"This new framework indicates that AI models could be as sensitive as traditional template matching algorithms, but orders of magnitude faster. Furthermore, these AI algorithms would only require an inexpensive graphics processing unit (GPU), like those found in video gaming systems, to process advanced LIGO data faster than real time.
The AI ensemble used for this study processed an entire month — August 2017 — of advanced LIGO data in less than seven minutes, distributing the dataset over 64 NVIDIA V100 GPUs. The AI ensemble used by the team for this analysis identified all four binary black hole mergers previously identified in that dataset, and reported no misclassifications."
Impressive : what can't these wonder GPUs do now ?
But starting at ~ $8.5K per unit makes 'inexpensive' a relative term.
Now, what to do when the seven minutes is up ? LOL ... <insert your favorite game here>, CIV VI maybe ? :-0
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
To find something is one important side, but you also have to characterize how sure your finding is. Up to now it´s international accepted standard to have >5 Sigma statistical error for true evidence. I´m missing such in the paper.
I believe, it will take some time to work out such an international accepted standard for work with AI-methodes, which is neccessary for daily practize, like here at E@H.
Nevertheless, it seems to me an important step forward.
I thought I was talking about including my fellow volunteers.
If the GPU resources needed to run useful AI for data analysis, could be low enough that our current crop of GPU's could do the job, I would love to join the race to AI.
Context. This is a public message thread on the e@h website. So it would apply to e@h? No?
I get that user group message threads can wonder all over.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Could this be applied to the data analysis we are doing for e@h? We certainly have enough raw data to process. I am speculating that every pulsar discovery would be the data needed for training.
Ditto on Gravity Waves.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
We are not solving any PDEs here at E@H. It is basically alot of arithmetic done quickly, Fourier transforms, from which we get an exact result per the given data set. AI comes in when post processing the candidate signals, followed by the human touch to claim discoveries. I cannot see a PDE approximator offering anything here. That's not to say it would not help other goals in astropbysics.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
"This new framework indicates
)
"This new framework indicates that AI models could be as sensitive as traditional template matching algorithms, but orders of magnitude faster. Furthermore, these AI algorithms would only require an inexpensive graphics processing unit (GPU), like those found in video gaming systems, to process advanced LIGO data faster than real time.
The AI ensemble used for this study processed an entire month — August 2017 — of advanced LIGO data in less than seven minutes, distributing the dataset over 64 NVIDIA V100 GPUs. The AI ensemble used by the team for this analysis identified all four binary black hole mergers previously identified in that dataset, and reported no misclassifications."
Impressive : what can't these wonder GPUs do now ?
But starting at ~ $8.5K per unit makes 'inexpensive' a relative term.
Now, what to do when the seven minutes is up ? LOL ... <insert your favorite game here>, CIV VI maybe ? :-0
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Hallo! The according paper
)
Hallo!
The according paper you´ll find here.
To find something is one important side, but you also have to characterize how sure your finding is. Up to now it´s international accepted standard to have >5 Sigma statistical error for true evidence. I´m missing such in the paper.
I believe, it will take some time to work out such an international accepted standard for work with AI-methodes, which is neccessary for daily practize, like here at E@H.
Nevertheless, it seems to me an important step forward.
Kind regards, say healthy and happy crunching
Martin
Have we made any progress
)
Have we made any progress that could be run by e@h volunteers?
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M wrote: Have we made
)
Tom, forgive for asking such a dumb question, but could you define "progress" in the way that you meant it?
Do you mean you want AI to assist us at E@H ?
Do you mean you want E@H volunteers to run the tasks that AI is doing ?
Do you mean that you want those that use AI to allow us at E@H to 'double-check' what AI is doing ?
And if you're asking any of these questions, why not include BOINC volunteers ?
.....an no..... I have not been drinking! ;*)
Proud member of the Old Farts Association
Are you sure you haven't been
)
Are you sure you haven't been drinking? ;)
I thought I was talking about including my fellow volunteers.
If the GPU resources needed to run useful AI for data analysis, could be low enough that our current crop of GPU's could do the job, I would love to join the race to AI.
Context. This is a public message thread on the e@h website. So it would apply to e@h? No?
I get that user group message threads can wonder all over.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
https://phys.org/news/2024-01
)
https://phys.org/news/2024-01-technique-efficiently-partial-differential-equations.html
Could this be applied to the data analysis we are doing for e@h? We certainly have enough raw data to process. I am speculating that every pulsar discovery would be the data needed for training.
Ditto on Gravity Waves.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
We are not solving any PDEs
)
We are not solving any PDEs here at E@H. It is basically alot of arithmetic done quickly, Fourier transforms, from which we get an exact result per the given data set. AI comes in when post processing the candidate signals, followed by the human touch to claim discoveries. I cannot see a PDE approximator offering anything here. That's not to say it would not help other goals in astropbysics.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal