SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
User avatar
outrun
Posts: 3913
Joined: April 29th, 2016, 1:40 pm

Re: Special Topics by Outrun

September 26th, 2017, 1:00 pm

Yes, numbersytems are symbolic representations of the actual numbers. We somehow all agree in the meaning of those actual numbers, probably based on how those numbers relate based on operators like +? Eg Apple + banana = pineapple
 
User avatar
Traden4Alpha
Posts: 22965
Joined: September 20th, 2002, 8:30 pm

Re: Special Topics by Outrun

September 26th, 2017, 1:28 pm

Exactly!

And it can lead to constructs like abstract algebras in which we define new sets of "numbers" and alternative definitions of "+" and "*" to create things like multi-valued logics, fuzzy logics, rings, fields, etc. for nonmonotonic reasoning, cryptography, state machine design, computer circuit design, etc.

A number need not represent a quantity. It could be an index on to a more abstract set. (But, of course, that must be clearly stated so that everyone understands how the number-symbols and operator-symbols have been overloaded with different functionality for the purposes of a given research effort or field of math).
 
User avatar
outrun
Posts: 3913
Joined: April 29th, 2016, 1:40 pm

Re: Special Topics by Outrun

October 5th, 2017, 4:07 pm

Excited about the letter we got: all Dutch will soon get a govt supplied batch of iodine pills in the mail!
 
User avatar
Traden4Alpha
Posts: 22965
Joined: September 20th, 2002, 8:30 pm

Re: Special Topics by Outrun

October 12th, 2017, 11:39 am

outrun wrote:
Excited about the letter we got: all Dutch will soon get a govt supplied batch of iodine pills in the mail!

Looks like it may be just in time:
https://www.thesun.co.uk/news/4659356/s ... om-russia/
 
User avatar
outrun
Posts: 3913
Joined: April 29th, 2016, 1:40 pm

Re: Special Topics by Outrun

October 12th, 2017, 11:51 am

Traden4Alpha wrote:
outrun wrote:
Excited about the letter we got: all Dutch will soon get a govt supplied batch of iodine pills in the mail!

Looks like it may be just in time:
https://www.thesun.co.uk/news/4659356/s ... om-russia/

Wow, I'd better eat the whole batch right away!
 
User avatar
Traden4Alpha
Posts: 22965
Joined: September 20th, 2002, 8:30 pm

Re: Special Topics by Outrun

October 12th, 2017, 12:50 pm

outrun wrote:
Traden4Alpha wrote:
outrun wrote:
Excited about the letter we got: all Dutch will soon get a govt supplied batch of iodine pills in the mail!

Looks like it may be just in time:
https://www.thesun.co.uk/news/4659356/s ... om-russia/

Wow, I'd better eat the whole batch right away!

And bring home 50 kg of lead sheeting for the egg chair on your bicycle.
 
User avatar
outrun
Posts: 3913
Joined: April 29th, 2016, 1:40 pm

Re: Special Topics by Outrun

October 27th, 2017, 3:41 pm

I've been trying to buy one of those plastic divider bars at the supermarket for half an hour now! She keeps putting it back.
 
User avatar
Traden4Alpha
Posts: 22965
Joined: September 20th, 2002, 8:30 pm

Re: Special Topics by Outrun

October 27th, 2017, 4:44 pm

outrun wrote:
I've been trying to buy one of those plastic divider bars at the supermarket for half an hour now! She keeps putting it back.
Put it in a baguette bag with a barcode-label and you'll breeze through.
 
User avatar
outrun
Posts: 3913
Joined: April 29th, 2016, 1:40 pm

Re: Special Topics by Outrun

December 1st, 2017, 9:20 pm

"A year in machine vision", a very nice overview of the most significant computer vision advances this year.
 http://www.themtank.org/a-year-in-computer-vision

So many spectacular results. Reminds me of Bladerunner
Image


"Rotate"
Image

"enhance"
Image
 
User avatar
ISayMoo
Posts: 351
Joined: September 30th, 2015, 8:30 pm

Re: Special Topics by Outrun

December 2nd, 2017, 12:41 pm

I have a slight issue with the fact that all those groundbreaking image recognition models are tested on the same unchanging dataset (ImageNet). #overfitting
 
User avatar
Cuchulainn
Posts: 53307
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Special Topics by Outrun

December 3rd, 2017, 1:27 pm

ISayMoo wrote:
I have a slight issue with the fact that all those groundbreaking image recognition models are tested on the same unchanging dataset (ImageNet). #overfitting

Noobie question: is it possible to learn NN without studying images of baboons' derrieres? Or are the techniques  indispensible for an understanding of other, more numerically-oriented areas? I don't really feel like brushing up computer graphics and CAD at the moment.
http://www.datasimfinancial.com

“Sir Walter Scott created rank & caste in the South and also reverence for and pride and pleasure in them"

Mark Twain
 
User avatar
ISayMoo
Posts: 351
Joined: September 30th, 2015, 8:30 pm

Re: Special Topics by Outrun

December 3rd, 2017, 1:53 pm

Of course it is.
 
User avatar
Traden4Alpha
Posts: 22965
Joined: September 20th, 2002, 8:30 pm

Re: Special Topics by Outrun

December 3rd, 2017, 1:55 pm

Cuchulainn wrote:
ISayMoo wrote:
I have a slight issue with the fact that all those groundbreaking image recognition models are tested on the same unchanging dataset (ImageNet). #overfitting

Noobie question: is it possible to learn NN without studying images of baboons' derrieres? Or are the techniques  indispensible for an understanding of other, more numerically-oriented areas? I don't really feel like brushing up computer graphics and CAD at the moment.
You can learn NN using any type of data you prefer. Image recognition is popular because that's traditionally been an "impossible" task for computers and an easy task for humans. Likewise speech recognition is popular in NN circles. Those are potentially very high value ML systems for a wide range of applications.

But other examples range across:
* data center power management
* predictive maintenance
* fraud detection in financial transaction systems
* customer service (providing helpful replies based on customer queries)
* forecasting
* business process automation
* agriculture (optimizing sowing date, fertilizer, irrigation, pesticides, etc.)

You can train an NN on any set of {[X,Y]} you like and see how well it interpolates/extrapolates Y = f(X) assuming you have (or retain) additional out-of-sample values of {[X,Y]} to testing.
 
User avatar
katastrofa
Posts: 4963
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Special Topics by Outrun

December 3rd, 2017, 2:20 pm

Cuchulainn wrote:
ISayMoo wrote:
I have a slight issue with the fact that all those groundbreaking image recognition models are tested on the same unchanging dataset (ImageNet). #overfitting

Noobie question: is it possible to learn NN without studying images of baboons' derrieres? Or are the techniques  indispensible for an understanding of other, more numerically-oriented areas? I don't really feel like brushing up computer graphics and CAD at the moment.

I use NN to discover data structure for my simulations (some data dependencies have characteristic shapes, which NN can parameterise and e.g. extrapolate).
 
User avatar
ISayMoo
Posts: 351
Joined: September 30th, 2015, 8:30 pm

Re: Special Topics by Outrun

December 3rd, 2017, 2:22 pm

Neural networks are also used in modern reinforcement learning models (e.g. AlphaGo), to approximate the policy or value function.
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...