Jiri Hron

jiri.m.hron@gmail.com

WHO AM I? A PhD student at the University of Cambridge, supervised by Rich Turner and Zoubin Ghahramani, a member of the Machine Learning Group and Trinity College. For most of my PhD, I was a student researcher at Google Brain in the group of Jascha Sohl-Dickstein, working particularly with himself, Roman Novak and Jeffrey Pennington. Most recently, I was a visiting scholar in Michael I. Jordan's group at UC Berkeley.

MY RESEARCH

  1. Neural network scaling (finding ways to make large NN training easier, and to understand their behaviour).

  2. Human-algorithm and algorithm-algorithm interactions with focus on information retrieval (studying how algorithms affect content creator incentives, and how interacting algorithms optimising different objectives affect each other).

  3. Neural net library (related to 1.), exploration in deep RL, and more (see Google Scholar).

I am a co-author of the original paper that proved deep NNs behave as Gaussian processes when layer widths are sufficiently large.

I am also one of the original authors of Neural Tangents, a neural net Python library which automates computation of large width NN limits, prediction of optimisation paths of wide finite networks, and much more.

Among else, this research enabled training of NNs with thousands of layers, orders of magnitude faster uncertainty estimation, and finding optimal hyperparameters for the largest current NNs using a fraction of compute needed for a single training run.

Wessel and I have worked through the excellent book High-Dimensional Statistics: A Non-Asymptotic Viewpoint by Professor Wainwright . We published our exercise solutions hoping they will help others who want to learn about modern estimation theory.

I wish more people knew about the fantastic and important work being done by the Humane League, the Albert Schweitzer Foundation, and the Animal Charity Evaluators.