Development of a biomimetic spiking neural network for modelling plasticity in the adult nervous system.
Lead Research Organisation:
University College London
Department Name: Structural Molecular Biology
Abstract
Strategic Research Priority: Industrial biotechnology and bioenergy
Abstract
The research will focus on creating a biomimetic model of a spiking neural network. Most simulations work by assigning changing synaptic weights to a set of permanent neural connections. The goal of this project is to create a model which additionally includes neural activity-induced changes in the number of synapses and neurons, based on human and animal data. The student will gain training in modelling neural networks using evolutionary algorithms and based on biological data in order to understand the links between information processing and wiring of the nervous system.
Project
Plasticity is a feature emerging in systems with a certain levels of complexity. The research will focus on creating a biomimetic model of a spiking neural network written in C. Most of neural networks developed as artificial intelligence tools do not use biologically-realistic neuron models, Hebbian learning and architectural plasticity, and work by assigning changing synaptic weights to a set of fixed neural connections. The goal of this project is to create a bio-inspired artificial intelligence tool which can be used data processing, and utilises biologically-relevant neuron models and unsupervised spike-timing-dependent plasticity. Additionally, the project explores the potential benefits of architectural plasticity, i.e. changes in the number of synapses and neurons caused by synaptogenesis, adult neurogenesis and synaptic pruning, on information processing.
The role of synaptic plasticity in learning is highly conserved in the animal kingdom, whereas activity-induced adult neurogenesis emerges in the vertebrates. We are going to use primarily mammalian data from the hippocampus and cortex to look at hardwired and plastic aspects of information processing in neural networks. We will use molecular biology and imaging data, and techniques such as evolutionary approaches (genetic algorithms), ensemble learning and self-organising maps. In addition, the model will extensively utilise data from published brain atlases, connectomes and electrophysiological experiments. Evolutionary and spiking neural network algorithms will be used to create a model which simulates dynamic behaviour of the nervous system at levels of ranging from a single neuron to a network. Such a model will allow to perform experiments without the technical constrains of an animal model or prior to using one, and to analyse the link between the properties of networks, such as size and redundancy, and their potential for plasticity. Depending on student's progress and interests, developed plastic network can be further adapted to model age-related degeneration or neural reorganisation in response to injury
Abstract
The research will focus on creating a biomimetic model of a spiking neural network. Most simulations work by assigning changing synaptic weights to a set of permanent neural connections. The goal of this project is to create a model which additionally includes neural activity-induced changes in the number of synapses and neurons, based on human and animal data. The student will gain training in modelling neural networks using evolutionary algorithms and based on biological data in order to understand the links between information processing and wiring of the nervous system.
Project
Plasticity is a feature emerging in systems with a certain levels of complexity. The research will focus on creating a biomimetic model of a spiking neural network written in C. Most of neural networks developed as artificial intelligence tools do not use biologically-realistic neuron models, Hebbian learning and architectural plasticity, and work by assigning changing synaptic weights to a set of fixed neural connections. The goal of this project is to create a bio-inspired artificial intelligence tool which can be used data processing, and utilises biologically-relevant neuron models and unsupervised spike-timing-dependent plasticity. Additionally, the project explores the potential benefits of architectural plasticity, i.e. changes in the number of synapses and neurons caused by synaptogenesis, adult neurogenesis and synaptic pruning, on information processing.
The role of synaptic plasticity in learning is highly conserved in the animal kingdom, whereas activity-induced adult neurogenesis emerges in the vertebrates. We are going to use primarily mammalian data from the hippocampus and cortex to look at hardwired and plastic aspects of information processing in neural networks. We will use molecular biology and imaging data, and techniques such as evolutionary approaches (genetic algorithms), ensemble learning and self-organising maps. In addition, the model will extensively utilise data from published brain atlases, connectomes and electrophysiological experiments. Evolutionary and spiking neural network algorithms will be used to create a model which simulates dynamic behaviour of the nervous system at levels of ranging from a single neuron to a network. Such a model will allow to perform experiments without the technical constrains of an animal model or prior to using one, and to analyse the link between the properties of networks, such as size and redundancy, and their potential for plasticity. Depending on student's progress and interests, developed plastic network can be further adapted to model age-related degeneration or neural reorganisation in response to injury
Organisations
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| BB/M009513/1 | 01/10/2015 | 31/03/2024 | |||
| 1627508 | Studentship | BB/M009513/1 | 01/10/2015 | 02/01/2020 | Katarzyna Kozdon |
| Description | I developed a spiking neural network for unsupervised clustering of spatio-temporal data. I published three papers: 1. I demonstrated that the algorithm can be used as an ensemble of networks which process data in parallel. 2. I used evolutionary approach to automatically develop network parameters, which make the networks suitable for processing various types of data thus excluding the need for manual adjustment of the algorithm when working on a different data set. 3. The third paper discusses methods of stabilising spiking activity in brain-inspired neural networks. A paper describing the use of evolutionary algorithms to create developmental rules for spiking neural networks is currently in preparation. This research is relevant to any field wihich uses data to make strategic decisions. |
| Exploitation Route | I develop and analyse biologically-inspired algorithms for data processing. They can be used to perform in silico experiments related to neuroscience and neuroevolution, but also to process data, as an inspiration to modify existing algorithm. |
