Automated non-invasive brain stimulation parameter selection using bayesian optimisation
<p>An increasingly utilised tool in cognitive neuroscience, which is capable of drawing causal relationships between brain function and behaviour, is transcranial electrical stimulation (tES). However, this non-invasive brain stimulation technique is commonly shown to have high levels of varia...
| Yazar: | |
|---|---|
| Diğer Yazarlar: | |
| Materyal Türü: | Tez |
| Dil: | English |
| Baskı/Yayın Bilgisi: |
2020
|
| Konular: |
| _version_ | 1826315867332280320 |
|---|---|
| author | Reed, T |
| author2 | Cohen Kadosh, R |
| author_facet | Cohen Kadosh, R Reed, T |
| author_sort | Reed, T |
| collection | OXFORD |
| description | <p>An increasingly utilised tool in cognitive neuroscience, which is capable of drawing causal relationships between brain function and behaviour, is transcranial electrical stimulation (tES). However, this non-invasive brain stimulation technique is commonly shown to have high levels of variability when used to modulate cognitive abilities. A potential cause of this variability is the requirement for researchers to select a number of stimulation parameters such as current intensity and frequency prior to delivering stimulation to an individual. As there is a very large number of possible parameter combinations, the selection of an optimal combination with which to modulate the behaviour of interest is highly challenging. In this thesis I employ a machine learning technique called Bayesian Optimisation (BO) to automatically search through the large stimulation parameter space to identify the optimal parameters for improving cognitive performance in complex behavioural tasks. Initially, I utilise a 'standard' BO algorithm to identify optimal parameters for improving performance in a behavioural task investigating rapid instructed task learning (Chapter 3). In a subsequent chapter I validate the results of this BO algorithm in a separate study (Chapter 4). I then attempt to extend the BO algorithm to enable it to suggest personalised stimulation parameters based on an individual's personal variables. I first apply this to a sustained attention task whilst utilising a personalised variable of individual theta:beta ratio (Chapter 5). Subsequently, I apply the personalised BO (pBO) to an arithmetic task whilst considering an individual's baseline ability as the personalised variable (Chapter 6). Finally, I validate the results of the pBO algorithm in a separate study (Chapter 7). Overall, this thesis explores the use of BO algorithms to suggest parameters for neurointerventions in complex cognitive behaviours, both at a 'one-size-fits-all' and a personalised level. Whilst results from initial optimisation studies are promising, conflicting results in validation studies highlight the difficulties of applying BO to complex human behaviour and the need for validation in further work.</p> |
| first_indexed | 2024-03-06T19:38:19Z |
| format | Thesis |
| id | oxford-uuid:1fc8c158-6e9a-48e4-8b2b-016c76cf175d |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-12-09T03:34:00Z |
| publishDate | 2020 |
| record_format | dspace |
| spelling | oxford-uuid:1fc8c158-6e9a-48e4-8b2b-016c76cf175d2024-12-01T17:34:43ZAutomated non-invasive brain stimulation parameter selection using bayesian optimisationThesishttp://purl.org/coar/resource_type/c_db06uuid:1fc8c158-6e9a-48e4-8b2b-016c76cf175dNeurosciencePsychologyEnglishHyrax Deposit2020Reed, TCohen Kadosh, RWoolrich, M<p>An increasingly utilised tool in cognitive neuroscience, which is capable of drawing causal relationships between brain function and behaviour, is transcranial electrical stimulation (tES). However, this non-invasive brain stimulation technique is commonly shown to have high levels of variability when used to modulate cognitive abilities. A potential cause of this variability is the requirement for researchers to select a number of stimulation parameters such as current intensity and frequency prior to delivering stimulation to an individual. As there is a very large number of possible parameter combinations, the selection of an optimal combination with which to modulate the behaviour of interest is highly challenging. In this thesis I employ a machine learning technique called Bayesian Optimisation (BO) to automatically search through the large stimulation parameter space to identify the optimal parameters for improving cognitive performance in complex behavioural tasks. Initially, I utilise a 'standard' BO algorithm to identify optimal parameters for improving performance in a behavioural task investigating rapid instructed task learning (Chapter 3). In a subsequent chapter I validate the results of this BO algorithm in a separate study (Chapter 4). I then attempt to extend the BO algorithm to enable it to suggest personalised stimulation parameters based on an individual's personal variables. I first apply this to a sustained attention task whilst utilising a personalised variable of individual theta:beta ratio (Chapter 5). Subsequently, I apply the personalised BO (pBO) to an arithmetic task whilst considering an individual's baseline ability as the personalised variable (Chapter 6). Finally, I validate the results of the pBO algorithm in a separate study (Chapter 7). Overall, this thesis explores the use of BO algorithms to suggest parameters for neurointerventions in complex cognitive behaviours, both at a 'one-size-fits-all' and a personalised level. Whilst results from initial optimisation studies are promising, conflicting results in validation studies highlight the difficulties of applying BO to complex human behaviour and the need for validation in further work.</p> |
| spellingShingle | Neuroscience Psychology Reed, T Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title | Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title_full | Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title_fullStr | Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title_full_unstemmed | Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title_short | Automated non-invasive brain stimulation parameter selection using bayesian optimisation |
| title_sort | automated non invasive brain stimulation parameter selection using bayesian optimisation |
| topic | Neuroscience Psychology |
| work_keys_str_mv | AT reedt automatednoninvasivebrainstimulationparameterselectionusingbayesianoptimisation |