Value-based decision making via sequential sampling with hierarchical competition and attentional modulation

(Colas, 2017)

Value-based decisions are defined as preferential decisions by this researcher. This research took an empirical stance in an attempt to select one of the many algorithmic models available to explain a “two-alternative forced-choice” paradigm in the context of subjective foods. The researcher discusses Sequential Sampling Models and in particular, the Drift Diffusion and the Race Models. He suggests that a “neurocentric modelling” can emulate human behaviour. Reaction time (RT) complemented choice selection and provided information to infer the neurophysiological and mental processes underlying behaviour.

Colas suggests that computational models of decision making range in a spectrum from simple, abstract cognitive models to modelling of individual neurons. He is attempting to suggest a connectionist model. He states that the present study took modulatory effect of attention into consideration in conjunction with various neuroalgorithmic models (ie. feedforward-inhibition model, leaky -competing-accumulator model, and the divisive normalisation Model).

A comparison of various models were made. The study itself involved use of EEG and fMRI whilst choosing one of two snack foods according to preference.A total of 7 different studies were undewrtaken with different configurations with an average of 31 subjects per study and an average of 9,157 data points captured per study (294 data points per person)

Data was ‘fitted’ to models, namely

• the race model,
• divisive-normalisation-or-feedforward-inhibition model
• the neural drift-diffusion model
• subtractive normalisation-or-feedforward-inhibition model
• subtractive competing accumulator model (also the supralinear subtractive competing-accumulator model)
• divisive competing accumulator model
• competing-accumulator model

Analysis

Reaction Times for the different choices were compared using two-tailed independent-samples t tests. The results contain a discussion of the various models used. Their conclusions pointed to the supralinear subtractive competing-accumulator model as being the most efficient

 

MY NOTES

• Used parameters that are statistically independent across trials
• Check: feedforward-inhibition model, leaky -competing-accumulator model, and the divisive normalisation Model
• Check: computational modelling, the race model, the neural drift-diffusion model and the subtractive normalisation-or-feedforward-inhibition model
• Used EEg and fMRI
• NB algorithmic computations are given for the different models
• Data was simulated based on the models to simulate 20,000 trials – find out how this is possible statistically

REFERENCE

Colas, J. T. (2017). Value-based decision making via sequential sampling with hierarchical competition and attentional modulation. PLoS ONE (Vol. 12).

.