Hi! So, I just finished 3 back to back projects in neural network, including supervising a MS thesis work as well. They were based on medical image processing, relating to classification tasks, and another was a multi-modal biometrics data fusion work for same classification and regression. I used MATLAB for 2 of them, and Octave for another one.
Scratch: The Octave one included writing backpropagation from scratch, but used LM linearization for optimization. As you shall know in MATLAB they are by default settings. So, I'm able to understand completely "proximal backpropagation" and "unbiasing backprop" paper too. Anyway, these are just to relate the terms. Similarly, reversible backprop is a feedback included technique that is normally absent in the MATLAB toolkit too.
Objective:
So, my understanding is that you are fine with any ONE of the papers implemented, to test the hypothesis claimed in them.
a) these are methods papers
b) theoretical and mathematical operations
c) not an application development task, so no GUI etc
Regarding your project:
Kindly share more information so that I can give a better estimate of timeline. Below are a few of them:
Queries:
1) Are you looking for any additional improvements etc sort of stuff at the end?
2) What's your deadline?
Which PAPER to implement?
I'm settling on the 1st one in the rack, local feature aggregation one. I'm working currently in feature engineering, hence. We s use CIFAR and VGG-Net as given.
About me:
Am a research assistant in image processing and machine learning, hence my interest in taking this project.
Please initiate a chat so that I can verify everything. IF not online, please leve your messages and queries, I shall reply ASAP. If everything goes well, then I can begin by tomorrow.