Abstract: The Bayes Factor is a Bayesian tool for comparing two hypotheses, which is gaining popularity in psychological research and being suggested to replace classical t-tests. However, the Bayes Factor requires the specification of a prior distribution for the parameter of interest, which cannot be done unambiguously. In many situations, when further research is needed, information is not complete. This problem can be solved in the context of imprecise probabilities by using only the available (incomplete) knowledge. In this approach, a set of prior distributions is used instead of a single prior, yielding a set of Bayes Factor results, which is called the Robust Bayes Factor.
 
In my talk, I will present the result of a project, in which the Bayes Factor was generalized to imprecise probabilities in a two-sample context with normally distributed data. The effect size between the two groups serves as parameter of interest and its prior was modeled as a set of normal distributions.

 
 
 
 
Abstract: Usually the presence of uncertainty attached to an event is represented by a probability distribution according to probability theory. This approach has proved to work just fine when all the parameters and data needed for the analysis are provided. However, when information is scarce, vague, or conflicting, a generalization of probability theory must be employed since it is hard to find a unique probability distribution. For this reason, imprecise probabilities become important because they can represent the available knowledge as well as provide the tools to model and work with weaker states of information. In this talk, I will briefly describe my impressions and experiences with Imprecise Probability methodologies during the summer school that took place in the Sciences Faculty of the University of Oviedo in Spain.

About Sarah: I am an Associate Clinical Professor in Vascular Rheumatology, and Honorary Consultant Rheumatologis.

 
 
 
Abstract: Computer models are often created because we need to make extrapolations in regions where there is no data. This makes validation challenging - how can we ensure that the model is suitable if it is to be applied in a region where there is no measurement data? The current paper proposes a method which can reveal flaws in a model which may be difficult to identify by other calibration/validation approaches. It specifically targets the situation where one is attempting to model a dynamical system which, it is believed, possesses some time-invariant calibration parameters. With our method we essentially allow these parameters to vary with time, even though it is believed that they are time-invariant. It is through this approach that we aim to identify key discrepancies - indications that a model has inherent flaws and, as a result, should not be used to influence decisions in regions where there is no data. The proposed method isn't necessarily a predictor of extrapolation performance, rather, it is a stringent test that, the authors believe, should be applied before extrapolation is attempted. The approach could therefore form a useful part of wider validation frameworks in the future.
 
Bio: Dr Peter Green (PG) became a lecturer in Uncertainty and Engineering at the UoL in 2015. His background is in structural dynamics, but he now develops uncertainty quantification and machine learning methods for engineering disciplines.

Lecture of the Imprecise Tuesday Colloquium Series

Validation and predictive capability of imperfect models with imprecise data

Abstract: Many sophisticated models in engineering today incorporate randomness or stochasticity and make predictions in the form of probability distributions or other structures that express predictive uncertainty. Validation of such models must contend with observations that are often sparse or imprecise, or both. The predictive capability of these models, which determines what we can reliably infer from them, is assessed by whether and how closely the model can be shown to yield predictions conforming with available empirical observations beyond those data used in the model calibration process. Interestingly, a validation match between the model and data can be easier to establish when the predictions or observations are uncertain, but the model’s predictive capability is degraded by either uncertainty. It is critical that measures used for validation and estimating predictive capability not confuse variability with lack of knowledge, but rather integrate these two kinds of uncertainties (sometimes denoted ‘aleatory’ and ‘epistemic’) in a way that leads to meaningful statements about the fit of the model to data and the reliability of predictions it generates.

Scott Ferson is director of the Institute for Risk and Uncertainty at the University of Liverpool in the UK. His research focuses on statistical tools when empirical information is very sparse or imprecise. This talk is a reprise of an address at the ASME 2018 Verification and Validation Symposium held two weeks ago in Minneapolis.

 
 
 
 
 
 
Abstract: Sensors used to capture time-history data will never provide perfect digital reconstructions of the the processes they originally recorded. At best, a sensor will have an ideal working tolerance and defined accuracy bounds, and at worst will fail, leaving gaps in the data. When estimating power spectra from these data, it is important to consider the effect that such uncertainties could have on the output model. In this talk, some common missing data reconstruction techniques and their shortfalls will be presented in the context of power spectrum estimation, as well as methods to quantify power spectrum uncertainties under incomplete data.
 
Bio: Liam Comerford graduated with a Bachelor in Aerospace Engineering from the University of Liverpool in 2009. He received his PhD in 2015 from the Institute for Risk and Uncertainty at the University of Liverpool. He then began his academic career as a Research Associate in Leibniz University Hannover, Germany, within the Institute Risk and Reliability. In May 2018 he returned to the UK to work in software development in the private sector. He currently maintains academic links through two European funded research projects in the areas of Stochastic Process Simulation and Compressive Sensing.

Career advice:

Some important aspects 

 

Abstract: Going for an academic career means to set out for a dynamic pathway with often unexpected obstacles and restrictions and also chances. Getting prepared early helps to reduce uncertainties and to increase chances significantly. The presentation will highlight a number of facts that are crucial for success but are often not considered or considered too late. It will be explained how to develop an impactful CV as a basis for applying for academic jobs worldwide. It will also be explained how applications are assessed, and how to prepare them to make the most important items clear to the panels and referees. This presentation is based on own experience in different academic systems and on experience from assessing applications and writing and assessing reference letters.

 

Michael Beer

MB is Professor and Head of the Institute for Risk and Reliability, Leibniz Universität Hannover. Guest Professor, Tongji University, College of Civil Engineering, Dept. of Structural Engineering & Shanghai Institute for Disaster Prevention and Relief & International Joint Research Center for Engineering Reliability and Stochastic Mechanics.

Publish successfully:

Some important aspects

Abstract: The presentation will highlight some key issues to take of when preparing journal papers. Advice will be provided on how to select the right journal for the paper, how to define the right paper type, and how to work out a research paper with emphasis on the key items that make it a research paper. In addition to these basic items, information will be provided on how to address reviewer comments, how to understand editorial decisions, and how to deal with conflicting information and problems. In overall, the information provided targets at increasing the chance for success when submitting journal papers. This presentation is based on own experience as an author, reviewer and handling and associate editor for several international journals.

 

Michael Beer

MB is Professor and Head of the Institute for Risk and Reliability, Leibniz Universität Hannover. Guest Professor, Tongji University, College of Civil Engineering, Dept. of Structural Engineering & Shanghai Institute for Disaster Prevention and Relief & International Joint Research Center for Engineering Reliability and Stochastic Mechanics.

 
 
 
 
 
 
Abstract: Blockchain has been described as the most important invention of the millennium so far. Made famous by the explosion of the cryptocurrency Bitcoin, blockchain facilitates decentralisation of all manner data, communication and other transactions. Blockchain is not just about currency. Blockchain provides an infrastructure for communication between parties without needing to trust one’s counterpart in the interaction. It is already disrupting many industries, from traditional banks developing blockchain technology to handle correspondent transactions, to healthcare providers securely and quickly sharing patient medical records. The talk will include a discussion of other potential applications of the technology in our research.
 
Bio: Dominic Calleja is a PhD student working in uncertainty quantification of plasma-facing components in future nuclear fusion reactors.