1

Question 1

What does the HGF do?

Answer 1

Allows the computation of multiple, hierarchically related, precision-weighted PEs when fitted to neurophysiological and behavioural time series.

Question 2

What does the HGF allow you to achieve?

Answer 2

To investigate with high temporal resolution the neural representation of two crucial quantities: lower-level PEs that update beliefs about stimulus probabilities, and higher-level PEs that update estimates about environmental volatility.

Question 3

What are you investigating?

Answer 3

The computational underpinning behind the ways in which healthy individuals create and sustain cognitive biases that affect well-being and interpersonal relations.

Question 4

how will you show this?

Answer 4

For the first time, I propose certain factors (such as intrinsic motivation, emotions or ritualised beliefs) distort the mechanism that assesses the reliability of our prior beliefs, expectations, and incoming sensory evidence. And that this in turn creates self-fuelling biases that affect learning and belief formation by selectively sampling information in ways that support prior belief, and work against the formation of a veridical perception.

Question 5

what’s predictive processsing?

Answer 5

By this account, we are continually in the process of predicting and explaining away sensory information in relation to our prior expectations whilst assessing the reliability of both sources.

The difference between prediction and sensory information generates a prediction-error signal that is passed up the cortical hierarchy to update beliefs about the expected causes of events - the impact of which is multiplied based on the calculated reliability of the information. Thus, high precision or reliable information has a greater impact on belief, whilst low precision has lessser impact as the information is more unreliable

Question 6

what is your methodological approach?

Answer 6

I propose a novel methodological approach recording electrophysiological (EEG) responses whilst participants conduct cognitive tasks involving attention and learning under probabilistic uncertainty, subsequently analysing the data using the Hierarchical Gaussian Filter (HGF) computational model.

Question 7

what is your research question?

Answer 7

Do factors such as intrinsic motivation, anxiety and ritualistic beliefs bias computations of uncertainty and adjust attention and learning processes?

Question 8

What’s your hypothesis?

Answer 8

The overall hypothesis is that these factors will place greater or lesser precision on prior expectation and sensory input (dependant on the factor – think anxiety reversal) that bias perception and learning. This in turn allows more specific predictions (in every experiment) centred around the Bayes-optimal learning rates developing over the time-course of experiments, about PEs being attenuated (altered in size) and developing over time and as they propagate throughout the brain, and specific predictions about the location of PE neural correlates (being in frontal, occipital, and parietal regions and characterised by gamma oscillations).

Question 9

what is the impact of your project?

Answer 9

Understanding the mechanisms underlying cognitive biases in healthy individuals that can impact well-being.

Demonstrating the limitations of Bayes-optimal theories to advance our theoretical understanding, but also to further research in clinical cases such as OCD, SCHIZO, DEPRESSION and ANXIETY disorders.

Important to understand how the predictive computations of the brain are calibrated and adjusted in the healthy brain.

Linking anxiety to evolutionary concerns for reducing uncertainty

Question 10

What are the applications?

Answer 10

Advancing theoretical understanding of both healthy and clinical cases by demonstrating the limitations of Bayes-optimal processing accounts.

Generating research able to devise therapeutic and help based programmes to identify negative cognitive biases which impact people’s lives. For those suffering from anxiety, well-being, depressive or OCD disorders, this research can illuminate what neural computations subserve negative or detrimental thought processes.

Question 11

How will youanalyse your design? What tests?

Answer 11

Single –trial classification EEG methods / HGF / Permutation tests

Question 12

Tell us a little bit about your Masters project

Answer 12

Does WM affect error awareness. ERN / Pe / Single trial / PES and Error rates

Question 13

Can you take us through a timeline of what your planning as the project sounds quite ambitious?

Answer 13

I should have all the skills necessary to begin preparing experiments from September. (?I plan to hone further MATLAB programming skills alongside creating the experiments in psychtoolbox?) – following this, recruitment and testing, analysis and write up of the first experiment will precede the viva upgrade. This will then repeat with the second experiment to the second year review / talk in Sept 2019 – then the 3rd experiment – and the last 6 months in write up.

