Classical statistical methods were developed at a time when data collection was expensive. Recent advances in science and computing technology has resulted in an explosion of available data, in fields as diverse as medicine, finance, marketing and biology. This has led to the development of new statistical methodologies, aimed at meeting the challenges associated with processing and understanding “big data”.

In this problem-solving oriented module, you will develop hands-on skills and techniques needed to turn complex data sets into useful information, implementing techniques developed in data mining and machine learning, and learning how to apply these in various data analytics packages and their open source versions.

The module spans over two terms. In the first term, it will develop basic understanding of data analytics and encourage group works, followed by more involved big data problems in the second term.

Prerequisite modules: “Scientific Computing 1" (ECM1914) and “Statistical Modelling” (ECM2907) or equivalent.

This module aims to lay the foundations for an understanding of statistical learning approaches and multivariate statistics. It aims to provide practical skills for implementing these techniques in practice, and how to effectively analyse and present “big data” effectively.

On successful completion of this module you should be able to:

- Heterogeneous, multimedia datasets like image, audio, video, text processing, financial time-series, bioinformatics, remote sensing and medical image; benchmark big datasets from engineering, physical/life sciences, business and social sciences; Data cleaning and pre-processing, data visualization; descriptive statistics, feature extraction, Introduction to statistical learning paradigms: supervised learning, unsupervised learning, semi-supervised learning, connections to statistical signal processing and information theory [3 hours];

- Big-data management and processing, parallel computing on CPUs and GPUs, algorithm scalability, signal/image filtering, wavelets, colour image processing, challenges in computer vision [3 hours];

- Multivariate analysis and dimensionality reduction: principal component analysis, independent component analysis [3 hours];

- Regression: Linear and nonlinear, univariate, multiple and multivariate regression, least square, regularization and shrinkage methods, model selection and resampling methods [3 hours];

- Introduction to Gaussian process and kernel methods; spatial and temporal random processes [3 hours];

- Classification: probabilistic and non-probabilistic classifiers, feature selection, logistic regression, discriminant analysis, k-nearest neighbour, support vector machine, decision tree, ensemble learning, generalised linear models, multilayer perceptron [3 hours];

- Clustering: k-means, mixture-model and expectation-maximisation, hierarchical and spectral clustering [3 hours];

- Recent advances in artificial intelligence and machine learning in particular deep learning (convolutional neural networks, auto-encoder, transfer learning, recurrent neural networks, deep generative models) [6 hours];

- Fuzzy inference, single and multi-objective swarm/evolutionary optimisation algorithms, reinforcement learning; Bayesian optimisation [3 hours];

- Sampling and inference, graphical models, Bayesian machine learning, combining models [3 hours].

 

If a module is normally assessed entirely by coursework, all referred/deferred assessments will normally be by assignment.

If a module is normally assessed by examination or examination plus coursework, referred and deferred assessment will normally be by examination. For referrals, only the examination will count, a mark of 40% being awarded if the examination is passed. For deferrals, candidates will be awarded the higher of the deferred examination mark or the deferred examination mark combined with the original coursework mark.

Basic reading:

ELE: http://vle.exeter.ac.uk

Reading list for this module: