps the design of a human is so functional it is closely copied all over the universe. (Or it was just too expensive to create unique alien costumes for every episode.)
 
Where would we go?
1635138322029.png


Not here

*****CHUCKLE*****



:)
 
I have never heard of a "chip location program".
But there is no way to program a network.
You can only program individual computers with what will be executed on that computer.
And I see no way one could program for any chip because new chips are constantly being created, so will need to be have new resources and definitions added for new chips constantly.
I don't know where you are getting your ideas, but you are terribly out of date. You are talking about the technology of the 19980s. I'm talking about the 2000s.

There are thousands of old chip designs and new ones coming in all the time. The die bonder or pick and place machines have to train any new chip or old chip in seconds so the assembly line won't be hung up. An unskilled operator puts the first chip or IC package in the machine and presses a button. The algorithm "learns" the new chip pattern in a fraction of a second by changing decision weights on its own.

The computer and software do not have to be adjusted or reprogrammed for any new chip. Only internal parameters are automatically changed. For larger chips a different machine with different optics may be needed, but the computer and algorithm do not change. I've been there and done all that.

I'm not surprised you never heard of a "chip location program". It is highly proprietary for the several companies in the business. You will not find any thing on the web unless you are willing to pay for articles in the IEEE journals or other specialty journals. Look up computer vision machine learning for an overview.

.
 

Forum List

Back
Top