Raizada Lab

• Home
• People
• Papers
• Teaching
• News
• Jobs

Some tutorial Python and Matlab neural network programs

Here are a few extensively commented Python and Matlab programs that I wrote, which I hope might be useful for teaching a course on how to program neural nets. Please feel free to mail me with any questions or comments.

• Backpropagation_using_only_NumPy.ipynb. This program shows step-by-step how to write classic Rumelhart & McClelland backprop using just matrices and vectors. This code might be useful for teaching the basic structure of the backprop algorithm, or for showing how to avoid for-loops by vectorising. It is a live Python notebook, hosted on Google Colab, so it should run inside any web browser, without needing any additional software (i.e. you don't even need to have Python installed on your computer).


• bkprop.m - The original Matlab version. I posted this code to the newsgroup comp.ai.neural-nets and it was subsequently used for teaching in a course at the University of Stirling.
  
  
• sofm.m - This program shows how a Self-Organising Feature Map (or SOFM, also often called a Kohonen map) can learn a particular input distribution, in this case a letter 'A'.
  
  
• fuzzyartmap_demo.m - This program shows graphically how Fuzzy ARTMAP learns the circle-in-the-square task by growing category boxes in response to the presented inputs. Detailed comments explain the algorithm's match, reset and learning stages.