Skip to main content
Dr Yuhua Li

Dr Yuhua Li

Senior Lecturer

School of Computer Science and Informatics

Users
Available for postgraduate supervision

Overview

I have conducted fundamental and applied research in machine learning, pattern recognition, data science, semantic similarity analysis and condition monitoring. My top cited original research paper has been cited more than 1000 times.

My experience in machine learning and pattern recognition includes statistical and geometrical methods and neural networks for feature/pattern selection and data analysis, knowledge discovery and inference.

My contribution to machine learning includes the development of anomaly/novelty detection methods for safety-critical systems which have limited or no data/knowledge on rare events, and the study of informative observation selection techniques for sensors/measurements location optimization for problems such as effective monitoring and process control. My work in semantic similarity analysis is among the top cited publications in this area and has been adopted in real systems.

I have led and carried out research projects funded by government and industry. I have obtained extensive experience in working with different sizes of national and international companies. My research has been applied to solve problems in digital manufacturing, condition monitoring, financial engineering, etc.

Publications

2021

2020

2018

2017

2016

2015

2014

2013

2012

2011

2009

2007

2006

2005

2003

2002

2001

2000

Teaching

I received a postgraduate certificate in higher education, I am a Fellow of the HEA. I am currently teaching:

  • CMT307 Applied Machine Learning
  • CMT316 Applications of Machine Learning: Natural Language Processing/Computer Vision
  • CMT219 Algorithms, Data Structures and Programming
  • CM1210 Object Oriented Java Programming

My research interests include:

  • Machine learning, pattern recognition
  • Novelty detection, anomaly detection
  • Data science, Big Data, text mining
  • Neural networks, deep learning
  • Condition monitoring and signal processing
  • Machine learning and AI applications, e.g., cyber secuirity, finance, manufacturing

Selected publications (more publications on Google Scholar).

  • Aboozar Taherkhani, Ammar Belatreche, Yuhua Li, Liam Maguire (2018)
    "A supervised learning algorithm for learning precise timing of multiple spikes in multilayer spiking neural networks," 
    IEEE Transactions on Neural Networks and Learning Systems. DOI
  • Yi Cao, Yuhua Li, Sonya Coleman, Ammar Belatreche, Martin McGinnity (2016)
    "Detecting wash trade in financial market using digraphs and dynamic programming," 
    IEEE Transactions on Neural Networks and Learning Systems, vol. 27, no. 11, pp. 2351-2363. DOI
  • Junxiu Liu, Jim Harkin, Yuhua Li, Liam Maguire (2016)
    "Fault tolerant networks-on-chip routing with coarse and fine-grained look-ahead" 
    IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 35, no. 2, pp. 260-273. DOI
  • Aboozar Taherkhani, Ammar Belatreche, Yuhua Li, Liam Maguire (2015)
    "DL-ReSuMe: A delay learning based remote supervised method for spiking neurons," 
    IEEE Transactions on Neural Networks and Learning Systems , vol.26, no.12, pp. 3137- 3149. DOI
  • Xuemei Ding, Yuhua Li, Ammar Belatreche, Liam Maguire (2015)
    "Novelty detection using level set methods," 
    IEEE Transactions on Neural Networks and Learning Systems. vol. 26, no. 3, pp. 576-588. DOI
  • Yi Cao, Yuhua Li, Sonya Coleman, Ammar Belatreche, Martin McGinnity (2015)
    "Adaptive hidden Markov model with abnormal states for price manipulation detection," 
    IEEE Transactions on Neural Networks and Learning Systems, vol.26, no.2, pp. 318-330. DOI
  • Haider Raza, Girijesh Prasad, Yuhua Li (2015)
    "EWMA model based shift-detection methods for detecting covariate shifts in non-stationary environments," 
    Pattern Recognition, vol. 48, no. 3, pp. 659-669. DOI
  • Yuhua Li, Liam Maguire (2011)
    "Selecting critical patterns based on local geometrical and statistical information," 
    IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 6, pp. 1189-1201. DOI
  • Yuhua Li (2011)
    "Selecting training points for one-class support vector machines," 
    Pattern Recognition Letters, vol. 32, no. 11, pp. 1517-1522. DOI
  • Yuhua Li, David McLean, Zuhair Bandar, James O'Shea, Keeley Crockett. (2006)
    "Sentence similarity using semantic nets and corpus statistics," 
    IEEE Transactions on Knowledge and Data Engineering, vol. 18, no. 8, pp. 1138-1150. DOI
  • Yuhua Li, Zuhair Bandar, David McLean. (2003)
    "An approach for measuring semantic similarity using multiple information sources," 
    IEEE Transactions on Knowledge and Data Engineering, vol. 15, no.4, pp. 871-882. DOI
  • Yuhua Li, Michael Pont, Barrie Jones (2002)
    "Improving the performance of radial basis function classifiers in condition monitoring and fault diagnosis applications where "unknown" faults may occur," 
    Pattern Recognition Letters, vol.23, no.5, pp. 569-577. DOI

