Marmosets can rapidly and robustly recognize faces invariant to pose, lighting & background changes

Jeon, You-Nah
Cho, Hector
Issa, Elias B
May 04, 2022

Simian Collective Meeting (SimCo) 2022, 1, Salk Institute

For humans, faces serve as an important social cue from which we can extract information about an individual such as identity, age, and emotional states. Humans’ outstanding ability to recognize faces is reflected by functional specialization in the brain, with the presence of face-selective brain regions. In macaques, these brain regions are highly interconnected and show a hierarchical functional arrangement: areas becoming increasingly selective for view-invariant identities. The common marmoset, a New World Monkey that has recently garnered attention as a next-generation primate model for neuroscience, serves as a candidate for studying such high-level face processing in a small animal. Previous studies have found face patches in a marmoset temporal cortex, similar to macaques. Also, that marmosets make eye contact and facial expressions to communicate has been used as evidence that they can utilize visual face information. Nevertheless, whether marmosets actually have the ability to recognize faces has not been thoroughly tested. Examining the marmoset face recognition behaviors will be critical to evaluate marmosets as a model to understand the neurobiological mechanisms of human face processing. Here, we show that marmosets can robustly discriminate faces under challenging conditions, and display deficiency to conditions that are known to be difficult for humans. Three marmosets were trained to discriminate between two different artificial face identities in varying poses, size, illumination, background, and position. In order to minimize the usage of texture as cues to discriminate faces, the experimental face stimuli were hairless and greyscale. All monkeys followed a learning curriculum that started with no variations i.e. front-facing faces, and slowly more poses were shown. Once they reached above 70% on the 2AFC task with faces of different poses spanning 180 degrees horizontally (yaw) and 90 degrees vertically (pitch), they were trained and tested with all the variations, adding lighting and background variation. One monkey performed around 80% across different conditions, while the other two averaged around 65%. In preliminary experiments in the first monkey trained, face inversion and contrast reversal appeared to significantly reduce this marmoset’s ability to recognize faces even though the monkey had become very proficient at the face discrimination (>80% performance). Future work will assess whether marmosets and humans share similar signatures in face processing such as the inverted face effect or deficiencies for reversed contrast, but our emerging results indicate that they share a common behavioral repertoire for face recognition at a coarse level, endorsing the argument for marmosets as a promising model of visual and social neuroscience.