Neurons have been recorded that reflect in their firing rates the confidence in a decision. Here we show how this could arise as an emergent property in an integrate-and-fire attractor network model of decision making. The attractor network has populations of neurons that respond to each of the possible choices, each biased by the evidence for that choice, and there is competition between the attractor states until one population wins the competition and finishes with high firing that represents the decision. Noise resulting from the random spiking times of individual neurons makes the decision making probabilistic. We also show that a second attractor network can make decisions based on the confidence in the first decision. This system is supported by and accounts for neuronal responses recorded during decision making and makes predictions about the neuronal activity that will be found when a decision is made about whether to stay with a first decision or to abort the trial and start again. The research shows how monitoring can be performed in the brain and this has many implications for understanding cognitive functioning.

The study of brain communication based on fMRI data is often limited because such measurements are a mixture of session-to-session variability with subject- and condition-related information. Disentangling these contributions is crucial for real-life applications, in particular when only a few recording sessions are available. The present study aims to define a reliable standard for the extraction of multiple signatures from fMRI data, while verifying that they do not mix information about the different modalities (e.g., subjects and conditions such as tasks performed by them). In particular, condition-specific signatures should not be contaminated by subject-related information, since they aim to generalize over subjects. Practically, signatures correspond to subnetworks of directed interactions between brain regions (typically 100 covering the whole brain) supporting the subject and condition identification for single fMRI sessions. The key for robust prediction is using effective connectivity instead of functional connectivity. Our method demonstrates excellent generalization capabilities for subject identification in two datasets, using only a few sessions per subject as reference. Using another dataset with resting state and movie viewing, we show that the two signatures related to subjects and tasks correspond to distinct subnetworks, which are thus topologically orthogonal. Our results set solid foundations for applications tailored to individual subjects, such as clinical diagnostic.