The Naked Scientists

The Naked Scientists Forum

Author Topic: Can Field Programmable Gate Arrays (FPGA) be used to design a neural net?  (Read 4052 times)

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
Can Field Programmable Gate Arrays (FPGA) be used to design a neural net - it seems that 2 million neurons represented as configurable logic elements should do so. They can be run at 2 GHz - so a ms or 1 ~ 10 kHz in a biological system may be replicable.

The human brain has about 100 billion neurons so it could be simulated in hardware and 2 million software cells could be reapplied to tasks each millisecond - neural mass times time I guess.

Would we want to replicate humans though? Perhaps the design is a useful basis but imperfect?

Has anyone here put a neural net on silicon? Doing it in MathCad is slow - parallel processing is always better than sequential processing.



Mod edit - formatted the subject as a question.  Please try to do this to help keep the forum tidy and easy to navigate - thanks!
« Last Edit: 23/07/2008 13:03:40 by BenV »


 

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
oops forgot to say thanks to Neilip for the info on posting images!

 

Offline graham.d

  • Neilep Level Member
  • ******
  • Posts: 2208
    • View Profile
You could put some sort of neural net on an FPGA but it would not be very efficient. I would expect this would have been done although I don't know for certain. There will be a paper on the web somewhere I expect. Most neural networks tend to be some sort of learning algorithms based on conventional logic or as a computer emulation. Actual representations of neurons have also been made, but involve a rather horrid mix of analogue and digital circuits which is not very compact. I don't think anyone has made a system which can work in any similar way to how the brain works. The silicon can't behave in any similar way so it all has to be emulated at some level.

I think we are quite a long way from getting close to modelling the human brain or approaching what the AI people call the "Singularity". There was a good set of articles in a recent issue of the IEEE Spectrum magazine which I think you can access on-line even if you are not a member.

http://www.spectrum.ieee.org/singularity

 

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
Hi Graham

http://ian.r.scott.tripod.com/nn_structure.htm

I am not unfamiliar with NN concepts.

There is little need to replicate biology but it has some lessons that we may choose to avoid

 

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
ps

If you place a link at least please make that link relevant to the subject matter!

Why waste peoples time?

Why armchair hypothesize also - there is no intellectual value in this.

I was hoping to find a reply from someone that actually programmed FPGA devices Alterra, Zilinx etc. These are not easy to program and the maths is advanced for what I seek.

Others please don't speculate as you add little value.

 

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
BenV

"I'm afraid I don't really understand this one either - this is an online science discussion forum, with a mixture of scientists and non-scientists who enjoy discussing science.  Sometimes this is science they fully understand, sometimes it is an opportunity to learn.  Does allowing people to learn and get involved constitute nannying?"

Do you think I am evil or uncaring?

Maybe I need to learn - but I am as I am and we cannot easily change our ways in life
 

Offline graham.d

  • Neilep Level Member
  • ******
  • Posts: 2208
    • View Profile
Ian, you wittered on about the human brain and speculated on whether it was worth emulating so why do you expect others to realise that what you wanted was a direct answer to a simple question? Why not say, "I am experienced and Neural Net technology and am interested to know if anyone has direct experience of programming a Neural Net on to an FPGA" and say no more. This way you will target the replies more in the way you want.

You should also note that the forum is not just for your own personal edification so whilst you may consider some of the answers beneath your intellectual level, others, who may not be so well versed in a particular subject, may find other contributions interesting or educational.

To program your FPGA, I would recomend you to get to grips with writing HDL (either Verilog or VHDL) although, from your background, you may well be aware of this. The tools for doing the synthesis are not cheap if you are trying to do this self funded. The bigger FPGAs are also very expensive now although you could prove out a principle on a small one I guess.
 

Offline Ian Scott

  • Sr. Member
  • ****
  • Posts: 101
    • View Profile
graham aren't you a hate child+
get a room
 

Offline Soul Surfer

  • Neilep Level Member
  • ******
  • Posts: 3345
  • keep banging the rocks together
    • View Profile
    • ian kimber's web workspace
Also Ian that topic is probably best placed in the technology forum
 

The Naked Scientists Forum


 

SMF 2.0.10 | SMF © 2015, Simple Machines
SMFAds for Free Forums