The main code (and paper) is found in the final folder.
We present an alternative technique for similarity estimation under locality sensitive hashing schemes. By utilising control variates, we are able to achieve better theoretical variance reductions compared to methods that rely on maximum likelihood estimation. We show that our method obtains equivalent results, but slight modifications can provide better empirical results at lower dimensions. Finally, we compare the various methods' performances on the MNIST and Gisette dataset, and show that our model achieves better accuracy and stability.