ANC Workshop - 12/11/2019

Chair:  Conor Durkan


Speaker: Borislav Ikonomov

Title: Robust Optimisation Monte Carlo

Abstract: The talk will consider Bayesian inference for parametric statistical models that are defined by a stochastic simulator which specifies how data is generated. Exact sampling is possible but evaluating the likelihood function is expensive. Approximate Bayesian Computation (ABC) is a likelihood-free inference framework to perform approximate inference in such situations. Optimisation Monte Carlo (OMC) has recently been proposed as an efficient method that leverages optimisation to accelerate the inference. The talk will demonstrate an important previously unrecognised failure mode of OMC - it generates strongly overconfident approximations by collapsing regions of similar or near-constant likelihood into a single point - and will propose an efficient, robust generalisation of OMC that corrects this.


Speaker: Cole Hurwitz



In this talk, I will give my perspective on being part of a multi-site, open source software project to improve reproducibility and analysis in neuroscience research. The overall aim of this project, called SpikeInterface, is to solve the task of automatically isolating single-neuron activity from extracellular recordings. Although much development has been directed towards improving the performance and automation of software designed for this task, there are challenges, such as file format incompatibility and reduced interoperability, that hinder benchmarking and preclude reproducible analysis. I will explain how the project begun and talk about our experience so far with developing, maintaining, and promoting this software package.

Nov 12 2019 -

ANC Workshop - 12/11/2019

Borislav Ikonomov and Cole Hurwitz

G.03, IF