ABSTRACT
Variational autoencoders (VAEs) are a popular generative model used to approximate distributions. The encoder part of the VAE is used in amortized learning of latent variables, producing a latent representation for data samples. Recently, VAEs have been used to characterize physical and biological systems. In this case study, we qualitatively examine the amortization properties of a VAE used in biological applications. We find that in this application the encoder bears a qualitative resemblance to more traditional explicit representation of latent variables.
ABSTRACT
Zahavi's "Bond Testing Hypothesis" states that irritating stimuli are used to elicit honest information from social partners regarding their attitudes towards the relationship. Two elements of the Cebus capucinus vocal repertoire, the "gargle" and "twargle," have been hypothesized to serve such a bond-testing function. The greatest threat to C. capucinus infant survival, and to adult female reproductive success, is infanticide perpetrated by alpha males. Thus, we predicted that infants (<8 months), pregnant females and females with infants would gargle/twargle at higher rates than the rest of the population, directing these vocalizations primarily to the alpha male. Over 16 years, researchers collected data via focal follows in 11 habituated groups of wild capuchins in Lomas Barbudal, Costa Rica. We found some support for our hypothesis. Infants and females with infants (<8 months) vocalized at higher rates than the rest of the population. Pregnant females did not vocalize at relatively high rates. Infants (age 8-23 months) were the only target group that vocalized more when the alpha male was not their father. Monkeys gargled and twargled most frequently towards the alpha male, who is both the perpetrator of infanticide and the most effective protector against potentially infanticidal males.