Biology subjects

Renoult, J. P.

Publications and source records attributed to Renoult, J. P..

10 recordsLinked to original sources

bioRxiv2024

A deep learning approach to detect and visualise sexual dimorphism in monomorphic species

Sex recognition is facilitated by dimorphism in some traits. However, humans often fail to find the traits that allow to distinguish between sexes in other species. Deep learning has the potential to surpass humans in identifying cryptic differences between sexes, but, so far, has rarely been used to assess sexual dimorphism. In this study, we evaluated (i) the ability of a fine-tuned classification neural network, EfficientNet, to find differences between sexes in a species that appears monomorphic to humans, the sociable weaver (Philetairus socius). We then assessed (ii) the benefits of Grad-CAM visualisation techniques to understand which parts of the individuals are used by the network to differentiate the sexes. We trained 10-folds cross-validation models on more than 4,500 pictures of the head from more than 1,300 individuals. Our results show that the network can predict sex of sociable weavers with an accuracy of 76%, which is considerably higher than humans performance, and that the model was similarly good at predicting females and males. When interpreting the probability of being classified to one sex, our results further reveal an effect of the interaction of sex with age on the confidence score of the models which shows that younger males are less masculine than older ones, and older females more masculine than younger ones. Finally, using Grad-CAM we found that the model mostly uses the beak region to predict the sex of individuals. Overall, this work shows that artificial intelligence has the potential to be a non-invasive sexing tool, surpassing human capabilities and aiding in pinpointing potential cryptic dimorphic body parts that have yet to be identified. In birds, half of the worlds species appear sexually monomorphic to humans, and re-evaluation of species dimorphism with this type of methods could deepen our understanding of the effect of selection on animal traits.

zoology↗

bioRxiv2023

From natural to sexual selection: Revealing a hidden preference for camouflage patterns

Background-matching visual patterns are selected to minimize detection by predators. However, once detected, what effects do these patterns have on conspecific perception? In humans, considerable evidence supports a "processing bias" by which people prefer images that match the spatial statistics of natural scenes, likely because the brain has evolved, or develops, to process such scenes efficiently. A direct but untested prediction of this bias is that people should prefer background-matching patterns compared to similar patterns that mismatch the spatial statistics of their surroundings. We conducted an online experiment to test this prediction on over 1700 participants. Our results show that a preference for the most frequent value of Fourier slope (a measure of spatial structure) observed in natural scenes can be modulated by an additional preference for background-matching. Because the initial stages of perception that underlie the efficient processing of visual scenes are widely conserved among animals, especially vertebrates, we expect this preference for background-matching patterns to be shared by other organisms. Two overlooked potential implications are that some animals may choose display sites with spatial statistics matching their own phenotype, and that the evolution of visual signals will, variously, be constrained or facilitated by such preferences for background-matching patterns. Highlights- Processing bias aims to explain the origins of preference for sexual signals - It predicts visual properties of background-matching patterns make them attractive - An online experiment with over 1700 human participants confirms that prediction - Results have important implications for the preference and design of animal signals

evolutionary biology↗

bioRxiv2023

Using generative artificial intelligence to test hypotheses about animal signal evolution: A case study in an ornamented fish.

The sensory drive hypothesis of animal signal evolution describes how animal communication signals and preferences evolve as adaptations to local environments. While classical approaches to testing this hypothesis often focus on preference for one aspect of a signal, deep learning techniques like generative models can create and manipulate stimuli without targeting a specific feature. Here, we used an artificial intelligence technique called neural style transfer to experimentally test preferences for color patterns in a fish. Findings in empirical aesthetics show that humans tend to prefer images with the visual statistics of the environment because the visual system is adapted to process them efficiently, making those images easier to process. Whether this is the case in other species remains to be tested. We therefore manipulated how similar or dissimilar male body patterns were to their habitats using the Neural Style Transfer (NST) algorithm. We predicted that males whose body patterns are more similar to their native habitats will be preferred by conspecifics. Our findings suggest that both males and females are sensitive to habitat congruence in their preferences, but to different extents, requiring additional investigation. Nonetheless, this study demonstrates the potential of artificial intelligence for testing hypotheses about animal communication signals. Highlights- Neural style transfer was used to test preferences for patterns in a colorful fish - Results trend in support of the processing bias hypothesis, a component of sensory drive - This study demonstrates the use of generative AI for animal behavioral studies

animal behavior and cognition↗

bioRxiv2022

Primate mothers promote proximity between their offspring and infants who look like them

