A) Talks to the LGC
1. January 24th, 2024, Peter Froggatt Centre , QUB .
Jonathan Henderson of Queen's University Belfast gave a talk to the Local Group entitled:
'Detecting Glaucoma Early: The Power of Fractal Measures and Change Point Detection in
Retinal Assessment.'
This study explored the potential of fractal geometry within the context of point patterns, showcasing their efficacy in identifying glaucoma in its incipient stages using rodent disease models. The use of these fractal measures is further considered in constructing model parameters for the disease process based on an approximate Bayesian computation algorithm. This approach allows for a change-point detection algorithm to be applied that can be used to identify regions of the retina at risk of future loss. This can be used to inform clinical practice through earlier diagnosis of disease and better prediction of therapeutic intervention efficacy.
This was an excellent talk full of interest and innovation and was very well received.
It was a hybrid event with Jonathan present with an in-room audience of 10 and an online audience of 12. The talk is available on our
YouTube channel.
2. February 21st, 2024, Peter Froggatt Centre, QUB.
Dr. Brieuc Lehmann of University College, London gave a talk to the Local Group entitled: 'Fusing experimental and observational data with multi-task Gaussian processes to estimate heterogeneous treatment effects .'
Randomised control trials (RCT) are widely considered the gold standard for making causal predictions and assessing treatment effects, due to the internal validity guaranteed from randomisation. However, the strict eligibility criteria lead to small samples not representative of the target population and as a result to lack of external validity. In contrast, observational studies, which typically come in larger sample sizes, are often more representative of the target population, but lack internal validity due to confounding leading to biased results. Brieuc presented a Bayesian nonparametric approach to obtain estimates of heterogeneous treatment effects for the target population. The approach is based on multi-output Gaussian Processes, where the limited RCT observations are used to correct the confounding effect of the observational data.
This excellent talk fell into the LGC 's planned series of talks contrasting RCTs with Observational studies. Brieuc's particular potential outcomes approach and subsequent analysis promises to be a useful line of attack.
The talk was well received by a small in-room audience of 7 with 20 participants on-line, and it is available to
watch online.
3. March 14th, 2024. in 18 College Green, near QUB.
Dr. Nema Dean from the University of Glasgow, Scotland, gave a talk to the Local Group entitled: 'Automatic Architecture Selection for Hierarchical Mixture of Experts Models'
Hierarchical mixture of experts (HMEs) models are an example of a tree-structured modelling technique based on the divide and conquer principle. They present a powerful prediction methodology due to their flexibility and ability to adapt their complexity. They also allow for interpretability that can be attractive in practice. However, deciding on the architecture or form of the model/tree is not a trivial task. Rather than fitting large numbers of competing models and choosing via some performance metric, we propose growing trees during the model fitting process instead of selecting the architecture in advance.
We present a flexible and adaptive way of performing automatic architecture selection using reversible jump in the Bayesian inference setting. The common issue of low acceptance rates for reversible jump algorithms is ameliorated with an adaptive smart proposal mechanism. The results of applying this and competing approaches is presented for Glasgow rental property prices data.
Nema's interesting application of RJMCMC was enjoyed by the local in-room audience of 9, some of whom were working on this topic and really appreciated Nema coming in person to deliver the talk. There were 9 participants online.
View the
recording of the talk.
4. April 17th, 2024, in the Peter Froggatt Centre, QUB.
Dr. Cliona McDowell from the Northern Ireland Clinical Trials Unit gave a talk to the Local Group
entitled: 'The importance of Randomised Controlled Trials and the Northern Ireland Clinical Trials Unit'
The talk dealt with the importance of RCTs and the current portfolio in the NICTU. It covered the types of trials taken on, funding streams, the clinical trial pathway. It also emphasised the variety of expertise held in the NICTU with a specific emphasis on Statistics.
Cliona gave her talk in person. She highlighted the workings of an active CTU and showed, inter alia, the importance of the regulatory framework governing the design of randomised controlled trials (in contrast, notably, to the lack of such a framework governing observational studies). Cliona also produced a table showing the current state of the CTU's trial activity in Northern Ireland – there were approximately 20 clinical trials open or being set up. This was one measure of NI's contribution to innovation in the field. The table excluded cancer trials which were organised by another unit and not always home-grown.
The talk, attended by an in-room audience of 9 and an on-line audience of 12, was very well received and can be
viewed online.