View My GitHub Profile


These are some projects that have been successful. The older ones will be summarised in a blog later, they are quite long, I got excited about writing about my first projects and they may be too long. It is also easier to correct mistakes that I made and misunderstandings that I had, than rewrite everything.

Comparing model confidence using noise

Comparing the confidence of models trained with different cost functions and final layers by looking at how well they distinguish noise from legitimate inputs.

Combatting and detecting FGSM and PGD adversarial noise

Testing pre-trained pruned models against the RobustML framework for robustness and confidence estimates.

Output analysis

This blog looks at the output values of models used in blog 1. Outputs can be used to assess the quality of the images,

Pruning, cost functions and adversarial noise

My first blog, where I have changed the cost function and pruned weights using 2 different methods. I will link here soon to a blog summarising this so nobody has to read that much.