Kyan Percevault

Deep Learning in Hidden Markov Models

Rather than the traditional Baum-Welch Algorithm, I will utilise deep learning to train a Hidden Markov Model from a given observation sequence. This aims to address the non-uniqueness in local optima returned by Baum-Welch and to allow an unknown number of underlying states before training the model. I will compare this hybrid HMM-Deep Learning model to alternative methods (including existing hybrid-HMM models) and determine the ideal model depending on characteristics of the training data.

Kyan Percevault

The University of Adelaide

In 2024, Kyan completed the second year of his Bachelor of Mathematical Sciences (Advanced) at the University of Adelaide, and he is working towards a double major in statistics and applied mathematics. Therefore, Kyan’s AMSI research project is an excellent
opportunity to delve deeper into stochastic modelling, as he will investigate the differences and potential harmony between hidden Markov models and modern deep learning techniques. Outside of academics, Kyan enjoys running, swimming, and trips to the beach.

You may be interested in

Lucy Dowdell

Modelling Chemical and Biological Clogging of Permeable Reactive Barrier when Treating Acidic Groundwater

Edward Mirco

Solving Graph Colouring Problems via Convex Optimization

Geneva Birtles

Spatial Modelling of Lesion Development Informed by Multiple Sclerosis Patient MRIs

Brenton Horne

Investigating spontaneous symmetry breaking of spatial Kerr solitons in fractional spatial dimensions using Fourier spectral methods.