Deep learning applications for disaster management and conservation

Speaker: Dr. Olga Isupova (University of Bath)

Title: Deep learning applications for disaster management and conservation

Time: Wed 7th July 2021 at 15:00

Venue: MS Teams


Deep learning has enormous success in a variety of applications, but it normally relies on vast labelled datasets. In this talk I will discuss how deep learning can be applied in the area of disaster management and conservation where there is a lack of well-prepared data. For example, if during the disaster relief campaign all we have is point labels of damaged buildings from volunteers and we would like to train a neural network to label the whole map where damaged buildings are to help disaster responders to plan their actions. I will talk how we can combine two normally separated tasks on crowdsourcing aggregation and classifier training. This talk also demonstrates what different applications can benefit from the use of satellite images if we employ neural networks to process the imagery efficiently.

Short Bio:

Olga Isupova is a Lecturer in AI at Department of Computer Science at University of Bath. Before joining Bath, Olga was a Research Assistant in Machine Learning at the Department of Engineering Science at the University of Oxford. She received her PhD, 2017, from the University of Sheffield and the Specialist (eq. to M.Sc.) degree in Applied Mathematics and Computer Science, 2012, from Lomonosov Moscow State University, Moscow, Russia. Her research is on machine learning for disaster response and environment protection, Bayesian nonparametrics, anomaly detection.

An overview of latent class analysis and its application to multimorbidity clustering

Speaker: Dr. Mizanur Khondoker, UEA

Title: An overview of latent class analysis and its application to multimorbidity clustering

Time: Wed 12nd May 2021 at 13:00

Venue: MS Teams

Abstract: The co-occurrence of two or more health conditions within one person is termed multimorbidity. This poses a substantial burden and challenge on the healthcare system as it may be with poorer health outcomes, more complex clinical management, a higher use of health services and associated cost. Identification of patterns multimorbid conditions is important in multimorbidity research, as the findings facilitate quantifying the effect of multimorbidity on health-related outcomes. This talk will present a model-based clustering method, Latent Class Analysis (LCA), for identifying patterns of multimorbidity. LCA, a special case of a more general structural equation modelling (SEM), will be introduced using combinations of causal diagrams and mathematical notions followed by maximum likelihood estimation and class prediction using maximum posterior probability. An application of the method for identifying multimorbidity patterns will be presented using the UK Biobank community cohort. Application will demonstrate selection of number of clusters using information statistics (AIC and BIC) in a training sample followed by cluster validation using an independent test sample.

Multi-state survival models for disease progression

Speaker: Dr. Ardo van den Hout (UCL)

Title: Multi-state survival models for disease progression

Time: Wed 17th March 2021 at 13:00

Venue: MS Teams

Abstract: Multi-state models are routinely used in research where change of status over time is of interest. In epidemiology and medical statistics, for example, the models are used to describe health-related processes over time, where status is defined by a disease or a condition. In social statistics and in demography, the models are used to study processes such as region of residence, work history, or marital status.

Part of the talk will be an introduction to continuous-time multi-state survival models. I will discuss longitudinal data requirements, the link with stochastic processes, and maximum likelihood inference. An important distinction is whether or not exact times are observed for transitions between the states. In many applications, we do not have exact times and it is important to take this into account in the statistical analysis.

Two applications will be discussed. (i) For an illness-death process, I will illustrate estimating time spent in states. In ageing research, this can be used to distinguish total residual life expectancies from healthy residual life expectancy. (ii) For cancer progression, I will discuss a model based on data from a two-arm design with a screening group and a control group.