Nature Inspired Computing and Engineering Research Group
Nature presents the best examples of how to solve complex problems efficiently and effectively.
The missions of the Nature Inspired Computing and Engineering (NICE) group are to develop computational models and algorithms inspired from natural intelligence found in biological, physical an social systems, in particular the human brain, nervous systems, gene regulatory systems, and natural evolution, and to solve real-world problems including manufacturing, healthcare, security, energy and environment.
We have three approaches for our strategy:
Real-world applications include:
- Engineering design optimization
- Medical image analysis
- Natural language processing
- Complex networks
- Autonomous systems
- Swarm and self-organizing robotic systems
Our top-down research approach aims to build up computational models for understanding biological and social intelligence found in nature.
We are particularly interested in understanding natural evolution, human perception, reasoning and learning, neural information processing in the brain, and genetic and cellular mechanisms governing the neural and morphological development from the evolutionary perspective.
Our bottom-up research approach is concerned with developing efficient mathematical and statistical, machine learning and optimization algorithms for solving complex problems found in:
- Optimization, decision-making and control
- Computer vision and image processing
- Data science and knowledge extraction
- Signal processing and pattern recognition
- Self-organization of collective systems.
Computational intelligence and machine learning
Computational intelligence embodies a wide range of techniques, from those that are inspired from biological, physical and social systems.
In this theme we investigate in computational environments natural intelligent systems at different levels, ranging from populations and societies, to nervous systems and brains, and to genes, proteins and metabolites.
- Data-driven surrogate-assisted evolutionary optimization, transfer optimization, multi-objective evolutionary optimization, and evolutionary optimization in the presence of uncertainty
- Statistical machine learning, Bayesian machine learning, logic-based machine learning, graph-based learning, deep learning, and human-like computing
- Privacy-preserving machine learning, robust machine learning, evolutionary learning, and meta-learning.
Funder: EPSRC and National Physical Laboratory
Period: July 2018 – June 2022
PI: Yaochu Jin
Collaborator: Spencer Thomas (NPL)
The project aims to develop new and innovative machine learning algorithms to analyse the data from the new 3D OrbiSIMS instrument (mass spectrometry) in a time and memory efficient manner.
Current techniques limit the volume of data that can be analysed, and currently there are no methods to integrate the different modalities produced by the instrument. The 3D OrbiSIMS is the first of its kind and is involved in a large number of projects relating to antimicrobial resistance, cancer research, and material characterisation. The project offers a unique opportunity for candidates to contribute to a wide range of disciplines and impact a broad scientific base.
Funder: Honda Research Institute Europe
Period: January 2019 - December 2021
PI: Yaochu Jin
Bayesian approaches to the optimisation of complex systems have attracted much research in recent years and have achieved encouraging success.
The project has mainly two aims:
- Develop new training algorithms and new optimisation methods that can deal with very low amount of training data for surrogate models and optimisation evaluations.
- Develop new infill criteria for Bayesian approaches to optimisation which integrate multiple models for estimating different criteria of a multi-objective problem or constraints.
Funder: Honda R&D Europe
Period: April 2018 – March 2021
PI: Yaochu Jin
The project aims to improve digital development process for vehicle dynamics in the light of efficient many-objective optimisation and smart visualisation.
Funder: Royal Society
Period: March 2018 - March 2020
PI: Yaochu Jin
In this project, we study multi-source side information fusion assisted Bayesian optimisation models and algorithms. The aim of this study is to fully exploit the side information to reduce the number of computational times of expensive fitness functions, and, meanwhile, to accurately construct response surface in the parameter space for effectively searching and recognising the global optimum.
Funder: Honda Research Institute Europe
Period: November 2016 - October 2019
PI: Yaochu Jin
This project investigates the current state of the art methods and algorithms relevant for decision making support systems. Focus points of the investigation are multi- and many-objective evolutionary optimisation methods as well as non-evolutionary MCDM methods and methods from portfolio management in relation to decision making support systems in which user of a system are supported in the task of selecting solutions from a numerically identified Pareto front.
Period: July 2015 - June 2018
PI: Yaochu Jin
Co-I: John Doherty
This research proposal aims to permit the application of evolutionary algorithms, a class of global search metaheuristics, to fluid dynamic optimisation of highly complex industrial systems by exploiting surrogate models and modern machine learning techniques.
