Projects

These are some of the projects I've tried out for fun.

• Using multiple GPUs in Tensorflow

  An alternative to using the tf.distribute.Strategy API approach that has been useful for me.
  
  Using multiple GPUs in Tensorflow (Blog post - approx 10 mins read)



• TrueSkill

  Neat application of applying a Bayesian approach to ML. I wanted to understand the algorithm Microsoft use to rank players in online matchmaking so I implemented the original TrueSkill paper.
  
  Understanding TrueSkill (Blog Post - approx 2 hours read)
  On ranking, factor graphs, expectation propagation and how useful Gaussians are! One acid test for me to see if I've learnt something is to try explain it to someone else so this blog post is my attempt at that. The original paper is old in ML years but super interesting so please check it out!
  Python implementation of TrueSkill (Repo)



• Network Pruning

  There is a lot of interesting literature on sparse networks. In particular I enjoyed this paper on the lottery ticket hypothesis from ICLR 2019. Had a crack at some weight and network pruning experiments in Colab.
  
  Python pruning notebook (Notebook)
  The notebook can be opened in Colab so please have a go at running through yourself! I learnt several cool things in this project like how to stop weight updates on pruned weights and did a bit of exploration into taking advantage of sparsity in Tensorflow to speed up inference.