Fig 1.
Theoretical model of associative recognition.
Boxes indicate processing stages based on machine-learning analyses of EEG data; associated regions-of-interest are based on MEG evidence. Represen. = representation.
Fig 2.
Architecture of the brain model.
A) high-level overview of the seven main modules with their connections and rough localization on the brain, B) detailed functional architecture including neuron counts. DLPFC = dorsolateral prefrontal cortex; STN = subthalamic nucleus; GPe = globus pallidus externus, GPi = globus pallidus internus; Mem. = memory; Comp. = comparison.
Fig 3.
Both familiarity memory and declarative memory receive input from other parts of the model. (A) Familiarity memory learns to associate output with a certain input through supervised PES learning. (B) In addition, declarative memory creates very sparse representations in the main module through unsupervised Voja± learning, and uses recurrent strengthening to bind patterns together through unsupervised BCM learning.
Fig 4.
MEG data (A-E), fMRI data (F), and model fit. Shaded areas indicate time windows associated with the respective processes. RP foil = re-paired foil. A-C show stimulus-locked data, D-E response-locked. Because the hippocampus and perirhinal cortex are adjacent, both regions contribute to ‘familiarity’ and ‘retrieval’ data. Model fits for these regions are mixtures of the familiarity and memory populations. S2 Fig shows pure familiarity and declarative memory activity.
Fig 5.
Response time data and model fit.
Error bars indicate standard errors of the mean. RP foil = re-paired foil.
Table 1.
Parameters that were adjusted for individual models.
Table 2.
Fit measures of the individually fitted models and the average model.
Root-mean-square deviation (RMSD) is only reported for behavioral data, as we did not predict source activity quantitatively.
Fig 6.
Error rates data and model fit.
Error bars indicate standard errors of the mean. RP foil = re-paired foil.