We examined the neural correlates of facial attractiveness by presenting pictures of male or female faces (neutral expression) with low/intermediate/high attractiveness to 48 male or female participants while recording their electroencephalogram (EEG). Subjective attractiveness ratings were used to determine the 10% highest, 10% middlemost, and 10% lowest rated faces for each individual participant to allow for high contrast comparisons. These were then split into preferred and dispreferred gender categories. ERP components P1, N1, P2, N2, early posterior negativity (EPN), P300 and late positive potential (LPP) (up until 3000 ms post-stimulus), and the face specific N170 were analysed. A salience effect (attractive/unattractive > intermediate) in an early LPP interval (450–850 ms) and a long-lasting valence related effect (attractive > unattractive) in a late LPP interval (1000–3000 ms) were elicited by the preferred gender faces but not by the dispreferred gender faces. Multi-variate pattern analysis (MVPA)-classifications on whole-brain single-trial EEG patterns further confirmed these salience and valence effects. It is concluded that, facial attractiveness elicits neural responses that are indicative of valenced experiences, but only if these faces are considered relevant. These experiences take time to develop and last well beyond the interval that is commonly explored.
MULTIFILE
We present a novel architecture for an AI system that allows a priori knowledge to combine with deep learning. In traditional neural networks, all available data is pooled at the input layer. Our alternative neural network is constructed so that partial representations (invariants) are learned in the intermediate layers, which can then be combined with a priori knowledge or with other predictive analyses of the same data. This leads to smaller training datasets due to more efficient learning. In addition, because this architecture allows inclusion of a priori knowledge and interpretable predictive models, the interpretability of the entire system increases while the data can still be used in a black box neural network. Our system makes use of networks of neurons rather than single neurons to enable the representation of approximations (invariants) of the output.
LINK
