Continuing development on the CNN before the due date

The week has passed fast, and a lot has happened. Turns out CNN’s are a little bit more complex than I thought. Although they are great for the whole convolution and finding patterns by defining the convolution and there is great benefit to using them, the implementation isn’t fast as a good understanding of the theory behind it is crucial, especially if you’re trying to implement it from scratch or get back propagation to work.

I found this out a bit too late, that being the week before the assignment was due. I have been working on the CNN from the start of the term, as last term was mainly focused on Winstogram due to the PyCon presentation. Therefore I spent most of term 4 dedicated to machine learning. I started out with the ANN and developed on that for about 1-2 weeks to get the idea of machine learning, but then learnt about convolutional neural networks. As more and more time went on, I realised how far I have been from the endpoint of the project, and how much more I had to do.

I continued to look into the theory behind CNNs to try and understand what it was I was trying to implement, and even down to the final week I still hadn’t gotten a full grip on the theory behind it all. The project is slightly late and will be I think a day late before being submitted for assessment, but I think it was probably worth more than it would lose in marks, therefore I believe it was a good decision to continue working on it.

The project will be finalised and submitted shortly, with the neural network hopefully working. The main thing I need to work on now is the back propagation implementation, and then the CNN should be functional.