Funder: Honda Research Institute Europe
Period: January 2013 - March 2016
PI: Yaochu Jin
This project aims to address the main challenges in evolutionary many-objective optimisation using model-based techniques and surrogate-assisted evolutionary optimisation.
To this end, the objectives of the project include:
- Develop a model-based evolutionary algorithm, thereby making it easier to represent the Pareto-optimal solutions. The start point can be the model developed in 
- To develop a preference-based approach to guide the evolutionary search.
In additional to the use of preference for modifying the dominance, an inverse model that can map the preferred search space in the objective space to the decision space will be constructed. With the help of the inverse model, the search can be biased toward the preferred region in the decision space. An on-line adaptation of the preferred solution will be considered.
Period: October 2011 - September 2012
PI: Yaochu Jin
Co-I: Andrew Crocombe
This project takes a detailed look at the design and use of materials in the aerospace industry, and will deliver a fully designed structure for use in aircraft design, and joins up a number of key themes in weight reduction, namely a reduction of fuel consumption and the knock-on environmental effects of this.
It has been estimated that reduction in 1kg mass of the panel can lead to a saving of 1.5 to two million Euros based on today’s fuel prices. The project also pin-points the safety implications which must be taken into account when superseding already advanced aerospace composite materials.
Period: January 2011 - June 2014
PI: Yaochu Jin
Machine learning or pattern matching problem consists of two parts.
Firstly, a set of features or statistics must be extracted from the object. The aim is to select features which include as much information relevant to the problem as possible, by avoiding unnecessary features.
The second part is the classifier, like a support vector machine or artificial neural network. We will invest most of the time on feature extraction, because the features must be tailored to our particular problem of recognition. If the features are well-chosen, any classifier should be able to demonstrate some positive effect.
Further work on the classifier design may improve results if the features are well-chosen, but may have no effect if they are properly not.
Computational and cognitive neurosciences
Computational models of brain function play an important role in helping us understand the brain and behaviour.
Working with psychologists and biologists, we explore brain function through computational neuroscience models.
- Spike-based neural information processing
- Activity-dependent synaptic and neuronal plasticity
- Bottom-up and top-down information processing
- Learning and memory
Funder: MRC (Medical Research Council) MR/M023281/1
Period: September 2015 - August 2018
PI: Norman Poh
Chronic Kidney Disease (CKD) is a significant cause of morbidity and mortality across the developed world. Patients with CKD have increased risk of death from cardiovascular disease and End Stage Kidney Failure, leading to dialysis and kidney transplant. Indeed, according to an NHS Kidney Care report in 2012, CKD was estimated to cost £1.45 billion in 2009-10; 1.8 million people were diagnosed with CKD in England; and, there were potentially 900,000 to 1.8 million people with undiagnosed CKD. Therefore, the importance and urgency of managing CKD cannot be overemphasised.
An overarching objective of this MRC research is to revisit the problem of modelling the progression of CKD using state-of-the-art machine learning techniques and methodologies. We introduce three innovations in this project.
First, we shall investigate statistical models that directly predict key clinical variables so that clinicians can make more informed decisions. This approach is more consistent with guidelines-based prescribing that is used by general practitioners.
Second, we will develop a way to identify patient groups by using data-driven methods. The approach used is similar to 'market segmentation' used in Business Intelligence. The hypothesis is that patients can be divided into groups not by their disease or stages (as currently practised) but by their patient records that essentially capture their health history. In essence, this method of grouping will naturally group patients with similar treatments including drugs and procedures and similar physiological and pathological characteristics.
Third, as part of the process in predicting the efficacy of kidney function, we will develop a risk model for predicting and detecting Acute Kidney Injury (AKI). This novel model will inform clinicians how likely it is that a patient will suffer from AKI. In short, we propose a unified framework to predict the efficacy of kidney function that also considers the possibility of AKI. This represents a potential advancement in modelling and understanding CKD because the risks of end-stage of CKD and AKI are so far often treated independently.