Behavioral discrimination of kin is a key process structuring social relationships in animals. In this study, we provide a first example of discrimination towards non-kin by third-parties through a mechanism of phenotype matching. In mandrills, we recently demonstrated increased facial resemblance among paternally-related juvenile and adult females indicating adaptive opportunities for paternal kin recognition. Here, we hypothesize that mothers use offsprings facial resemblance with other infants to guide offsprings social opportunities towards similarly-looking ones. Using deep learning for face recognition in 80 wild mandrill infants, we first show that infants born to the same father or conceived during the tenure of the same alpha male resemble each other the most, independently of their age, sex or maternal origin, extending previous results to the youngest age class. Using long-term behavioral observations on association patterns and controlling for matrilineal origin, maternal relatedness and infant age and sex, we then demonstrate that, as hypothesized, mothers are spatially closer to infants that resemble their own offspring more, thereby facilitating associations among similar-looking infants. Using theoretical modeling, we describe a plausible evolutionary process whereby mothers gain fitness benefits by promoting nepotism among paternally related infants. This mechanism, that we call "second-order kin selection", may extend beyond mother-infant interactions and has the potential to explain cooperative behaviors among non-kin in social species, including humans.

evolutionary biology↗

bioRxiv2020

Deep Neural Network and field experiments reveal how transparent wing windows reduce detectability in butterflies

Lepidoptera - a group of insects in which wing transparency has arisen multiple times - exhibit much variation in the size and position of transparent wing zones. However, little is known as to how this variability affects detectability. Here, we test how the size and position of transparent elements affect predation of artificial moths by wild birds in the field. We also test whether deep neural networks (DNNs) might be a reasonable proxy for live predators, as this would enable one to rapidly test a larger range of hypotheses than is possible with live animals. We compare our field results with results from six different DNN architectures (AlexNet, VGG-16, VGG-19, ResNet-18, SqueezeNet, and GoogLeNet). Our field experiment demonstrated the effectiveness of transparent elements touching wing borders at reducing detectability, but showed no effect of transparent element size. DNN simulations only partly matched field results, as larger transparent elements were also harder for DNNs to detect. The lack of consistency between wild predators and DNNs responses raises questions about what both experiments were effectively testing, what is perceived by each predator type, and whether DNNs can be considered to be effective models for testing hypotheses about animal perception and cognition.

evolutionary biology↗

bioRxiv2019

Deep learning-based methods for individual recognition in small birds

O_LIIndividual identification is a crucial step to answer many questions in evolutionary biology and is mostly performed by marking animals with tags. Such methods are well established but often make data collection and analyses time consuming and consequently are not suited for collecting very large datasets. C_LIO_LIRecent technological and analytical advances, such as deep learning, can help overcome these limitations by automatizing data collection and analysis. Currently one of the bottlenecks preventing the application of deep learning for individual identification is the need of hundreds to thousands of labelled pictures required for training convolutional neural networks (CNNs). C_LIO_LIHere, we describe procedures that improve data collection and allow individual identification in captive and wild birds and we apply it to three small bird species, the sociable weaver Philetairus socius, the great tit Parus major and the zebra finch Taeniopygia guttata. C_LIO_LIFirst, we present an automated method that allows the collection of large samples of individually labelled images. Second, we describe how to train a CNN to identify individuals. Third, we illustrate the general applicability of CNN for individual identification in animal studies by showing that the trained CNN can predict the identity of birds from images collected in contexts that differ from the ones originally used to train the CNNs. Fourth, we present a potential solution to solve the issues of new incoming individuals. C_LIO_LIOverall our work demonstrates the feasibility of applying state-of-the-art deep learning tools for individual identification of birds, both in the lab and in the wild. These techniques are made possible by our approaches that allow efficient collection of training data. The ability to conduct individual identification of birds without requiring external markers that can be visually identified by human observers represents a major advance over current methods. C_LI

ecology↗

bioRxiv2019

Same father, same face: deep-learning reveals paternally-derived signalling of kinship in a wild primate

