OxCSML seminars next week: Kerrie Mengersen & O. Deniz Akyildiz
23 views
Skip to first unread message
Hai Dang Dau
Oct 13, 2023, 6:38:42 AM10/13/23
to oxc...@googlegroups.com, oxcsml...@googlegroups.com
Dear all,
Next week, we welcome two speakers at our OxCSML seminars: Kerrie
Mengersen from Queensland University of Technology and O. Deniz Akyildiz
from Imperial College London. Note that the two talks are on different
days. Please see the details below.
Looking forwards to seeing you. Kind regards,
Saif & Hai-Dang
=================
Speaker: Prof. Kerrie Mengersen (Queensland University of Technology)
Time and date: 3.30 pm - 4.30 pm, Thursday 19 October
Place: Large Lecture Theatre (LG01), Department of Statistics
Zoom: https://zoom.us/j/98992013160?pwd=RUQwUWVMbENXRkxkOHJNNFNVaVlBQT09
Title: Dealing with Sensitive Data
Abstract: Many datasets of interest to statisticians are subject to privacy conditions. This can constrain access, analysis, sharing and release of results. In this presentation, we will consider two ways in which this issue might be addressed. The first is through federated learning, in which the analysis is undertaken in such a way that the data remain in situ and private. The second is synthetic generation of the data, such that the simulated data retains salient characteristics but retains the required privacy. We provide some extensions to the class of models that can be considered in federated learning, and an overview of synthetic generation of tabular data. The exposition of these ideas will be motivated by the creation of an Australian Cancer Atlas.
This research is in collaboration with QUT colleagues Conor Hassan and Dr Robert Salomone, and is funded by the Australian Research Council and Cancer Council Queensland.
Papers:
C Hassan, R Salomone, K Mengersen (2023) Federated variational inference methods for structured latent variable models. arXiv preprint arXiv:2302.03314
C Hassan, R Salomone, K Mengersen (2023) Deep generative models, synthetic tabular data and differential privacy: an overview and synthesis. arXiv preprint arXiv:2307.15424
Time and date: 3.30 pm - 4.30 pm, Thursday 19 October
Place: Large Lecture Theatre (LG01), Department of Statistics
Zoom: https://zoom.us/j/98992013160?pwd=RUQwUWVMbENXRkxkOHJNNFNVaVlBQT09
Title: Dealing with Sensitive Data
Abstract: Many datasets of interest to statisticians are subject to privacy conditions. This can constrain access, analysis, sharing and release of results. In this presentation, we will consider two ways in which this issue might be addressed. The first is through federated learning, in which the analysis is undertaken in such a way that the data remain in situ and private. The second is synthetic generation of the data, such that the simulated data retains salient characteristics but retains the required privacy. We provide some extensions to the class of models that can be considered in federated learning, and an overview of synthetic generation of tabular data. The exposition of these ideas will be motivated by the creation of an Australian Cancer Atlas.
This research is in collaboration with QUT colleagues Conor Hassan and Dr Robert Salomone, and is funded by the Australian Research Council and Cancer Council Queensland.
Papers:
C Hassan, R Salomone, K Mengersen (2023) Federated variational inference methods for structured latent variable models. arXiv preprint arXiv:2302.03314
C Hassan, R Salomone, K Mengersen (2023) Deep generative models, synthetic tabular data and differential privacy: an overview and synthesis. arXiv preprint arXiv:2307.15424
=======================
Speaker: O. Deniz Akyildiz, Assistant Professor at Imperial College London.
Title: Interacting Particle Langevin Algorithm for Maximum Marginal Likelihood Estimation
Time and date: 14.00 to 15.00, Friday 20 October.
Place: Room LG03 (Small lecture theatre), Department of Statistics.
Zoom option: https://zoom.us/j/94450025045?pwd=OElJc0hhREFROFNEZjFUeW1FZnhEdz09
Abstract: I will introduce a class of interacting particle systems for implementing a maximum marginal likelihood estimation (MMLE) procedure to optimize over the parameters of a latent variable model. To do so, I will consider a continuous-time interacting particle system which can be seen as a Langevin diffusion over an extended state space, where the number of particles acts as the inverse temperature parameter in classical settings for optimisation. Using Langevin diffusions, I will present nonasymptotic concentration bounds for the optimisation error of the maximum marginal likelihood estimator in terms of the number of particles in the particle system, the number of iterations of the algorithm, and the step-size parameter for the time discretisation analysis. This is joint work with Francesca Romana Crucinio, Mark Girolami, Tim Johnston, and Sotirios Sabanis.
Title: Interacting Particle Langevin Algorithm for Maximum Marginal Likelihood Estimation
Time and date: 14.00 to 15.00, Friday 20 October.
Place: Room LG03 (Small lecture theatre), Department of Statistics.
Zoom option: https://zoom.us/j/94450025045?pwd=OElJc0hhREFROFNEZjFUeW1FZnhEdz09
Abstract: I will introduce a class of interacting particle systems for implementing a maximum marginal likelihood estimation (MMLE) procedure to optimize over the parameters of a latent variable model. To do so, I will consider a continuous-time interacting particle system which can be seen as a Langevin diffusion over an extended state space, where the number of particles acts as the inverse temperature parameter in classical settings for optimisation. Using Langevin diffusions, I will present nonasymptotic concentration bounds for the optimisation error of the maximum marginal likelihood estimator in terms of the number of particles in the particle system, the number of iterations of the algorithm, and the step-size parameter for the time discretisation analysis. This is joint work with Francesca Romana Crucinio, Mark Girolami, Tim Johnston, and Sotirios Sabanis.
=========================
Reply all
Reply to author
Forward
0 new messages