| Sectors | Aerospace, Defence and Marine,Agriculture, Food and Drink,Chemicals,Construction,Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Electronics,Energy,Financial Services, and Management Consultancy,Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology,Retail,Security and Diplomacy,Transport,Other |
| Description | I have been contacted by private companies and journalists regarding my research. I have been invited by schools to talk about my research at career events. |
| First Year Of Impact | 2018 |
| Sector | Digital/Communication/Information Technologies (including Software),Education,Financial Services, and Management Consultancy |
| Impact Types | Cultural,Societal |
| Title | Evolvable Spiking Neural Networks with Hebbian learning |
| Description | This model allows exploration of the mechanisms used for information processing and hebbian learning in neural networks. Neuronal parameters such as resistance and firing threshold are adjustable and can be modelled on any biological (human / mammalian / non-mammalian) or theoretical parameters not found in nature. Additionally, the model utilises evolutionary pressure in order to develop neural networks optimised for particular tasks. |
| Type Of Material | Model of mechanisms or symptoms - human |
| Year Produced | 2018 |
| Provided To Others? | No |
| Impact | Results obtained using this model have been submitteed to The Genetic and Evolutionary Computation Conference and are currently under review. |
| Title | Ensemble of Spiking Neural Networks for Unsupervised processing of spatio-temporal data |
| Description | The software is written in C and is an ensemble of spiking neural networks for parallel unsupervised clustering of spatio-teporal data. The updated version uses evolutionary approach to develop learning hyperparameters. |
| Type Of Technology | Software |
| Year Produced | 2017 |
| Impact | This software was used to produce data for my conference research paper (Wide Learning, Kozdon K and Bentley P 2018). The updated version was used to produce data for my conference research paper (The Evolution of Training Parameters for Spiking Neural Networks with Hebbian Learning, Kozdon K and Bentley P 2018). |
| Description | National Geographic interview |
| Form Of Engagement Activity | A press release, press conference or response to a media enquiry/interview |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Media (as a channel to the public) |
| Results and Impact | National Geographic contacted me asking to help the readers understand what AI is and isn't. Plans for further future engagements were made as well. |
| Year(s) Of Engagement Activity | 2019 |
| URL | https://www.nationalgeographic.com/science/2019/02/what-trump-american-artificial-intelligence-initi... |
| Description | PhD student Mini-conference |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Postgraduate students |
| Results and Impact | PhD student Mini-conference |
| Year(s) Of Engagement Activity | 2018 |
| Description | Professional twitter account |
| Form Of Engagement Activity | Engagement focused website, blog or social media channel |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | I keep a professional Twitter account - its focus is specifically the intersection of neuroscience and computer science (my PhD project). The intended purpose was to develop relations with researchers from the same field but also media and policy makers etc, and to promote my research and boost interest in this field of research. The account is linked to both my LinkedIn accound and website. Thanks to this online presence, I have been contacted and my research is being followed by researchers in my field with whom I probably would not be in touch otherwise, private companies, third sector organisations and media. |
| Year(s) Of Engagement Activity | 2017,2018 |
| URL | https://twitter.com/kasia_kozdon |
| Description | Professional website |
| Form Of Engagement Activity | Engagement focused website, blog or social media channel |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Other audiences |
| Results and Impact | I have created my own website - its focus is specifically the intersection of neuroscience and computer science (my PhD project). The intended purpose was to provide a platform where researchers from the same field but also media and policy makers etc could find details of my research, and to make it easier for them to contact me electronically and in person at various conferences. The website is linked to both my LinkedIn accound and professional twitter account. Thanks to this online presence, I have been contacted and my research is being followed by researchers in my field with whom I probably would not be in touch otherwise, private companies, third sector organisations and media. |
| Year(s) Of Engagement Activity | 2017,2018 |
| URL | https://bioaiblog.wordpress.com |
| Description | School visit, Inspire the Future |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Schools |
| Results and Impact | The visits usually involved presenting to groups of students (~30 students) or conversations in smaller groups (~10 students). The meetings involved all age groups from primary and secondary schools in various parts of London. The intended focus was familiarising pupils with modern, interdisciplinary research - including research done as a part of my current PhD project - and potential careers available to science graduates based on the example of my research project. The visits always sparked many questions both from the students and the teaching staff. Interdisciplinary approach and the ethical aspects of the research (cruelty-free, cheap) were always viewed in a very positive way. The listeners frequently admitted that the talks majorly improved their understanding of how modern science can look like. At every meeting, I felt that seeing a female STEM researcher was extremele important for girls - I was told by one of the teachers that the visit has inspired their female student to consider careers in STEM. |
| Year(s) Of Engagement Activity | 2015,2016,2017 |
| Description | UCL Neuroscience Symposium poster presentation |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Professional Practitioners |
| Results and Impact | UCL Neuroscience Symposium poster presentation |
| Year(s) Of Engagement Activity | 2018 |
| Description | talk at ALIFE 2018 conference, Tokyo, Japan |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | talk at ALIFE 2018 conference, Tokyo, Japan |
| Year(s) Of Engagement Activity | 2018 |
| URL | https://youtu.be/LctEcwyXzHE |
| Description | talk at National Institute of Informatics, Tokyo, Japan |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | talk at National Institute of Informatics, Tokyo, Japan |
| Year(s) Of Engagement Activity | 2018 |
| Description | talk at the London Hopper Colloquium, London (I was selected as a finalist of the flash talk competition) |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Professional Practitioners |
| Results and Impact | talk at the London Hopper Colloquium, London (I was selected as a finalist of the flash talk competition) |
| Year(s) Of Engagement Activity | 2018 |
| URL | http://www.cs.ucl.ac.uk/london_hopper_2018/ |