Machine Learning Projects
Here are some interesting research projects in machine learning and statistics.
Consistency Results in Estimating the Number of Clusters
Supervised by A/Prof Clara Grazian.
We prove a new result on the consistency of Pitman-Yor models in estimating the number of clusters in a mixture model. We show a positive consistency result for the case of a Uniform prior on the concentration parameter.
Summary | Full Report
Sampling Approaches to Graph Clustering and Evaluation on the StringDB PPI
Supervised by Prof Georg Gottwald.
We explore a new approach to clustering networks by sampling from a distribution of possible modularity resolutions, computing the Louvain partition for each resolution, and then taking a consensus result. We then explore a way of evaluating our clusters with stochastic block models.
Full Report
Hidden Markov Models for Analysing Stress Levels in Working Dogs
Supervised by A/Prof Clara Grazian, funded by Australian Mathematical Sciences Institute.
We explore novel methods for improving the analysis of multivariate hidden Markov models on time series with both fast and slow dynamics. We also explore data-cleaning methods for very noisy biological sensor data.
Summary | Full Report
Bayesian Model Selection for Logistic Regression via Variational Bayesian Interference
Supervised by A/Prof John Ormerod.
We develop a novel method of performing simultaneous model selection and regression using the reverse collapsed variational Bayesian method. Our algorithm outperforms $k$-NN and random forests in cross-validation MSE performance, and is available as an R
package.
Summary | Full Report | Code.