Title: Postgraduate Students' Presentations
Date: Tuesday 10 May
Time: 11:00
Venue: Maths and Physics Building, Lab 602
During the month of May, a series of departmental seminars will be held in order to celebrate 25 years since the founding of the Department of Statistics and Operations Research. Following the first seminar, the second will be held on Tuesday May 10, and it will be comprised of four 15-minute talks given by Postgraduate Students. Below are the names of the speakers, together with the relative abstracts:
Kurt Dylan Buttigieg
Ensemble learning and its implementation for event and activity detection in sports
Ensemble methods combine multiple statistical learning models with the aim of obtaining a better overall predictive model which gives a favourable generalisation of the population being studied. Three of the most prevalent ensemble algorithms shall be discussed, particularly bagging, boosting and majority voting. Furthermore, the reliability of these techniques shall be investigated on the eminent problem in football, that of offside detection.
Gianluca Ursino
Phylodynamic analysis of Covid-19 in Malta using the birth-death skyline model
Phylodynamics is the combination of immunodynamics, epidemiology and evolutionary processes and the aim is to see how these processes act and possibly interact with each other to give a complete picture of viral phylogenies. In other words, genetic data can be used to infer epidemiological parameters such as the effective reproductive number as well as infer the evolutionary history and relationships between them. In this presentation, I shall demonstrate the phylodynamic model in the Bayesian framework known as the birth-death skyline plot model to a sample of the COVID-19 viral sequences collected from the Maltese population to see how the effective reproductive number changed throughout the pandemic in Malta as well as to analyse the relationships between the different variants.
Dylan Zammit
Nonparametric Bayesian estimation of the volatility function of stock data
Volatility is a measure of how much the stock price fluctuates at a particular point in time. Larger movements in the markets, such as those during Covid-19, mean that the volatility is higher than other typical periods. Measuring volatility is an important concept for financial institutions so they can manage their risk and avoid large losses due to unexpected events. In my work, I model the sock price using a Stochastic Differential Equation, and attempt to estimate the volatility function using nonparametric Bayesian techniques. This way, we can give a value to the volatility at any point in time, helping us understand which periods were the most volatile and helping us understand the characteristics of the volatility of a particular stock price.
Samuel Zammit
Estimation of cosmological parameters using Gaussian processes and viable alternatives
Cosmology is the study of the large-scale structure of the Universe. One of the main disagreements within this field relates to the value of a cosmological constant known as the Hubble parameter. Varying methodologies for estimating this parameter result in different estimates that do not agree with each other. This talk deals with the use of Gaussian process regression to obtain estimates for this parameter based on different cosmological datasets and prior values from the existing literature. Gaussian processes, which are a natural extension of the Gaussian distribution, are useful in Bayesian statistical modelling and highly flexible for such a task. Other viable alternatives, such as Markov chain Monte Carlo techniques and Student's t-processes, are also being explored.
