Invited Talks

Guy Theraulaz: How do fish interact with their neighbors when swimming in a school?

Schools of fish and flocks of birds display an impressive variety of collective movement patterns that emerge from local interactions among group members. These collective phenomena raise a variety of questions about the interactions rules that govern the coordination of individuals’ motions and the emergence of large-scale patterns. While numerous models have been proposed, there is still a strong need for detailed experimental studies to foster the biological understanding of such collective motion phenomena. I will first describe the methods that we developed in the recent years to characterize social interactions between individuals involved in the coordination of swimming in groups of Rummy-nose tetra (/Hemigrammus rhodostomus/) from data gathered at the individual scale. This species of tropical fish performs burst-and-coast swimming behavior that consists of sudden heading changes combined with brief accelerations followed by quasi-passive, straight decelerations. Our results show that both attraction and alignment behaviors control the reaction of fish to a neighbor. Then I will present how these results can be used to build a model of spontaneous burst-and-coast swimming and social interactions of fish, with all parameters being estimated or directly measured from experiments. This model shows that the simple addition of the pairwise interactions with two neighbors quantitatively reproduces the collective behavior observed in groups of five fish. Increasing the number of interacting neighbors does not significantly improve the simulation results. Remarkably, and even without confinement, we find that groups remain cohesive and polarized when each fish interacts with only one of its neighbors: the one that has the strongest contribution to the heading variation of the focal fish, dubbed as the “most influential neighbor". Overall, our results suggest that fish avoid information overload when they move in large groups since individuals only have to acquire a minimal amount of information about the behavior of their neighbors for coordinating their movements.

Iain Couzin: The Geometry of Decision-Making

While I have spent much of the last 15 years considering collective decision-making in animal groups - from insect swarms, to fish schools, bird flocks and wild primate groups - recently we have discovered that there exist fundamental mathematical principles that unite apparently disparate collective decision-making systems. This suggests that common underlying processes occur across vast scales of biological organization, including among neural collectives. I will present a new, simple theoretical framework that predicts the existence of common emergent geometric principles during spatial decision-making by both organismal collectives (animal groups) and neural collectives (individual organisms), and demonstrate that these result from shared intrinsic (critical) system dynamics. These dynamics are shown to both facilitate highly-effective decision-making and to spontaneously allow the group/brain to decompose complex multi-option decisions into a series of repeated binary decisions in space-time. I will then present evidence from immersive Virtual Reality (VR) experiments conducted with two evolutionarily (very) distant organisms, fruit flies and zebrafish, demonstrating that individuals of both species behave exactly as predicted by our theory during perceptual decision-making tasks. Finally, I will show that this geometric theory can also predict the movement decisions made by wild free-ranging primates.

Nir Gov: Emergent problem solving behavior during collective transport by ants

During collective transport, an ant group cooperates to carry a heavy food item efficiently to the nest. How does the group handle obstacles along the way, which do not block the passage of ants but prevent the much larger food item from passing. It is observed that upon hitting an obstacle the ants sometimes persist in pushing the food against the narrow opening, while at other times they quickly abandon the blocked passage and perform large-amplitude "searches" along the obstacle. Small objects that may be squeezed through the opening are more likely to be pushed against the narrow passage, while large objects are always carried on large excursions along the obstacle, as if looking for a way around it. We demonstrate that this rational problem-solving behavior arises spontaneously from the normal behavior of individual ants, which is unchanged by the presence of the obstacle. This means that no ant needs to realize that there is a problem (obstacle), that needs to be solved, but rather that the problem solving behavior is emergent on the collective (group) level.

Ariana Strandburg-Peshkin: Communication and collective behavior in animal societies

Group-living animals face a wide array of coordination challenges, from coming to consensus with group mates about when and where to move, to avoiding competition when searching for food, to collectively defending shared resources from external threats. For animals that live in stable social groups, social relationships are often multi-faceted and can persist over an individual’s lifetime. These complexities may introduce heterogeneity into the rules individuals employ when making decisions, with potential consequences for group-level outcomes. Furthermore, many species have evolved sophisticated communication systems that can play a key role in shaping the processes of group coordination. Employing technologies such as lightweight GPS tags, accelerometers, and audio recorders enables us to monitor of the movements, behaviors, and vocalizations of multiple individuals simultaneously within wild animal groups, offering a new window into the mechanisms underpinning collective behaviors in natural contexts. In this talk, I will present recent and emerging collaborative work exploring the mechanisms by which animals living in stable social groups coordinate collective behaviors, focusing on three systems of social mammals: olive baboons, meerkats, and spotted hyenas.

