A Deep Siamese Neural Network Learns the Human-Perceived Similarity Structure of Facial Expressions Without Explicit Categories


In previous work, we showed that a simple neurocomputational model {The Model, or TM) trained on the Ekman & Friesen Pictures of Facial Affect (POFA) dataset to categorize the images into the six basic expressions can account for wide array of data (albeit from a single study) on facial expression processing. The model demonstrated categorical perception of facial expressions, as well as the so-called facial expression circumplex. Here, we extend this work by 1) using a new dataset, NimsStims, that is much larger than POFA, and is not as tightly controlled for the correct Facial Action Units; 2) using a completely different neural network architecture, a Siamese Neural Network (SNN) that maps two faces through twin networks into a 2D similarity space; and 3) training the network only implicitly, based on a teaching signal that pairs of faces are in either in the same or different categories.

Back to Table of Contents