The potential advantages of the proposed method include:
- Better tailoring of the method to patient subgroups via data-driven stratification
- Ability to exploit many more variables that are specific to each patient stratum
- Ability to predict eGFR (or estimated Glomerular Filtration rate, which characterises the efficiency of kidney function) and ACR (or Albumin:creatinine ratio) that can be used in conjunction with guidelines-based prescribing
- Ability to predict the risk of acute kidney injury.
These outputs are significant in the following ways:
First, although risk models for CKD exist, there is no predictor for eGFR and ACR to date. Directly predicting these variables have significant clinical implication because the approach is consistent with guidelines-based prescribing. Unlike risk models, by directly predicting the observable outputs, these predictors convey the notions of severity and uncertainty at the same time (whilst risk models often predict only the worst outcome).
Second, there is no AKI risk model and its association with CKD remains unclear; our model considers AKI risk when predicting eGFR, thus, combining the two pieces of related information in a principled way via the Bayesian framework.
Third, we propose data-driven patient stratification as an alternative to disease and state-specific stratification. This will lead to a better understanding of CKD pathways and patient profiling. Indeed, our proof-of-concept experiment suggests that patients who have eGFR may be categorised into some 60 clusters. This patient stratification strategy inherently considers co-morbidities and disease-staging at the same time.
Funder: EU Horizon 2020 Human Brain Project (HBP)
Period: 1 April 2014 - 31 March 2016
PI: André Grüning
Collaborators: This project was a part of Subproject 4 Theoretical Neuroscience of the Human Brain Project, a multi-billion EU Horizon 2020 Flagship project, and involved collaboration with the European Institute of Theoretical Neuroscience (EITN) in Paris (the hub of the Subproject 4).
In the project we worked on constructing a framework for the formulation of goal-oriented learning rules for spiking neural networks.
Existing algorithms do not relate to each other, use different terminologies and notations without making connection to each other. We aimed at developing a mathematical and conceptual framework within which goal-oriented learning algorithms which could be formulated clearly and concisely for ease of their application, understanding and generalisation.
This was important for the understanding of how neuron-scale plasticity "conspires" to bring about goal-oriented human and animal learning behaviours at the cognitive and behavioural scale.
Period: January 2013 - June 2016
PI: Saeid Sanei
Co-I: Yaochu Jin
Collaborators: Gonzalo Alarcon, Antonio Valentin (Imperial College)
In this project new algorithms will be developed to initially use a set of previously recorded data (in their so called training phase) to best model the neural pathways from deep medial temporal source to scalp potential patterns. Solving this problem, we can then perform separation of the weak spikes from noise-like scalp signals, and localise the sources. In this direction, the major problems are nonlinearity of the medium and interference of the cortical potentials which are usually recognised as the scalp EEG of a normal brain.
Funder: EPSRC MILES (EP/I000992/1)
Period: 3 January 2012 - 21 December 2012
Collaborators: Dr. Theresa A Hague, FHMS, PGMS
The incidence of obesity in the UK has risen significantly with treatment costs rising to over £5 billion by 2025. The many social and biological causes of obesity have been explored in the Foresight obesity system map. Although useful for illustration, the qualitative nature and complexity of this map have prevented consensus on targets for public health intervention. In this project we will apply, for the first time, dynamic neural simulations to quantitatively model the social parameters influencing the increase in obesity rates to help identify variables best targeted by public health interventions.
The aim of this project is to develop a spiking neural network model to quantify the social parameters influencing the increase in overweight and obesity in the UK over the last 30 years.
Specific objectives are to:
- Identify key social parameters from the Foresight qualitative map (e.g. watching television, walking to school, media) and obtain the relevant epidemiological data from the Economic and Social Data Service and map to network.
- Develop network models using these parameters to enable measurement of the spatial dynamics of the interactions between individuals observed.
- Quantify outcomes, specifically: the emergence of polychronous groups, the average firing rates and the connectivity pattern.
- Future expansion could include other parameters within Foresight and would focus on likely parameters for public health intervention.
This project clearly meets the MILES assessment criteria in initiating a multidisciplinary collaboration between biologists, computer scientists and mathematicians. The research planned is an entirely novel exploitation of the dynamics of spiking neural networks in the context of how social parameters influence the development of obesity. The modelling of connectivity, influence and information flow within this system will be predictive and therefore offer the potential of identifying critical nodes for public health intervention. As such this project offers tremendous potential to lead to a successful bid securing external funding to continue this work.
