Welcome!
This repository serves as an archive to all notes and relevant resources to talks/lectures given in the past. This is free to anyone who wishes to look at what has been presented previously.
Enjoy! :)
Click for details!
Date: 22-May-2019 (Wed)
Pint of Science 2019 was the annual science festival that takes place every May and brings researchers to your local bar to share their research. One of such session was held at the Federick's bar in Liverpool where the theme was: "Not Yet Decided: The Value of Procrastination".
My presentation was one of the series of short 5-minute presentations as part of the "Shot of Science" segment of the event where I presented on Mozart's procrastination in composing the Don Giovanni’s overture.
Click for details!
Date: 28-Apr-2020 (Tue)
In this two-part e-lecture series, I gave a basic introduction to the concept of inverse problems, the motivation behind Bayesian Model Updating, and the tools to address Bayesian Model Updating problems.
In Part I, I gave a brief background behind the concept of model updating and the difference between deterministic and probabilistic model updating. From there, we establish that Bayesian Model Updating falls under the category of probabilistic model updating and its advantage lies in its ability to update one's knowledge from his/her apriori knowledge through making observations.
In Part II, I introduced 3 of the popular sampling techniques used to address Bayesian Model Updating problem: Markov Chain Monte Carlo (MCMC), Transitional Markov Chain Monte Carlo (TMCMC), and Sequential Monte Carlo (SMC). Details to each sampling algorithm will be introduced and described through illustrative flow-charts.
This two-part e-lecture series is also made available on YouTube:
Part I: https://youtu.be/A-cjvg741is
Part II: https://youtu.be/87b2-Fb4uas
Click for details!
Date: 12-Feb-2021 (Fri)
In this talk, I gave a basic introduction to Bayesian Model Updating, followed by an introduction to the sampling techniques employed (i.e. Markov Chain Monte Carlo, Transitional Markov Chain Monte Carlo, and Sequential Monte Carlo samplers). For each of the sampling techniques, we present simple engineering case-studies to demonstrate its implementation. Finally, we end off the discussion with a summary of the key advantages and disadvantages between the different sampling techniques. Notes and MATLAB codes to these numerical examples presented in this talk are also available here.
Click for details!
Date: 8-Dec-2021 (Wed)
In this lecture is divided into 2 segments. The first segment involves going through the theory of Model updating from which we proceed to discuss the topic on Bayesian Model Updating. This is then followed by an introduction to the sampling techniques employed (i.e. Markov Chain Monte Carlo, Transitional Markov Chain Monte Carlo, and Sequential Monte Carlo samplers). For each of the sampling techniques, we present simple engineering case-studies to demonstrate its implementation. Finally, we conclude the first segment discussion with a summary of the key advantages and disadvantages between the different sampling techniques.
In the second segment, we demonstrate the implementation of OpenCOSSAN to solve a simple Bayesian Model Updating problem involving a Linear Spring-Mass system using the Transitional Markov Chain Monte Carlo sampler.
The lecture notes, worksheet, MATLAB codes to the numerical examples presented in this talk, as well as the OpenCOSSAN MATLAB codes to the practical demonstration are also available here.
Click for details!
Date: 12-Jul-2022 (Tues)
The Semi-plenary talk was presented at the 10th International Conference on Modern Practice in Stress and Vibration Analysis (MPSVA 2022). In the talk, I gave a brief overview on the concept of Bayesian Model Updating, the Transitional Ensemble Markov Chain Monte Carlo sampler, and the Approximate Bayesian computing framework in performing Uncertainty quantification in structural dynamical problems.
As an illustration to the Uncertainty quantification framework involving the above concepts, an application problem was presented based on the recent NASA-Langley Uncertainty Quantification Challenge 2019 to which a brief comparison is made between the choice of the distribution model for the random model variables, the type of data used for the Black-box model calibration, and the choice of distance metric used for the Approximate Bayesian computation.
1st Joint TINT – Nuclear Malaysia ASEAN NPSR Technical Meeting on PSA & HRA on Nuclear Research Reactor in ASEAN Region:
Click for details!
Date: 18-Jul-2023 (Tues)
In this talk, I discussed how Uncertainty Quantification (UQ) methods and Probabilistic Safety Assessment (PSA) approaches can go hand-in-hand towards providing a robust framework towards risk assessment for Nuclear safety. We begin with a brief summary of what we currently know about PSA and the 3 key challenges it faces:
- Limited data;
- Independence assumption between events; and
- Uncertainty over the distribution models.
To address the above challenges, the presentation introduces the following UQ tools which can be used in tandem with PSA approaches:
- Bayesian model updating and Interval arithmetic to address Challenge 1;
- Frechet bounds and Fuzzy logic to address Challenge 2; and
- Bayesian model selection and Probability boxes to address Challenge 3.
An overview to the relevant tools to the above approaches will also be provided.
In summary, the twinning of PSA approaches with UQ techniques seeks to provide a more generalised and realistic framework which quantifies the uncertainty over the risk assessment given the information available as well as provide a systematic approach towards propagating the uncertainty in the calculation of the final probability of the severe accident being studied.
Click for details!
Date: 5-Apr-2024 (Fri)
In this talk, I provided an introduction to the concept of Bayesian Model Updating and Approximate Bayesian Computation (ABC) towards model calibration and validation. From which, we proceed to introduce some examples of the distance metrics used for ABC in the literature, namely:
- the Euclidean distance;
- the Bhattacharyya distance;
- the Bray-Curtis distance; and
- the 1-Wasserstein distance, along with their respective mathematical formalisms.
Following this, we proceeded to present a case study based on the 2008 SANDIA Thermal Challenge we we presented the comparison in the model validation performance of the temperature model calibrated via ABC using the different distance functions. A discussion is provided on the findings which concluded that the calibrated temperature model using the Euclidean distance-based ABC yielded the best model validation performance.