Diego Granziol

My main interest relies on the application of information theoretic/statistical mechanics concepts to a wide array of machine learning problems. Recent work has focused on the spectral approximation of large matrices, using a combination of stochastic trace estimation and the method of maximum entropy to form approximate, asymptotically optimal and cheap estimates of log determinants, matrix inverse traces and matrix inner products. Other work has looked at creating Novel bounds for the differential entropy of Gaussian mixture models, with applications in text summarisation, parameter estimation, tracking and Bayesian optimisation in deep neural networks, along with efficient community detection algorithms. Financial work includes a novel options pricing model using MaxEnt. Prior to joining the robots group, I completed a masters in Physics at Oxford and worked in the city of London for Goldman Sachs and CQS.


Published Research

Granziol, D and Roberts, S (2017). Entropic Determinants of Massive Matrices. arXiv. 1709.02702. ..
Granziol, D and Roberts, S (2017). Entropic Determinants. Proceedings of IEEE Big Data . 2017. ..