Serving the Quantitative Finance Community

 
User avatar
JohnLeM
Topic Author
Posts: 379
Joined: September 16th, 2008, 7:15 pm

100 millions time faster than ODE methods

November 14th, 2019, 2:19 pm

Mc Ghee is outrunned, NNs compute today 100 millions times faster than ODE methods !!!
https://arxiv.org/pdf/1910.07291.pdf
 
User avatar
Alan
Posts: 2957
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: 100 millions time faster than ODE methods

November 14th, 2019, 3:51 pm

Mc Ghee is outrunned, NNs compute today 100 millions times faster than ODE methods !!!
https://arxiv.org/pdf/1910.07291.pdf
 "Generating these (training) data required over 10 days of computer time".
 
User avatar
Paul
Posts: 6598
Joined: July 20th, 2001, 3:28 pm

Re: 100 millions time faster than ODE methods

November 14th, 2019, 4:42 pm

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 14th, 2019, 4:44 pm

 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 14th, 2019, 4:48 pm

Mc Ghee is outrunned, NNs compute today 100 millions times faster than ODE methods !!!
https://arxiv.org/pdf/1910.07291.pdf
 "Generating these (training) data required over 10 days of computer time".
Lies, damned lies and neural networks. No way, Jose.

For a real shot at finance SABR application, see 

https://www.datasim.nl/blogs/26/msc-the ... al-finance
Last edited by Cuchulainn on November 14th, 2019, 5:01 pm, edited 1 time in total.
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 14th, 2019, 4:58 pm

Mc Ghee is outrunned, NNs compute today 100 millions times faster than ODE methods !!!
https://arxiv.org/pdf/1910.07291.pdf
As Wittgenstein would say, it is a description and not an explanation. Not a single formula. Reads like a piece of marketing.
 
User avatar
Alan
Posts: 2957
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: 100 millions time faster than ODE methods

November 14th, 2019, 5:00 pm

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
That analogy may be better than you think. When your ten thousand hours does not help me -- and I still have to practice/train the damn thing -- maybe that's a good criterion for a 'useless' machine learning application. 

On the other hand, when the app can be trained by a third party and then I can use it  --  I am thinking Waymo self-driving cars -- that's a different story!    
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 14th, 2019, 5:03 pm

12 days training adds to the destruction of the planet.
 
User avatar
Paul
Posts: 6598
Joined: July 20th, 2001, 3:28 pm

Re: 100 millions time faster than ODE methods

November 14th, 2019, 5:06 pm

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
That analogy may be better than you think. When your ten thousand hours does not help me -- and I still have to practice/train the damn thing -- maybe that's a good criterion for a 'useless' machine learning application.  
Cheeky!  I always think my analogies are spot on!
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 14th, 2019, 5:29 pm

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
That analogy may be better than you think. When your ten thousand hours does not help me -- and I still have to practice/train the damn thing -- maybe that's a good criterion for a 'useless' machine learning application.  
Cheeky!  I always think my analogies are spot on!
Did you check for isomorphism?

https://en.wikipedia.org/wiki/Analogy
 
User avatar
Alan
Posts: 2957
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: 100 millions time faster than ODE methods

November 14th, 2019, 10:39 pm

Speaking of Waymo, which I was, they are right now doing an AMA over at reddit: link

oops -- it's over already and my question did not get answered! 
 
User avatar
katastrofa
Posts: 7430
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: 100 millions time faster than ODE methods

November 15th, 2019, 7:05 am

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
That analogy may be better than you think. When your ten thousand hours does not help me -- and I still have to practice/train the damn thing -- maybe that's a good criterion for a 'useless' machine learning application. 

On the other hand, when the app can be trained by a third party and then I can use it  --  I am thinking Waymo self-driving cars -- that's a different story!    
And, correct me if I'm wrong, they didn't test different values of hyperparameters (the whole tuning was only for two different pairs of "tuning" parameters). That's revolting! I considered this tuning in my paper, appendix A.5, very scanty: https://arxiv.org/pdf/1802.09427.pdf Doing this properly would take them much longer.
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

November 15th, 2019, 10:52 am

It's like learning to play the violin. Ten thousand hours of practice, but then you can rattle through all four seasons in minutes.

But 'ukulele is still better.
That analogy may be better than you think. When your ten thousand hours does not help me -- and I still have to practice/train the damn thing -- maybe that's a good criterion for a 'useless' machine learning application. 

On the other hand, when the app can be trained by a third party and then I can use it  --  I am thinking Waymo self-driving cars -- that's a different story!    
And, correct me if I'm wrong, they didn't test different values of hyperparameters (the whole tuning was only for two different pairs of "tuning" parameters). That's revolting! I considered this tuning in my paper, appendix A.5, very scanty: https://arxiv.org/pdf/1802.09427.pdf Doing this properly would take them much longer.
Has A.5 to do with cross-validation, 5-fold and confusion matrix use cases?
 
User avatar
katastrofa
Posts: 7430
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: 100 millions time faster than ODE methods

November 15th, 2019, 3:18 pm

Sort of. Yes.
 
User avatar
Cuchulainn
Posts: 20203
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: 100 millions time faster than ODE methods

December 3rd, 2019, 6:50 pm

In December 2015, Google announced that the D-Wave 2X outperforms both simulated annealing and Quantum Monte Carlo by up to a factor of 100,000,000 on a set of hard optimization problems.[30]