Event Details

Dr Christopher Nemeth and Dr James Edwards will present on their research in the university statistics section.

Chris Nemeth

Title: An introduction to scalable MCMC for tall data

Abstract:

Monte Carlo methods, and in particular Markov chain Monte Carlo techniques, have been the gold standard computational tool for Bayesian modelling over the past 30 years. These algorithms can be applied in general settings, from identifying traits in phylogenetic trees to detecting Earth-like planets in distant solar systems, their supporting theoretical guarantees have led them to be widely used by scientists and industry practitioners alike. However, a significant drawback is that traditional Monte Carlo algorithms scale poorly with large datasets, leading to a computational cost that grows at least proportionally with the data size. This leads to a prohibitive cost for modern-day machine learning and data science applications and has led practitioners towards scalable approximate alternatives, such as variational methods, which have no theoretical guarantees on the resulting approximation error.In this talk I'll give a brief introduction to one popular approach for scalable MCMC, known as stochastic gradient MCMC. This class of methods maintain many of the favourable theoretical properties of standard MCMC methods and are easily generalisable to real-world datasets and industrial-scale models.

James Edwards

Title: Learning and Earning with Dynamic Pricing

Abstract:

Choosing a price for a good or service that maximises revenue necessitates knowing how demand will vary with price. This is rarely known in advance and so must be learnt by observing sales. Choosing a single price will only give good information about demand local to that price so there is the need for both appropriate statistical modelling and price experimentation. However, any experimentation comes with a cost since it involves real money and sales. Dynamic pricing considers this problem of adaptively choosing prices to balance good short term results with learning for better long term outcomes. Even very simple versions of this problem can be surprisingly hard to solve and, furthermore, in practical applications standard simplifying assumptions usually break down. I will give an introduction to this area and some of the challenges, both academic and practical, that I have met whilst doing research for a real world pricing application.