Question 14

Where do you see yourself 10 years

Answer 14

research - teaching - emotions affect perceptual inference fmri research (amygdala, insula Cingulate gyrus, orbitofrontal cortex)

Question 15

Why PhD

Answer 15

History + The PhD represents the next step, where I can apply all the skills and knowledge I have cultivated so far, and focus them entirely on my research whilst learning and teaching others (the ideal environment for me). I am also looking forward to growing as a researcher, communicating with others in the department, and contributing to the development of an active and stimulating research environment.

Question 16

Why Goldsmiths

Answer 16

the interdisciplinary / academic rigor as well as creative freedom

Question 17

what skills you bring to Goldsmiths

Answer 17

design and analysis of cognitive psychology experiments, experience with neuroimaging (EEG), advanced quantitative techniques, and the computational approach. I even bring my experience from my 11 years as a professional musician, where I became proficient at creating, writing and collaborating with other professionals, and I honed skills in problem solving whilst creating and designing and programming electronical equipment.

Question 18

Why are you motivated to undertake this PhD project?

Answer 18

I genuinely enjoy the research process, I find it exciting and fulfilling – but I am also interested in contributing something valuable to our understanding of the computational neuroscience of cognitive biases, attention & learning processes. I am very devoted to cognitive neuroscience, it’s challenging and rewarding intellectually – I would be fortunate and exhilarated at the opportunity to continue working on research I truly love.

Question 19

What are your thoughts and approaches to teaching?

Answer 19

PhD students run tutorials of around 3 classes, and help cement understanding from lectures and facilitate learning and conversation about lecture and coursework topics.

‘fascinators’ such as an attention grabbing statement or image to generate conversation and interest in the topic.

Moreover, I encourage active learning where students create questions and talk with fellow classmates and myself in order for stimulating discourse/avenues/answers to emerge.

Question 20

Long term plan?

Answer 20

Extend PhD project by investigating the effect of emotional biases on perceptual inference. And that you could investigate as a Postdoc using fMRI (to access key areas involved in emotional processing & decision-making, e.g amygdala, insula, cingulate gyrus, orbitofrontal cortex).

Question 21

Relevant Experience/Specific Skills?

Answer 21

Bsc research / summer Jan, / third year -

maria 2 projects - British Academy / Leverhulme / applying computational models of Bayesian inference to investigate motor learning under conditions of uncertainty-EEG

Maria = MATLAB, EEGLab ICA, Bayesian Modelling, Hierarchical Gaussian Filter. I hope to combine all of this experience with my knowledge and creative approach from my experience in music…. Programming / problem solving / determination / attention to detail / high pressure environments and situations.

Question 22

What are the benefits for the Psychology department?

Answer 22

Developing our reputation as leading researchers using advanced neuroimaging and analysis techniques – and computational;

Question 23

What skills would you need to improve?

Answer 23

Computational modelling & advanced data analysis /some programming

Question 24

What teaching experience do you have?

Answer 24

drums

Question 25

How would you deal with someone who doesn’t engage with your teaching?

Answer 25

I would remaincalm, but attempt to re-engage the person and minimize the disruption to others learning. I would use fascinators such as am attention grabbing statement or image to generate conversation and interest in the topic. Moreover, I encourage active learning where students create questions and talk with fellow classmates and myself in order for stimulating answers to emerge.

I would want to make sure there was as little disruption as possible to the tutorial/lecture.

Question 26

musician / scientist

Answer 26

being a musician makes you a better scientist and viceversa

get inspiration from each discipline and take novel perspectives towards your work

many neuroscientists and psychologists are musicians -

PAM HEATON -

Peter Vuust, head of the Centre for Music and The Brain in Aarhus -

RAYMOND MACDONALD MMB external examiner (also a professional jazz musician) Raymond is Head of Music and Professor of Music Psychology and Improvisation at The University of Edinburgh

Question 27

Do you have any questions for us?

