Neural Networks and Reservoir Computing

[kbharadwaj]

Hi,

I am taking a Neural Networks course this Fall in my university and my instructor is very inclined towards its application in the field of finance. I would like to know (from the ones who have applied neural networks for finance and from others who are in the field of quant finance) that how good is neural networks in the matter of forecasting and how good is it valued regarding the job market?

I am a FE major and am trying to fit Neural Networks in to my field of study so that I could apply it to Finance. My instructor said that he would like to apply reservoir computing to time series analysis using neural networks and that gives by far the most precise estimation when compared to the other methods using neural networks. He asked me to start coding in C. So do you think C is a good base considering C++ is better and has OOP advantage?

 

[lalala]

I am not from finance, but I think my research field is kind of moving from NN to SVM. NN has been around for quite some time, but seems that not many really understand the principles behind beyond neurons. SVM comes in as a very 'scientific' method for machine learning. Basically the idea of finding an optimal hyperplane to separate the classes in high dimensional space. I think the key advantage here is that we know clearly how things are done. Feature is defined, mathematics is solid. About research, seems that my field is crazy about kernel nowadays:P

Personally, I believe NN is the future of machine learning, but we might need some more inputs from neuro or other communities. I guess the brain models now might not be accurate/strong enough. Anyway, I don't think human brain learns in a very scientific way. Or at least not science defined currently.

If I recall correctly, I think there are uses on credit fraud detection of SOM. Just happen to get this website, you might want to have a look Support Vector Machines: Financial Applications

Just my two cents worth, I guess C has the advantage because of its real time use (so in many parallel computing stuffs, we can do scheduling bla bla bla), ability to play with memory stuffs, and of course speed and load. I am just a beginner to C++, not sure about it too. But C is definitely a popular real time language in embedded systems, I wrote a space invader game on an ARM9 processor board using C in my coursework Fun! And perhaps popular libs and support everywhere now? STL...

 

[kbharadwaj]

I agree that there are fewer inputs. But I am participating in this competition at the end of the semester which is all about stock forecasting using NN and my instructor said we should use Reservoir Computing for the application to be more accurate.
There is also a book called "Visual Explorations in Finance" which is all about SOM's, if you know what I mean. Even Som's are good but not practical. SVM is a good, however, it is more into data mining.

---------- Post added at 08:52 PM ---------- Previous post was at 08:51 PM ----------

Just my two cents worth, I guess C has the advantage because of its real time use (so in many parallel computing stuffs, we can do scheduling bla bla bla), ability to play with memory stuffs, and of course speed and load. I am just a beginner to C++, not sure about it too. But C is definitely a popular real time language in embedded systems, I wrote a space invader game on an ARM9 processor board using C in my coursework Fun! And perhaps popular libs and support everywhere now? STL...

C sure has the advantages that you talk about. But what about OOPs while dealing with something as concrete as stock forecasting?

 

[Daniel Duffy]

"So do you think C is a good base considering C++ is better and has OOP advantage? "

Yes, C is a god basis.

C++ (and OOP which is overhyped at times) will come in time.

<!-- / message -->

 

[kbharadwaj]

Thank you Prof. Daniel Duffy. I would be using C for my neural networks. In fact I started using it. Again, what do you think about neural networks in finance in general?