Bayesian nonparametric hidden Markov models for remote monitoring applications

Seminar on "Bayesian nonparametric hidden Markov models for remote monitoring applications" by Yordan Raykrov, Aston University

Abstract: In recent years, sensors embedded in smartphones, wearables and other consumer devices have become ubiquitous and have evolved to the point where they can be used in areas such as healthcare, security, environmental monitoring and transport. However, the principled analysis of large amounts of unconstrained time series monitoring data is difficult due to noise, lack of labels and large amount of uncertainty associated with the observed measurements. Applying black box unstructured methods leads to poor understanding of derived outcomes motivating the need for principled statistical models for reasoning in such data. In this talk, we will look at how Bayesian nonparametric hidden Markov models (HMMs) can be trained and augmented to handle some common problems in health monitoring using IoT sensors. The talk will start with a gentle review of nonparametric HMMs. Then we will look at two specific problems related to quality control of clinimetric testing and passive monitoring of individuals. We will consider how novel approximate scalable inference algorithms for nonparamtric HMMs can be used to train high order models and we will propose a new adaptive input HMM which can account for contextual factors affecting the monitored measures. 

Short bio: Yordan P. Raykov received BSc in mathematics from University of Leicester, UK and PhD in machine learning from Aston University, Birmingham, UK in 2013 and 2016 respectively. During his PhD, he developed scalable inference algorithms for some of the most common Bayesian nonparametric models. He then joined the R&D team in ARM Cambridge to work on the development of novel embedded device for occupancy estimation. Following ARM, he did a pharma funded postdoc with Aston University and Radboud unversity medical center where he developed probabilistic models for unsupervised and semi-supervised learning, with focus on health monitoring applications. From 2018, Yordan is a lecturer in the Mathematics Department in Aston where he studies scalable nonparametric graphical models and hybrid models for latent structure discovery across multiple domains including: health monitoring, medical imaging,  proteomics and finance.