Hello!
So, I've implemented neutral networks from scratch for some of my past projects, in MATLAB and in Python using numpy. In essence, I've coded weight updates, sigmoid activation, backpropagation, and even other loss functions like RELU and softmax. A few were for similar university projects and one was for a MS Thesis on biometrics authentication. I've used neural networks for image classification and image search as well on the famous Oxford Monuments dataset.
Thus, I'd like to know more about your project, as the things like i/p, o/p vectors or dataset is not clear, as you might know that neural networks requires a supervised dataset to begin with. Similarly, the hidden layers and other things have to be decided based on the kind of input and classes to be learnt. This is again, unfortunately, not properly mentioned in the docx. As a matter of fact, I belong to Robotics background, so am really interested to learn the whole setup, and the modules of it, to begin with here.
Tools:
Python 2 in Anaconda
LaTeX for documentation
MATLAB for algorithm development
Milestones:
It should be 3, over 2 week or higher (whichever is suitable) period, as I've assumed based on previous work.
Kindly message me whenever possible. If I'm not online, shall reply you ASAP. Thanks.