Dominic Richards

Research Interests

I am interested in the understanding the interplay between statistics and computation in the context of machine learning. Recently, I have been investigating the performance of estimators arising from gradient descent in settings where the learning problem is more benign. I have also worked on the interplay between statistics and network topology in context of decentralised machine learning.

• Google Scholar CV

Papers

• Comparing Classes of Estimators: When does Gradient Descent Beat Ridge Regression in Linear Models? (with E. Dobriban and P. Rebeschini) (submitted) [Paper]
• Stability & Generalisation of Gradient Descent for Shallow Neural Networks without the Neural Tangent Kernel (with I. Kuzborskij) (NeurIPS 2021) [Paper]
• Distributed Machine Learning with Sparse Heterogeneous Data (with S. N. Negahban and P. Rebeschini) (NeurIPS 2021)
• Learning with Gradient Descent and Weakly Convex Losses (with M. Rabbat) (AISTATS, 2021) [Paper]
• Asymptotics of Ridge(less) Regression under General Source Condition (with J. Mourtada and L. Rosasco) (AISTATS, 2021) [Paper]
• Decentralised Learning with Random Features and Distributed Gradient Descent (with P. Rebeschini and L. Rosasco) (ICML 2020) [Paper] [Code]
• Optimal statistical rates for decentralised non-parametric regression with linear speed-up (with P. Rebeschini) (NeurIPS 2019). [Paper]
• Graph-dependent implicit regularisation for distributed stochastic subgradient descent (with P. Rebeschini) (JMLR) [Paper] [Code]

Talks

• 16/10/2020 - Asymptotics of Ridge(less) Regression under General Source Condition (Department of Applied Mathematics and Theoretical Physics, University of Cambridge)
• 22/07/2020 - Asymptotics of Ridge(less) Regression under General Source Condition (Montréal Machine Learning and Optimization Group)
• 04/06/2019 - Decentralised Sparse Multi-Task Regression (Numerical Analysis Group Internal Seminar, Mathematical Institute, University of Oxford)
• 04/05/2019 - Optimal Statistical Rates for Non-parametric Decentralised Regression with Distributed Gradient Descent (Yale Probabilistic Networks Group, Yale University)