- matlab/ --------- matlab code (input generation and postprocessing)
- webapp/ --------- javasript code to create the experiment on webapp
- interface/ ------ interface to handle communication between matlab and webapp (online database handled manually)
- img/ ------------ output images from matlab
- report/ --------- report and presentation
- Generate correct responses u with a generative model. State x_1 obtained from a bernoulli distribution of p(u|cues) and s(x_2)
- Probability of u given cues as a step function p(u|cues) = (0.1, 0.9, 0.5, 0.1, 0.9)
- Generative model to generate hidden states x3
norm(.), x2norm(.), x1~bern(s(x2), p(u|cues)) - Cues obtained from a bernoulli distribution of u
- Correct responses given to the users are noisy, meaning that 10% of the trials are toggled randomly -> increase difficulty
- Export cues and ground truth to the webapp
- Objective test: biscuits in supermarket A (yes|no) ---> are there biscuits in supermarket B?
- Subjective test: sanitizers in supermarket A (yes|no) ---> are there sanitizers in supermarket B?
- Anxiety test: 20 questions ---> 4 levels of anxiety ---> only interested in 2 (anxious or not) = ground truth of patients
- Tests 1 and 2 have the same cues and ground truths
- Tests 1 and 2 ranodmly inverted to avoid side effects (boredom, learning, ...)
- Anxiety test bassed on STAI
For both test 1 (biscuits) and test 2 (virus):
- Estimate perceptual parameters om2, om3 and mu3 using tapas_Fit function (tapas_unitsq_sgm_config)
- Estimate ideal model using bayesian model in tapas_Fit
- K-Means to cluster the estimations into 2 groups (health controls and anxious patients)
- Compute score of K-means using the anxiety test as ground truth
- Compute MSE of users responses using the ideal model as ground truth
- Correlation between estimated parameters
- Hypothesis verified: lockdown has made people more anxious of not finding sanitizers in supermarkets.