Machine learning

Even amongst practitioners, there is no truly well accepted definition for machine learning. So, I’ll to provide two:

Pioneer machine learning researcher Arthur Samuel defined machine learning as: “the field of study that gives computers the ability to learn without being explicitly programmed”. This definition is beautiful in its simplicity though lacks a little formality. So, with a little more structure,..
Tom Mitchell states that “a computer program is said to learn from experience E, with respect to some task T, and some performance measure P, if its performance on T as measured by P improves with experience E”.

A classic practical application is the email spam filter. The email program watches which emails the user does or does not mark as spam and, based on that, learns how to better filter future spam automatically. In the parlance of Tim Mitchell’s definition, classifying the emails as spam or not span is the task, T, watching the user label emails as spam or not spam is the experience, E, and the fraction emails correctly classified could be the perform measure, P.

There are a great number of machine learning algorithms and, as such, they are often divided into three main types: supervised, unsupervised and reinforcement learning algorithms.

For queries about this topic, contact Ash Booth.

View the calendar of events relating to this topic.

Projects

Automated Algorithmic Trading with Intelligent Execution

Frank McGroarty, Enrico Gerding (Investigators), Ash Booth

In this project, we introduce the first fully automated trading system for real-world stock trading that uses time-adaptive execution algorithm to minimise market impact while increasing profitability com- pared to benchmark strategies.

Automated Trading with Performance Weighted Random Forests and Seasonality

Frank McGroarty, Enrico Gerding (Investigators), Ash Booth

This project proposes an expert system that uses novel machine learning techniques to predict the price return over these seasonal events, and then uses these predictions to develop a profitable trading strategy.

Automatic Image Retrieval with Soft Biometrics for Surveillance

Mark Nixon, John Carter (Investigators), Daniel Martinho-Corbishley

We're investigating ways to automatically describe and identify pedestrians from surveillance footage using human understandable, soft biometric labels. Our goal is to enable surveillance operators to search for pedestrians in a video network using soft biometric descriptions, and to automatically retrieve these descriptions from CCTV images.

BRECcIA - Building REsearch Capacity for sustainable water and food security In sub-saharan Africa

The BRECcIA project is aimed at developing research and researchers to understand water and food security challenges in sub-Saharan Africa

Centre for Doctoral Training in Next Generation Computational Modelling

Hans Fangohr, Ian Hawke, Peter Horak (Investigators), Susanne Ufermann Fangohr, Thorsten Wittemeier, Kieran Selvon, Alvaro Perez-Diaz, David Lusher, Ashley Setter, Emanuele Zappia, Hossam Ragheb, Ryan Pepper, Stephen Gow, Jan Kamenik, Paul Chambers, Robert Entwistle, Rory Brown, Joshua Greenhalgh, James Harrison, Jonathon Waters, Ioannis Begleris, Craig Rafter

The £10million Centre for Doctoral Training was launched in November 2013 and is jointly funded by EPSRC, the University of Southampton, and its partners.

The NGCM brings together world-class simulation modelling research activities from across the University of Southampton and hosts a 4-year doctoral training programme that is the first of its kind in the UK.

Deep Optimisation

Jamie Caldwell

The project will develop the implementation and application of a new optimisation technique. 'Deep optimisation' combines deep learning techniques in neural networks with distributed optimisation methods to create a dynamically re-scalable optimisation process. This project will develop this technique to better-understand its capabilities and limitations and develop GPU implementations. The protein structure prediction problem will be used as the main test application.

Development and application of powerful methods for identifying selective sweeps

Andrew Collins, Reuben Pengelly, Timothy Sluckin, Sarah Ennis (Investigators), Clare Horscroft

This project is about detecting regions of the genome which have experienced selective pressure. To achieve this, mathematical models will be developed and applied to human genomic data sets, as well as to those of other species.

Genetic studies to characterise the role of genetic factors in early-onset breast cancer

Andrew Collins (Investigator), Rosanna Upstill-Goddard

Breast cancer is a highly heterogeneous disease, with many distinct subtypes. In the majority of breast cancer cases the causative genetic component is poorly characterised. This study aims to explore both rare and common mutations in early-onset breast cancer patients and the contribution of such variants to disease using a variety of analytic approaches.

Mathematical tools for analysis of genome function, linkage disequilibrium structure and disease gene prediction

Mahesan Niranjan, Andrew Collins, Reuben Pengelly (Investigators)

This iPhD project uses a Gaussian Bayesian Networks framework through Machine learning methods to predict which genes are involved in the development of different diseases.

Mathematical tools for analysis of genome function, linkage disequilibrium structure and disease gene prediction

Andrew Collins, Mahesan Niranjan, Reuben Pengelly (Investigators), Alejandra Vergara Lope

This iPhD project uses a Gaussian Bayesian Networks approaches framework through machine learning approach to predict which genes are involved in the development of different diseases.

Mathematical tools for analysis of genome function, linkage disequilibrium structure and disease gene prediction

Mahesan Niranjan, Andrew Collins, Reuben Pengelly (Investigators)

This PhD project uses a Monte Carlo molecular simulation processes approach to predict which genes are involved in the development of different diseases.

New Forest Cicada Project

Alexander Rogers, Geoff Merrett (Investigators), Davide Zilli, Oliver Parson

Rediscover the critically endangered New Forest cicada with crowdsourced smartphone biodiversity monitoring techniques.

On the applicability of nonlinear timeseries methods for partial discharge analysis

Paul Lewin (Investigator), Lyuboslav Petrov

The governing processes of Partial Discharge (PD)
phenomena trigger aperiodic chains of events resulting in ’ap-
parently’ stochastic data, for which the widely adopted analysis
methodology is of statistical nature. However, it can be shown,
that nonlinear analysis methods can prove more adequate in
detecting certain trends and patterns in complex PD timeseries.
In this work, the application of nonlinear invariants and phase
space methods for PD analysis are discussed and potential pitfalls
are identified. Unsupervised statistical inference techniques based
on the use of surrogate data sets are proposed and employed for
the purpose of testing the applicability of nonlinear algorithms
and methods. The Generalized Hurst Exponent and Lempel Ziv
Complexity are used for finding the location of the system under
test on the spectrum between determinism and stochasticity. The
algorithms are found to have strong classification abilities at
discerning between surrogates and original point series, giving
motivation for further investigations.

Optimisation of Acoustic Systems for Perceived Sound Quality

Jordan Cheer (Investigator), Daniel Wallace

Acoustic systems have traditionally been optimised on the basis of minimising an objective acoustic measure, such as sound pressure level. The project investigates the use of subjective measures of sound quality, such as "loudness", "harshness" etc. in optimisation algorithms.