SERVING THE QUANTITATIVE FINANCE COMMUNITY

FaridMoussaoui
Posts: 507
Joined: June 20th, 2008, 10:05 am
Location: Genève, Genf, Ginevra, Geneva

Re: 100 millions time faster than ODE methods

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]
Are they still in the proof of concept paradigm?

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

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]
Are they still in the proof of concept paradigm?
I only discovered it via trackstar few weekends ago.

Managed to download (you need to register) + they give you a dashboard + 0.01 secs per month free.
That dampened by initial enthusiasm somewhat.
When will it be a commodity?
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

There are many ways to measure time.

My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

This is a trailer from a recent very good thesis on Heston and Rough Heston. Instead of ANN being $10^4$ faster, in this case it is $[8,17]$ times slower.

Thesis public domain soon..
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

tagoma
Posts: 18399
Joined: February 21st, 2010, 12:58 pm

Re: 100 millions time faster than ODE methods

This is a trailer from a recent very good thesis on Heston and Rough Heston. Instead of ANN being $10^4$ faster, in this case it is $[8,17]$ times slower.

Thesis public domain soon..
Was the very purpose of this research to compare the speed of "traditional" methods vs ANN? What methodologies were applied to prevent biases? And isn't it apples vs bananas, anyways?

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

No, there are several goals. Just be patient before jumping to conclusions.

A by-product is to debunk the myth than NN is 'fast'. In that sense the above snippet is relevant to the current thread.

//
I also discuss the software aspects in my recent Thalesians video. And the speech by Grady Booch is a gem ("Software design with AI components").
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

"What methodologies were applied to prevent biases?
1.  5-fold cross validation. $R^2 = 0.999$

2.  Other works take a more laissez-faire approach.

I have examined their code, which yield extremely high accuracy. However, they are evaluating the accuracy using in-sample data, which tells us nothing about the accuracy of the ANN on unseen/out-of-sample data."

Sneaky.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

tagoma
Posts: 18399
Joined: February 21st, 2010, 12:58 pm

Re: 100 millions time faster than ODE methods

I long to have a look at this piece of research.
Sincerely,
tag

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

I long to have a look at this piece of research.
Sincerely,
tag
tag,
I reckon 2-3 weeks and then it will hopefully be public domain
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

JohnLeM
Topic Author
Posts: 471
Joined: September 16th, 2008, 7:15 pm

Re: 100 millions time faster than ODE methods

This is a trailer from a recent very good thesis on Heston and Rough Heston. Instead of ANN being $10^4$ faster, in this case it is $[8,17]$ times slower.

Thesis public domain soon..
That is more or less what I was expecting. I guess the thesis used a customized finite difference 2D scheme ?

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

This is a trailer from a recent very good thesis on Heston and Rough Heston. Instead of ANN being $10^4$ faster, in this case it is $[8,17]$ times slower.

Thesis public domain soon..
That is more or less what I was expecting. I guess the thesis used a customized finite difference 2D scheme ?
Kind of: it is a rough Heston model that was solved as a fractional Riccati equation. It also uses the characteristic function and FFT.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

Cork student wins Young Scientist 2021 with 'deepfake' detection software

https://www.irishexaminer.com/news/muns ... 03374.html

The project uses a sophisticated artificial intelligence software program that can efficiently detect DeepFake media with state-of-the-art accuracy.

The software, which is over 150,000 lines of code, made significant improvements on speed and efficiency when compared to the current best model without sacrificing its ability to accurately detect the fake.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

tagoma
Posts: 18399
Joined: February 21st, 2010, 12:58 pm

Re: 100 millions time faster than ODE methods

How can someone even write 150K lines of proper code?

Cuchulainn
Posts: 63816
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: 100 millions time faster than ODE methods

How can someone even write 150K lines of proper code?
According to Mr. Tarr, this is wrong. Probably the journo who wrote it. He used the usual libraries.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl

JohnLeM
Topic Author
Posts: 471
Joined: September 16th, 2008, 7:15 pm

Re: 100 millions time faster than ODE methods

How can someone even write 150K lines of proper code?
I wrote much more than 150 k code lines :/