In 2007 24 per cent of adults and 17 per cent of children in the UK were obese; the estimated costs of treating the consequences of obesity were £1 billion in 2002 and projected to be £5.3 billion by 2025 (NHS Information Centre, Lifestyle Statistics, 2009). Numerous social and biological factors have been implicated in the aetiology of obesity, and in 2007 the Foresight Programme of the UK Government Office for Science published a conceptual obesity system map that illustrates the multifactorial nature of this disease. With 108 variables this qualitative atlas highlights the tremendous complexity underlying obesity and the need to move away from single intervention approaches. However, this very complexity and the qualitative nature of the map have also prevented the development of a consensus on where public health interventions are most likely to make a difference.
At first glance, an unlikely correlate of this multifactorial, highly interconnected and complex network of influences is the human brain. The complexities of the brain have been studied at many levels, from chemical reactions in individual neurons through to high-level cognitive architectures. These studies have demonstrated that small-scale behaviour, such as when and how neurons fire, can have a large-scale influence on a population. However, it is only recently that advances in modelling the dynamics of neural interactions have led to an increase in the scale of models that can be developed (Izhikevich & Edelman, 2008). In these models, neurons are represented by biologically plausible dynamic equations, with each neuron highly connected with others. By stimulating a handful of neurons over time, behaviour emerges from the model in the form of polychronous groups: Repeating patterns of neurons that learn to activate together. These patterns result from the stimulation but demonstrate the wide influence that single inputs can establish.
These emergent patterns of behaviour may, by analogy, represent the dissemination of social influences and consequent changes in behaviour. For example, we can treat neurons as people within a population who are highly connected to others. An increase by one person in, say, watching television, modelled as a pattern of neural activity, will influence those to whom the person is connected. Over time, this influence causes the emergence of polychronous groups tuned to this pattern of activity, and hence a wider number of individuals have been influenced. The challenges are identifying social parameters that we can model and developing a plausible architecture for the influences of the parameters to be observed. With the computational efficiency of these spiking neural networks, many such parameters and architectures can be evaluated and compared with the known increase in obesity of the last 30 years in the UK to establish a benchmark. We can then test different parameter values to find target interventions.
Bringing together a team of biologists, computer scientists and mathematicians, this project aims to develop a spiking neural network model to quantify key social parameters from the Foresight qualitative map influencing the increase in overweight and obesity in the UK over the last 30 years.
E.M. Izhikevich & G.M. Edelman (2008) Large-scale model of mammalian thalamocortical systems, PNAS, 105(9):3593-3598.
Funder: The Leverhulme Trust
Period: 1 July 2009 - 31 July 2012
A constraint spatial non-negative matrix factorization (CSNMF) technique is proposed for separation of event-related, movement-related and activated regions of the brain from the rest of the fMRI sequence. NMF is a powerful decomposition approach and an emerging technique for analysis of multivariate.
The goal of NMF is to decompose a data matrix X into two other matrices, which results in the extraction of some latent features whilst reducing some redundancies in the original data. Inherently, this is an ill-posed problem. Here, a solution to this optimisation problem will be given by incorporating EEG signals into a CSNMF system.
Systems biology and morphogenetic engineering
Natural techniques demonstrate how a variety of biological systems can organize and classify high dimensional and high volume data.
By looking at a wider variety of natural systems, we can understand different patterns of intelligence, while benefiting from their application to difficult real-world problems.
- Organizing principles in nervous evolution, e.g. energy constraint and coupling between brain, body and environment
- In silico modeling, analysis and synthesis of gene regulatory networks
- Morphogenetic self-organisation of collective systems, e.g., UVAs, swarm and modul robots, mobile sensors, computing resources, and structural design.
Funder: European Commission
Period: April 2013 - September 2016
Amount: EUR 523,571
PI: Yaochu Jin
This project aims to use gene regulatory networks and morphogen gradients governing the biological development process for self-organising large-scale swarm robots that can autonomously generating patterns for following and surrounding moving targets.
