Speaker: Paul Fearnhead (Lancaster)
Title: Statistical and computational challenges in changepoint detection
Abstract: Big data introduces some new statistical and computational challenges for changepoint detection. One example of this is the need to develop algorithms that scale well (close to linearly) with the amount of data. This talk will present a new dynamic programming approach for estimating changepoints using penalised likelihood methods that can allow for dependence across segments. It will also discuss some other generic issues that are particularly relevant for big data, such as dealing with model error, and the computational/statistical trade-off.