Jack J. Mayo
Researcher in Machine Learning and Mathematical Statistics

Hi, and welcome to my page! I'm a first-year PhD at the Korteweg-de Vries Institute for Mathematics at the University of Amsterdam. Under the supervision of Dr. Tim van Erven, I am currently working on the theory and mathematics of modern machine learning.

My aim is to develop novel algorithms, and complementary formal methods which adapt to intrinsic data characteristics of sequential descision making and learning problems, with special emphasis on 

  1. Instance Dependence - Formal guarantees of performace should be near minimax optimal, and yet depend on the natural notion of complexity of the problem at hand.
  2. RobustnessTo the extent that specification is neccesary, algorithms should be designed for robustness to outliers, adversarial attacks, misspecification, and general safety under possible changing circumstances.
  3. Adaptivity Algorithms should detect and leverage simplifying indicators in the domain of application for enhanced performance in the face of "easy" data characteristics.

What I really like is to have algorithms work "out of the box", without the burden of suboptimality and cost associated with hyperparameter tuning. I'm excited to apply the insights I'm deveoping across a variety of domains in machine learning. 

Generally, topics of interest to me at present largely fall into the following categories:

- Parameter-free online learning.

- Bayesian methods, and Online Convex Optimization.

- Reinforcement learning.

I also have some work in nonequilibrium thermodynamics and condensed matter theory, from studies undertaken during my BSc and MSc.

Recent News;

January 26th 2020 - Very excited to be TA'ing the Mastermath course in Machine Learning Theory from February 11th 2021. Tim and Wouter have worked in some new and very current topics under the broad heading of online convex optimisation. Sign up at the above link!

January 26th 2020 - I am co-organizing this years NeurIPS Debriefing along with Tim, which takes place at the end of February 2021. If you're interested in presenting some work (either your own, or something you'd like to emphasize from the conference) feel free to get in contact!

October 5th 2020 - Our work on "Unravelling intra-aggregate structural disorder using single-molecule spectroscopy" was awarded with an Editor's Pick in Journal of the American Chemical Society!

June 17th 2020 - Our work entitled "Full Counting Statistics of Topological Defects after Crossing a Phase Transition" was awarded with an Editor's Choice in PRL!

June 17th 2020 - Smitha Vishveshwara from the University of Illinois at Urbana-Champaign wrote this great article about our work on topological defects in Physics Magazine: Defect or No Defect: It’s a Toss Up!