Re: [microsound] Math Professors Solve 'Cocktail Party' Problem

[ãããããããã <ndkent@xxxxxxxxxxxxx>]

This does sound like a pitch for U.S. government mad money.

I wouldn't be surprised is the neural net is just a handy buzzword to  
make an AI claim.

For the person asking, an algorithm has a predetermined sequence of  
steps that result in a solution. A neural net tries to "learn" how to  
most successfully solve a problem by trying different solutions and  
somehow weighting it's own results so as to refine further solutions.
> The computer we use is doing the work without an algorithmic  
> program. It uses a system called a neural net, which is designed  
> for the computer to teach itself. Basically, it works on trial and  
> error," Casazza said. "This isn't consistent and cannot be  
> duplicated easily. We need to find a way to design an implementable  
> algorithm that could do this consistently and quickly."
Sounds like a total "gotcha" to me. The unsaid factor is how is the  
neural net being refined? I've got the feeling "trial and error"  
means there's a human involved informing the neural net how  
successful or unsuccessful it's output is. You know, how else would  
it know it was performing better in this situation as the results are  
perceptual rather than quantifiable? It's not likely figuring this  
out on it's own, it's probably right now just a set of potentially  
but not necessarily new noise reduction tools probably guided by the  
feedback of a human monitor. The issue is primarily qualitative not  
quantitative and sounds like it's eluded a single solution they might  
wish for.



---------------------------------------------------------------------
To unsubscribe, e-mail: microsound-unsubscribe@xxxxxxxxxxxxx
For additional commands, e-mail: microsound-help@xxxxxxxxxxxxx
website: http://www.microsound.org