Sparana is a software library I built to do experiments on artificial neural networks. These are some blogs about those experiments, and the software. I would recommend starting with the summary blog then moving on to the others if you want more information.
Summary blog
This is a summary of all of the experiments that I have put a significant amount of time into
Targeting Single Datapoints
Here I am trying to correct misclassified datapoints by pruning a small number of parameters.
Forgetting
Removing parameters to make neural networks forget stuff. The end goal is to target parameters that contribute to unwanted behaviour while fine tuning a model. This is my current set of experiments that I am working on with some promising results to start with.
Removing Activations
These are another set of ongoing experiments that I am slowly working on. Pruning models means that sometimes I can remove activations that aren’t doing anything, This leads to some interesting results.
I need to make some plots, but I have a couple ready to go soon.
The Basics
These are some techniques and definitions that I need to explain, but don’t need to put in every blog. A sort of resource. Might be helpful to understand how and why I do what I do.
About Sparana
Here is a blog about how I built the software, and a bit about adding a module. If you want to know a bit about how it works. The software was never really intended to be used by anyone else, but if you want to take the techniques from the code, it might help.
About Me
Bio and contact info. Would love to get some feedback.
Other Stuff
I have some old experiments I will write up when I have time.