ANC Workshop - Vaidotas Simkus, Sandy Nelson
Tuesday, 28th November 2023
Variational Gibbs Inference for Statistical Model Estimation from Incomplete Data - Vaidotas Simkus
Abstract: Over the last decade we have experienced a boom of highly expressive statistical models powered by neural networks. These modern models have demonstrated exceptional performance in representing clean, fully-observed data, while the real world often presents us with data sets riddled with missing data. But, these state-of-the-art models are typically specified under the assumption of complete data, posing significant challenges in adapting them to handle incomplete data. In this talk, I will uncover some of the challenges of efficiently estimating these statistical models from incomplete data and present our recent efforts aimed at mitigating these issues via a new method called variational Gibbs inference.
Parameter inference in synthetic biology with ensemble MCMC methods - Sandy Nelson
Abstract: MCMC methods and ensemble methods in particular are popular in synthetic biology for parameter inference. I will present some results from simulation experiments comparing an ensemble sampler with affine invariance (Emcee) to Hamiltonian Monte Carlo (HMC). I will contrast the two methods sensitivity to dimensionality and choice of prior.
Event type: Workshop
Date: Tuesday, 28th November
Time: 11:00
Location: G.07
Speaker(s): Vaidotas Simkus, Sandy Nelson
Chair/Host: Robbie Court