Period: 7 August 2006 - 6 October 2006
PI: Paul Sowden, Antony Browne
Our understanding of both natural and artificial cognitive systems is an exciting area of research that is developing into a multi-disciplinary subject with the potential for significant impact on science, engineering and society in general. There is considerable interest in how our understanding of natural systems may help us to apply biological strategies to artificial systems. Of particular interest is our understanding of how to build adaptive information fusion systems by combining knowledge from different domains. In natural systems, the integration of sensory information is learnt at an early stage of development. Therefore, through a better understanding of the structures and processes involved in this natural adaptive integration, we may be able to construct a truly artificial multi-sensory processing system. Conversely, knowledge from theoretical work on information fusion may give a better understanding of the biology and behaviour of natural sensory systems. Here then, psychological and physiological knowledge of multi-sensory processing, and particularly the low level influence that different modalities have on one another, can be used to build upon existing theoretical work on computational mechanisms, such as self organisation and the combination of multiple neural networks, to help build systems that can fuse together different information sources. However, success in this area depends upon a cross-discipline understanding of these subjects.
The workshop on biologically inspired information fusion was held to address this issue, attempting to bring together both life and physical scientists to discuss research from the perspective of the different disciplines, focused on the common theme of information fusion. The aim was to promote collaboration between the disciplines to develop an understanding of how to build adaptive information fusion systems. This initial workshop was targeted at bringing the disciplines together and helping improve our understanding of both the natural and artificial domains. This was achieved through a two-day programme of tutorials, discussion sessions, student presentations and brainstorming.
This project has successfully met its objectives with a workshop programme designed to promote training and discussion. Participation exceeded target with 47 international participants, drawn from the disciplines of biology, psychology, computer science and robotics, including international research leaders in each field. Papers and discussion session abstracts were submitted to the workshop from each of the disciplines, resulting in four tutorials, four discussion sessions and three student presentations, each of which provoked lively debate and demonstrated the necessity of cross-discipline collaboration. Workshop activities also included three brainstorming sessions to help form cross-discipline research priorities. As a result of these activities, cross-discipline collaboration has been achieved, with at least two follow-up projects already being planned to develop adaptive models of multi-sensory integration. Evaluation of the objectives was carried out at the end of the workshop through anonymous questionnaire returned by over half the participants, with the majority of respondents reporting that the aim of the workshop had been met. All materials resulting from the workshop have been widely distributed, with further dissemination of the work via a special issue of the Information Fusion journal.
This project was done in collaboration with Dr Hujun Yin of the School of Electrical and Electronic Engineering, University of Manchester.
The workshop consisted of invited tutorials from discipline leaders to help summarise current knowledge of their field for a multi-disciplinary audience. Rising to this challenge were Professor Barry Stein (Department of Neurobiology and Anatomy, Wake Forest University School of Medicine), Dr Gemma Calvert (Multisensory Research Group in the Department of Psychology, University of Bath), Dr Charles Spence (Department of Psychology, Oxford University), Professor John Foxe (School of Psychology, City College of New York), Dr Belur Dasarathy (Editor-in-Chief of the Elsevier Information Fusion Journal and technologies consultant) and Dr Gerard McKee (School of Systems Engineering, University of Reading). In addition to tutorials, discussion sessions were invited to provoke cross-discipline debate of ideas and questions. Accepted discussions were led by Professor Alex Thomson (Department of Pharmacology, University of London), Professor Hans Colonius (Department of Psychology, University of Oldenburg), Professor Robert Damper (School of Electronics and Computer Science, University of Southampton) and Professor Leslie Smith (Department of Computing Science and Mathematics, University of Stirling). Student papers were also invited for peer review, with three papers presented at the workshop.
Find an expert
Get in contact with our staff and postgraduate research students.
Head of the Group
Professor Yaochu Jin
Dr Frank Guerin
Professor in Complex Systems
Professor Paul Krause
Lecturer in Artificial Intelligence
Dr Yunpeng Li
Senior Lecturer in Complex Systems
Dr Sotiris Moschoyiannis
Lecturer in Machine Learning and Computational Intelligence
Dr Alireza Tamaddoni Nezhad
Reader in Artificial Intelligence
Dr H Lilian Tang