Exit Menu

Advanced robust deep learning

Deep neural networks have recently achieved great success and classification, where the Loss function is an indispensable part of deep learning. Various kinds of loss functions such as mean square error (MSE) and binary cross entropy (BCE), are available for different tasks, including image-based object recognition, face recognition, and speech recognition.

While many loss functions, such as MSE, are sensitive to noises or outliers, the Huber M-loss function is a popular choice for symmetric robust problems. However, the Huber-type of loss functions has limited abilities such that they can only represent certain types of error distributions resulting in a limited scope of applications. Recently, besides outlier issues asymmetry in multi-label classification, electricity pricing structure modelling, deep learning based prediction brings some challenges in symmetric loss functions for learning, classification and prediction.

This project aims to develop new algorithms for robust predictors or classification.

References:

  1. Loss and Loss Functions for Training Deep Learning Neural Networks by Jason Brownlee on January 28, 2019 in Deep Learning Performance 
  2. How to Choose Loss Functions When Training Deep Learning Neural Networks by Jason Brownlee on January 30, 2019 in Deep Learning Performance
  3. Common Loss functions in machine learning 
  4. Understanding Loss Functions in Machine Learning 

Candidates with high MSc or BSc degree grades in Statistics, Mathematics, Computing and AI are welcome to apply.

How to apply

If you are interested in applying for the above PhD topic please follow the steps below:

  1. Contact the supervisor by email or phone to discuss your interest and find out if you woold be suitable. Supervisor details can be found on this topic page. The supervisor will guide you in developing the topic-specific research proposal, which will form part of your application.
  2. Click on the 'Apply here' button on this page and you will be taken to the relevant PhD course page, where you can apply using an online application.
  3. Complete the online application indicating your selected supervisor and include the research proposal for the topic you have selected.

Good luck!

This is a self funded topic

Brunel offers a number of funding options to research students that help cover the cost of their tuition fees, contribute to living expenses or both. See more information here: https://www.brunel.ac.uk/research/Research-degrees/Research-degree-funding. The UK Government is also offering Doctoral Student Loans for eligible students, and there is some funding available through the Research Councils. Many of our international students benefit from funding provided by their governments or employers. Brunel alumni enjoy tuition fee discounts of 15%.

Meet the Supervisor(s)


Keming Yu - Keming Yu – Chair in Statistics Research Director (Impact) – in Mathematical Sciences Keming joined Brunel University London in 2005. Before that he held posts at various institutions, including University of Plymouth, Lancaster University and the Open University. Keming got his first degree in Mathematics and MSc in Statistics from universities in China and got his PhD in Statistics from The Open University, Milton Keynes.

Related Research Group(s)

Mathematical and Statistical Modelling

Mathematical and Statistical Modelling - Established in 2021 the Centre spotlights the transformative role of mathematics and statistics in innovative projects, their impact in real-world applications and in adding societal and economic value.