Answer 27

1. You can ask them whether you would get funds to travel to conferences
   (e. g human brain mapping / FENS (forum of neurosci in europe / ASSC association for scientific study of consciousness / the human mind conference /

> What is the thrust behind public outreach and communication between research departments in other institutions?

> Do you have many collaborations with other institutions and partners in thethird sector?

> What links between the computing department do we have to establish more coding and computational links between psychology departments?

Question 28

what is bayesian modelling?

Answer 28

gelman!

The essential characteristic of Bayesian methods is their:

> explicit use of probability for quantifying uncertainty in inferences based on data.

> Therefore, the use of Bayesian methods in the PP framework allows us to make inferences from data (and model how the brain makes those inferences) using probability models for quantities we observe and for quantities about which we wish to learn.

> Bayesian methods will allow me to evaluate how the brain of even healthy individuals makes biased estimations of uncertainty (precision) in sensory evidence or in prior beliefs, which will be crucial to addressing my main question:

Question 29

potential problems from methods or exp’s?

Answer 29

Clarification of anxiety reduction experiment using jigsaw to motivate - this need addressing

The result could reflect that anxiety leads to a higher motivation to achieve intrinsic rewards, instead of simply leading to higher motivation to reduce uncertainty during learning (when there is no additional reward). I absolutely agree in that this would be weaker evidence for the original proposal.

more notes:

The tasks will not be a problem because they have been successfully used in several studies. Participant attendance should not pose a problem because we are looking for healthy individuals (no history of neurological disease) without any specific requirement.
The only crucial point would be the computational modelling. Fitting the models to the behave and EEG data will require time and computational power, so this phase will be the most critical.

Question 30

what is your question for experiment 2?

Answer 30

Is anxiety a mechanism promoting learning to reduce uncertainty in inferences about the environment?

Question 31

Participants – Sample Size

Answer 31

N = 20, for each group, standard for EEG studies
The important point is that the model is fitted individually to the trials of each single subject, thereby allowing us to make inferences about each participants performance. What will be more crucial is the number of trials, so that the HGF can be fitted to a large number of data points in the behavioural time series (we will have between 200-300 trials).
The single-subject model estimates will enable us to account for inter-individual variability.
CHECK THE PROJECT PROPOSAL

Question 32

any outside collaboration?

Answer 32

Mathys inn Trieste

Question 33

how does the HGF and anxiety relate from maria’s work/ journal club presentation relate to your work?

Answer 33

indeed your idea to show how anxiety, as well as other factors such as intrinsic motivation, might drive our motivation to learn - after working with maria on these two projects (using the HGF) naturally fused with my ideas and increasing interest in bayesian predictive processing accounts

Question 34

cognitive biases is not maria’s area of expertise ? why is she doing this project ?

Answer 34

her expertise in neuroscience generally falls within the areas of error-monitoring and sensorimotor learning. Error-monitoring, reward-based learning and Bayesian modelling is one of my current key research areas, which has been granted the BA/Leverhulme Small Grant , etc etc.. and that I aim to pursue this line of research further.

Plus the ritual things

Question 35

what about the rituals experiment? what is this going to involve?

Answer 35

Experiment 3 will be developped more in detail based on the expertise you gain from modelling experiments 1 and 2. Knowing how to use the model in well-tested tasks will allow you to design the specific design you need for Exp 3.

but broadly speaking, it will involve the idea that rituals represent an over reliance or greater reliability on the power of prior beliefs in performing ritualised actions on having certain outcomes with the environment. Perhaps this is a way to feel in control, or yet another way to reduce anxiety - this will all develop from our work.

Question 36

you have no experience in bayesian computational models - why are you planning to run a phd using them ?

Answer 36

interested last year
developed skills this year with maria
gave sucessful presentation on bayesian stats in karina’s module
this intentionally is a project I want to grow into and learn from, i’d prefer to always be learning and ushing myself. So yes, I got the general idea and I’ll keep learning this year.