Animal faces convey important information such as individual health status1 or identity2,3. Human and nonhuman primates rely on highly heritable facial traits4,5 to recognize their kin6-8. However, whether these facial traits have evolved for this specific function of kin recognition remains unknown. We present the first unambiguous evidence that inter-individual facial similarity has been selected to signal kinship using a state-of-the-art artificial intelligence approach based on deep neural networks and long-term data on a natural population of nonhuman primates. The typical matrilineal society of mandrills, is characterized by an extreme males reproductive skew with one male generally siring the large majority of offspring born into the different matrilines each year9. Philopatric females are raised and live throughout their lives with familiar maternal half-sisters (MHS) but because of males reproductive monopolization, they also live with unfamiliar paternal half-sisters (PHS). Because kin selection predicts differentiated interactions with kin rather than nonkin10 and that PHS largely outnumber MHS in a mandrills social group, natural selection should favour mechanisms to recognize PHS. Here, we first show that PHS socially interact with each other as much as MHS do, both more than nonkin. Second, using artificial intelligence trained to recognize individual mandrills from a database of 16k portrait pictures, we demonstrate that facial similarity increases with genetic relatedness. However, PHS resemble more to each other than MHS do, despite both kin categories sharing similar degrees of genetic relatedness. We propose genomic imprinting as a plausible genetic mechanism to explain paternally-derived facial similarity among PHS selected to improve kin recognition. This study further highlights the potential of artificial intelligence to study evolutionary mechanisms driving variation between phenotypes.

evolutionary biology↗

bioRxiv2019

Looking for mimicry in a snake assemblage using deep learning

Batesian mimicry, with its highly colorful species and astonishing mimic-model resemblance, is a canonical example of evolution by natural selection. However, Batesian mimicry could also occur in inconspicuous species and rely on subtle resemblance. Although potentially widespread, such instances have been rarely investigated, such that the real frequency of Batesian mimicry has remained largely unknown. To fill this gap, we developed a new approach using deep learning to quantify the resemblance between putative mimics and models from photographs. We applied this method to quantify the frequency of Batesian mimicry in Western Palearctic snakes. Potential mimics were revealed by an excess of resemblance with sympatric venomous snakes compared to random expectations. We found that 8% of the non-venomous species were potential mimics, among which all were imperfect mimics. This study is the first to quantify the frequency of Batesian mimicry in a whole community of vertebrates, and shows that even concealed species can be reliably identified as potential models. Our approach should prove useful to detect mimicry in other communities, and more generally it highlights the benefits of deep learning for quantitative studies of phenotypic resemblance.

evolutionary biology↗

bioRxiv2019

Sexual signals of fish species mimic the spatial statistics of their habitat: evidence for processing bias in animal signal evolution

The diversity of animal visual displays has intrigued scientists for centuries. Sexual selection theory has explained some of this diversity, yet most of this effort has focused on simple aspects of signal design, such as color. The evolution of complex patterns that characterize many sexual displays remains largely unexplained. The field of empirical aesthetics, a subdiscipline of cognitive psychology, has shown that humans are attracted to visual images that match the spatial statistics of natural scenes. We investigated whether applying this result to animals could help explain the diversification of complex sexual signaling patterns. We used Fourier analysis to compare the spatial statistics of body patterning in ten species of darters (Etheostoma spp.), a group of freshwater fishes with striking male visual displays, with those of their respective habitats. We found a significant correlation between the spatial statistics of darter patterns and those of their habitats for males, but not for females. Our results suggest that visual characteristics of natural environments can influence the evolution of complex patterns in sexual signals.

evolutionary biology↗

bioRxiv2019

Modelling the Perception of Colour Patterns in Vertebrates with HMAX

O_LIIn order to study colour signals as animals perceive them, visual ecologists usually rely on models of colour vision that do not consider patterns-the spatial arrangement of features within a signal. C_LIO_LIHMAX describes a family of models that are used to study pattern perception in human vision research, and which have inspired many artificial intelligence algorithms. In this article, we highlight that the sensory and brain mechanisms modelled in HMAX are widespread, occurring in most if not all vertebrates, thus offering HMAX models a wide range of applications in visual ecology. C_LIO_LIWe begin with a short description of the neural mechanisms of pattern perception in vertebrates, emphasizing similarities in processes across species. Then, we provide a detailed description of HMAX, highlighting how the model is linked to biological vision. We further present sparse-HMAX, an extension of HMAX that includes a sparse coding scheme, in order to make the model even more biologically realistic and to provide a tool for estimating efficiency in information processing. In an illustrative analysis, we then show that HMAX performs better than two other reference methods (manually-positioned landmarks and the SURF algorithm) for estimating similarities between faces in a nonhuman primate species. C_LIO_LIThis manuscript is accompanied with MATLAB codes of an efficient implementation of HMAX and sparse-HMAX that can be further flexibly parameterized to model non-human colour vision, with the goal to encourage visual ecologists to adopt tools from computer vision and computational neuroscience. C_LI

evolutionary biology↗