Dear all,

it is time again for a new talk at the One WorldABC Seminar!

Our next speaker will be Michael Gutmann, who will talk about Neural Approximate Sufficient Statistics, on Thursday October 28, at 11.30am, with an abstract reported below.

To join the talk, please click on the Zoom link below

https://ed-ac-uk.zoom.us/j/82567695825
Meeting ID: 825 6769 5825
Passcode: hq1z8WYx

If you miss last talk by Matias Quiroz, you can access its recording here

https://www.youtube.com/watch?v=SBAqwEID6Lo

We look forward to seeing you on Thursday,

best wishes,

Massimiliano on the behalf of the One World ABC Seminar Organisers.

---

When: Thursday, October 28, 11.30am UK Time

Speaker: Michael Gutmann, University of Edinburg

Title: Neural Approximate Sufficient Statistics

Abstract: My talk will first review the concept of sufficient statistics and then explain that we can learn them by learning mutual information maximizing representations. I will then explain how we used the learned statistics to boost the performance of both classical and recent methods for Bayesian parameter inference when the likelihood is intractable but sampling from the model is possible.

The talk is based on the paper:
Neural Approximate Sufficient Statistics for Implicit Models. Y. Chen, D. Zhang, M. Gutmann, A. Courville, and Z. Zhu. In International Conference on Learning Representations (ICLR) 2021
https://openreview.net/pdf?id=SRDuJssQud