Randall is currently studying a Bachelor of Science and Engineering at Monash University in Melbourne, with majors in applied mathematics, statistics and electrical engineering. His main area of interest within mathematics is numerical computing – the use of approximate methods to solve problems where an exact solution is difficult to obtain, and the considerations which arise when implementing these on a computer. In his spare time, Randall enjoys playing the piano and a game of Mahjong.
Numerical Optimisation Methods for Big Data Analytics
The goal of this project is to investigate the algorithmic performance of the stochastic gradient descent (SGD) method and develop improvements and variants of the method that will enhance convergence speed and robustness. Techniques such as preconditioning and variance reduction which promise faster convergence and have been recently described in the literature, will be considered.
In addition, stochastic versions of second-order methods will be considered that promise to lead to important convergence advances over first-order approaches like SGD. The methods will be applied and tested on problems in machine learning and recommendation, including logistic regression and other classification algorithms.