Neural Nets

Go To Last Post
8 posts / 0 new
Author
Message
#1
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Guys, I am seriously thinking of a project involving Neural Nets.  It's an interesting topic which I covered in University a long time go.

What would be a simple example of the use of a neural network?

 

Maybe a virtual car trying to traverse a virtual course!

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Things I need to learn is how do you decided the actual architecture of the Neural network, I.e how many nodes and layers.

Edited: I guess the number of inputs would detected the architecture.

Last Edited: Tue. Mar 10, 2020 - 07:42 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Get yourself a raspberry pi, camera &  install openCV & Tensorflow...the sky is the limit.

 

Or just get this:

https://aiyprojects.withgoogle.com/vision

When in the dark remember-the future looks brighter than ever.   I look forward to being able to predict the future!

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 1

Fianawarrior wrote:
how do you decided the actual architecture of the Neural network,
Maybe have a look at:

 

https://www.renesas.com/in/en/solutions/automotive/soc/r-car-v3h.html

 

That contains a CNN unit ! From Renesas own marketing material:

 Convolutional neural networks (CNNs), are deep, feed-forward artificial neural networks that have been successfully applied for analyzing visual imagery and are increasingly used in automotive for applications like road detection or object classification.

More about Renesas e-AI here for example:

 

https://www.embeddeduse.com/2019/03/15/running-neural-networks-on-microcontrollers/  

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Thanks for the suggestions lads, but I think I'll learn from scratch.  Much more fun in learning.

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Thanks for the suggestions lads, but I think I'll learn from scratch.  Much more fun in learning.

Yep, all you really need is pencil & paper, since it is all math-based. 

 

More seriously, you can use programs like matlab (octave) to implement the calculations.   Use a neural net to determine when your hamburger is done, based upon several input factors.  It can learn when you cook & provide many batches of sample data.

When in the dark remember-the future looks brighter than ever.   I look forward to being able to predict the future!

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Fianawarrior wrote:
Thanks for the suggestions lads, but I think I'll learn from scratch.  Much more fun in learning.
Yeah but the point is this - Convolution Neural Networks really are going to be found in consumer projects but it won't be a "software only" thing. It will be backed by "special silicon" like the solutions from Renesas so the "skilled neural engineer" is going to be the person who knows how to drive silicon like that from Renesas and others.

 

Sure it wouldn't hurt to learn the theory of NN by some simple procedural code in C/C++ but the chances are these things are going to be a whole new approach/language which is going to need engineers trained in that specific skill,

 

As I work in an "intense computing" environment I see this already with parallelism. Various manufacturers are producing special silicon for image processing that requires a whole new mindset to program it and get the best out of it. While you can try and retrain C/C++ engineers to use these things what we can't find for love nor money (which means it could well be "loads of money" when we do find the right engineers) are "parallel processing experts". I imagine, now the trrend is towards neural nets that we're going to hit the same with those - no one who really understands the things. So an engineer that can get themselves into the position will likely have a very rosy future!

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Maybe use C++.  It will be my first time using C++ in anger.