FindA University Ltd Featured PhD Programmes
Sheffield Hallam University Featured PhD Programmes
FindA University Ltd Featured PhD Programmes

Scalable randomized search algorithms for high-dimensional machine learning (deep learning)


School of Computing

About the Project

The focus of neural network research in recent years has been on deep learning where stochastic gradient descent and its variants are dominant methods for training large-scale neural networks. Availability of affordable computing resources and accessibility of large volumes of data are at the core of the recent surge of success in the applications of deep learning. Despite being less emphasized, the field of neuroevolution has also been affected by the same factors and population-based randomized search algorithms have shown competitive results or even outperformed their classic deep learning counterparts on a range of application areas. In a different department, several techniques and algorithms have been developed in the field of large-scale global optimization to improve the scalability of population-based search algorithms by means of exploiting problem structure and incorporation of second-order information in the optimization processes. Deep neural networks, despite being high-dimensional optimization problems by definition, have not been studied through the lens of large-scale global optimization. The aim of this project is to improve the scalability of population-based randomized algorithms by means of developing efficient ways of exploiting problem structure and the second-order information to train large-scale neural networks for a wide range of application areas including computer vision and reinforcement learning. Use of second-order information to improve the convergence properties of optimization algorithms has proven to be beneficial in both classic mathematical optimization as well as population-based algorithms. In classic optimization, this takes the form of using Hessian matrix or its approximation to exploit problem structure. In population-based algorithms, this is manifested in the form of variable interaction analysis and problem decomposition.

Funding Notes

This is a fully-funded 3-year studentship which includes tax-free doctoral stipend of £15,285 per annum and tuition fee coverage of £23,750 per year for international students and £4,600 for UK/EU students (2020/2021 fees).

Email Now

Insert previous message below for editing? 
You haven’t included a message. Providing a specific message means universities will take your enquiry more seriously and helps them provide the information you need.
Why not add a message here

The information you submit to University of Leeds will only be used by them or their data partners to deal with your enquiry, according to their privacy notice. For more information on how we use and store your data, please read our privacy statement.

* required field

Your enquiry has been emailed successfully



Search Suggestions

Search Suggestions

Based on your current searches we recommend the following search filters.



FindAPhD. Copyright 2005-2020
All rights reserved.