Serving the Quantitative Finance Community

 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

May 21st, 2019, 7:04 pm

Funny, but I have a photographic memory. Faces I never forget.
It's the names I can't remember.
 
User avatar
tw
Posts: 592
Joined: May 10th, 2002, 3:30 pm

Re: If you are bored with Deep Networks

May 23rd, 2019, 3:29 pm

I guess this is starting to become a bit too news related to stay in a technical part of the forum but....

Amazon shareholders reject banning sale of facial recognition software to law enforcement
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 22nd, 2019, 11:49 am

some dark clouds on the horizon?

DEEPMIND'S LOSSES AND THE FUTURE OF ARTIFICIAL INTELLIGENCE

https://www.wired.com/story/deepminds-l ... elligence/


On the first question, there is reason for skepticism. DeepMind has been putting most of its eggs in one basket, a technique known as deep reinforcement learning. That technique combines deep learning, primarily used for recognizing patterns, with reinforcement learning, geared around learning based on reward signals, such as a score in a game or victory or defeat in a game like chess.

DeepMind gave the technique its name in 2013, in an exciting paper that showed how a single neural network system 
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 22nd, 2019, 12:02 pm

P.S. pity that the erstwhile ML forum cheerleaders are unusually quiet.
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

August 24th, 2019, 10:14 am

Maybe they're being slightly bored with receiving condescending one-liners as replies? ;-)

This stuff about DeepMind is widely misunderstood. But can't say more publicly.
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 24th, 2019, 2:06 pm

.
Last edited by Cuchulainn on August 24th, 2019, 2:18 pm, edited 1 time in total.
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 24th, 2019, 2:08 pm

Maybe they're being slightly bored with receiving condescending one-liners as replies? ;-)
Not really. Some did not buy into the euphoria based on the hype and waffle. CS and maths don't mix well..
BTW I have two MSc students doing ML-PDE-FDM  projects (Black Scholes, Heston, SABR). The theses are due in in a few weeks and a combination of C++ and Python are used. At the least, they can reproduce the handful of articles already written and is a first stake in the ground.
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 24th, 2019, 2:34 pm

P.S. I have never seen any articles or even blogs to support the hype here.
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

August 26th, 2019, 7:33 am

I usually respond to specific points that I understand. The problem is, often your points are vague and hard to understand. And you don't seem open to learning things. At least that's my impression.
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 26th, 2019, 11:19 am

I usually respond to specific points that I understand. The problem is, often your points are vague and hard to understand. And you don't seem open to learning things. At least that's my impression.
Seems I am not alone. And I am talking about PDE-ML.
The threads on Numerical are much better as they attempt at least to run to completion.

Many ML articles are preaching to the converted. Each to his own, but ML/CS does not understand PDE. Look here (JohnLeM is a expert PDE guy)

https://forum.wilmott.com/viewtopic.php?f=34&t=101716
And the authors never respond.

You can't say we didn't try. 
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

August 26th, 2019, 4:24 pm

Fair point about the PDEs. But there's more to ML than naive attempts at PDE solving.
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

August 26th, 2019, 4:33 pm

What do you think about this work? https://www.pnas.org/content/116/31/15344
 
User avatar
FaridMoussaoui
Posts: 327
Joined: June 20th, 2008, 10:05 am
Location: Genève, Genf, Ginevra, Geneva

Re: If you are bored with Deep Networks

August 26th, 2019, 7:46 pm

They shared the (python) source code here: https://github.com/google-research/data-driven-pdes

There is also a tutorial (python notebook) here: Advection diffusion tutorial
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 27th, 2019, 8:43 am

Fair point about the PDEs. But there's more to ML than naive attempts at PDE solving.
True; but ML-PDE js the talk of the town in the quant sphere.
 
User avatar
Cuchulainn
Posts: 20250
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

August 27th, 2019, 8:48 am

What do you think about this work? https://www.pnas.org/content/116/31/15344
That's interesting (as well a Farid's post. At the moment I am porting C++ code to Python (ADE, CN, MOL) as an aid to understanding PDE, with a focus on maintainabiliy and extendibiliity (read; design pattern).
PDE/ODE is an endangered species. And several MFE programmes demand such savoir faire for entry. And Python is easier to learn than C++ in this regard.