Supervision

I am interested in supervising PhD students in the areas of:

  • Machine learning, pattern recognition
  • Data science, Big Data, text mining
  • Neural networks, deep learning
  • Machine learning and AI applications, e.g., cyber secuirity, finance and engineering

I currently have vacancies for self-funded PhD students, you are welcome to contact me (LiY180@cardiff.ac.uk) if you have your own funding sources. Listed below are examples of PhD project proposals.

Project 1- Online classification with emerging new classes
Standard classification methods can only classify pre-defined classes, i.e., they classify a new instance into one (or multiple) of the known classes. For example, for building a classifier for viral respiratory diseases, we need to train the classification model on a dataset with pre-defined classes such as MERS and SARS. At the time of developing a model for disease classification, the classifier is trained on available data which contains only, e.g., MERS and SARS. Such a classifier can only classify MERS and SARS diseases, it will be unable to deal with the emergence of new diseases such as COVID-19 in the future. In order to deal with the emergence of new classes, a novel approach is needed to learn a classifier that is able to detect newly emerging classes and adapt the classifier accordingly. Such a classifier learning paradigm with new classes has numerous applications, e.g., self-driving cars manoeuvring in novel traffic scenarios, malware detector dealing with new type of network attacks, robotic soldiers navigating in new type of terrains, etc.

This project aims to develop a novel approach to learning a classifier that is capable to classifier emerging and novel classes. The proposed approach will address two main challenges: effective detection of emerging classes and just-in-time adaptation of classifiers for new classes. Emerging class detection will be built on the latest advances of novelty detection (novelty detection is a machine learning technique that learns a model based on only known classes to detect instances coming from a novel class), just-in-time adaptation will develop a novel incremental learning strategy to integrate new classes into current classifier. The developed algorithms will be evaluated on a use case in cybersecurity or Internet of Things (e.g., new type of network attacks).

Project 2- Learning concept evolution in data streams
In applications with concept evolution, new concepts emerge in data stream and existing/known concepts disappear over time, e.g., new types of attack in a computer network and new topics of interest in social medial data stream. This project aims to develop novel methods for tackling the challenging issue of concept evolution to enable the learned models to accommodate new concepts. It will achieve the following objectives: known concepts modelling; novel instances detection and accumulation; new concepts detection and integration; outdated concepts retiring.

Project 3- Explainable machine learning for securing Internet of Things (IoT) 
Internet of Things (IoT) consists of things, services, and networks, it connects interrelated smart devices, objects, animals or people to transfer data over a network to serve people better. The amount of data transferred with IoT systems is continuous, heterogenous and huge, which make IoT systems vulnerable than traditional network to malicious activities from attackers, so security and privacy of this highly automated network is a key challenge for the deployment of Internet of Things (IoT). It is constantly subject to adversarial attacks including denial of service, jamming, spoofing, eavesdropping, malware and privacy leakage. The limited resources (computation, battery, and memory) on IoT devices and the amount of data generated and communicated severely constrains the applicability of existing security measures to IoT systems. Even if a security system is effective at the time of deployment, it is prone to fail soon as attackers adapt more smarter strategies to foil the system and to avoid detection. Machine learning is a major tool for detecting adversarial attacks, human level explainability of detection results remain an open research in the security of IoT. 

This project aims to address these key challenges to secure future IoT systems with creative machine learning methods by: Investigating data streaming classification methods for effectively detecting known types of attacks and their variants in the future; Developing computationally cheaper machine learning algorithms as well as robustness against eavesdropping attacks; Optimising the offloading policy in dynamic radio environments to optimally distribute computational load over cloud, device and edge; Investigating adversarial machine learning techniques to tackle attackers’ changing strategies; Interpretating prediction results to support human to take trustworthy actions