Project 2: Neural mechanisms of shifting attention in working memory
Cognition involves holding relevant pieces of information in mind, and manipulating them. How could such an abstract task be performed by neurons?
The prefrontal cortex (PFC) is believed to be critical in these cognitive functions, being implicated in attention, working memory, and rule-following. However, very few mechanistic theories can explain how these functions could be achieved by neurons and synapses. This project will study the idea that rapid synaptic plasticity would allow neurons to perform these cognitive functions (Manohar et al. BioRxiv 2017).
The rapid plasticity model makes strong predictions about how neural representations change when we focus on them. Some of the predictions match the patterns of brain activity already observed during attention and working memory tasks. But several predictions have not yet been tested.
We aim to test specific predictions about the format in which information is represented in the brain. Some theories propose that consistent patterns of neural activity correspond to particular items or events in the world. This is the traditional ‘place-code’ or ‘labelled-line’ view, in which representation is a stable property. However more recent theories propose that for some neurons, for example in associative or general-purpose brain areas, there is no consistent mapping between patterns of activity and the world – instead, synaptic weights of neurons can change rapidly, and so both the input and readout of neurons is variable over time. These two views produce clear testable predictions about brain activity.
In this project we will design a human behavioural experiment to test some novel predictions of this model. If unattended memory items are held in synaptic weights, there should be specific interference effects across trials, where items are remembered worse if objects consist of novel conjunctions of features from the previous trial. Note that there is flexibility in the task design and question you would like to answer, and you are expected to contribute to designing the study.
After this, there is considerable flexibility on how you would like to continue the project. You will have the option to examine neurophysiological data that has already been acquired, re-analysing it to find characteristic signatures expected from prefrontal neurons. We have access to intracranial recordings from monkeys performing attention and working memory tasks, and MEG data from humans. We plan to look for the hallmark of flexible coding in these two datasets.
Alternatively, you could take the opportunity to develop the theory further by simulating the neural networks and computational modelling. We would like to develop a version of the model that can switch between tasks – an extension that would involve considering the basal ganglia as well as the co
The DPhil would be in collaboration with Prof Mark Stokes at the Oxford Centre for Human Brain Activity (OHBA) and Prof Masud Husain in Experimental Psychology.
The project would give you exposure to eye tracking, experimental design and some programming in Python or Matlab. You would acquire a number of statistical and data analysis skills. If you choose to examine neurophysiological data, you would learn Bayesian hierarchical statistics, signal filtering, dimensionality reduction and classification techniques from machine learning. Alternatively if you choose to extend the model, you will be guided through the computational modelling. In either case, you would need a preliminary understanding of neurophysiology and neuroanatomy, and ideally have some mathematical or programming knowledge.
You will work in a team comprising a postdoctoral researcher, a research assistant, other PhD students and undergraduates. There will be plenty of guidance from the team on day-to-day issues but you are expected to be responsible for your own task design and dataset; you will meet with Prof Manohar weekly, and attend a weekly lab meeting. Ethics is in place and we have a participant database ready to go.