SQUIDS Seminar - Bayesian Computation when privacy matters
| Dates: | 1 November 2023 |
|---|
| Times: | 14:30 - 16:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Department of Mathematics |
|---|
| Who is it for: | University staff, External researchers, Current University students |
|---|
| Speaker: | Andrew Duncan |
|---|
|
Schedule:
- 2:30pm: pretalk
- 3:05pm: research talk 'Bayesian Computation when privacy matters'
Abstract:
When performing Bayesian computation in practice, one is often faced with the challenge that components of the model and/or the data are spread across a network, and cannot be brought together onto a single machine, e.g., due to privacy concerns or sheer volume. I will present some strategies for performing Bayesian computation in this federated setting, including a new approach based on continuous-time Markov processes, which yields a computation -- and communication -- efficient family of Bayesian inference algorithms which enables full federation, while preserving the asymptotic exactness of classical MCMC. We also investigate scenarios where one of the nodes is compromised and quantify the degree of information leakage if communication between the nodes is intercepted.
Speaker
Andrew Duncan
Organisation: Imperial College, London
Travel and Contact Information
Find event
G.113
Alan Turing Building
Manchester