William Warren: A Visual Model of Collective Motion in Human Crowds

In experiments on humans "walking with" real and virtual crowds, we have found that pedestrians align their heading (direction of travel) and speed with a weighted average of their neighbors, in which the weight decays exponentially with distance (Rio, Dachner, & Warren, PRSB 2018). The decay rate is gradual to the nearest neighbor, then steeper within the crowd, forming a doughnut-shaped neighborhood of interaction (Warren, CDPS 2018). I will describe a new visual model that accounts for this behavior based on optical variables. The input to the model is the angular velocity and optical expansion of each neighbor, weighted by their visibility due to occlusion. The human visual system is sensitive to these variables, they are known to control following in dyads, and explain certain asymmetries in depth. The model reproduces the ‘double decay’ data in virtual crowds and more accurately simulates individual trajectories in real crowds. Importantly, we find that the gradual decay rate is a natural consequence of Euclid’s law of visual angles, while the steeper decay rate is an added effect of the partial occlusion of far neighbors. Consequently, the visual model accounts for the distance-dependence of collective motion without explicit distance terms. The doughnut-shaped neighborhood of interaction in human crowds can thus be explained by visual information.

Mehdi Moussaid: Individual and collective response to nonstationary threats

How do groups adapt to changing levels of risk in the environment? Real-life events such as the increased frequency of mass panics after a terrorist attack seem to suggest that groups overreact to a sudden increase of risk. Experimental studies of risk amplification are pointing towards the same conclusion. What are the behavioral mechanisms of this phenomenon ? In this talk, I will present the first results of an ongoing project trying to decipher the causes of this apparent maladaptive behavior.

Cristian Huepe: Relating Emerging Collective Dynamics to Interaction Network Structures

The emergence of collective dynamics in a broad range of systems is determined not only by the behavior of its components but also by the structure of its interaction network. Despite this, there is little understanding on how this interaction topology can affect collective behavior and decision making. In this talk, I will present two cases where the interaction structure plays a critical role. In the first one, I will show how the self-organization of two models of distributed consensus and collective motion can be affected by interaction topologies ranging from nearest neighbors to long-range random Erdős–Rényi or Scale Free topologies, discussing applications to animal group behavior and swarm robotics. In the second one, I will show how the coupling between the state of individual agents and their interactions can strongly affect collective opinion formation in social groups, discussing an example based on the analysis of Twitter data during a recent movement of social unrest.

Gonzalo Polavieja: Modelling animal interactions as the sensorimotor transformation compatible with a desired global behavior

I will describe a modelling approach that uses reinforcement learning to obtain which sensorimotor transformation gives a desired global behaviour. As examples of desired global behaviors we consider rotating mills, tornadoes and balls. We obtain that, despite only constraining the model by the global behavior, all solutions in the x-y plane are similar among themselves and to those of models obtained using experimental trajectories. The different global behaviors emerge mainly by differences in how information is processed in the z-direction. This is unpublished work obtained in the lab by Tiago Costa, Andres Laan and Francisco Heras.

Benjamin Judkewitz: Tiny brain but rich behavior in the singing fish D. translucida

Understanding how distributed neuronal circuits integrate sensory information and generate behavior is a central goal of neuroscience. However, it has been difficult to study neuronal networks at single-cell resolution across the entire adult brain in vertebrates because of their size and opacity. We address this challenge by introducing the fish Danionella translucida to neuroscience as a model organism. This teleost remains small and transparent even in adulthood, when neural circuits and behavior have matured. Despite having the smallest known adult vertebrate brain, D. translucida displays a rich set of behaviors, including courtship, shoaling, schooling and - remarkably - acoustic communication. In order to carry out optical measurements and perturbations of activity with genetically encoded tools, we de-novo sequenced the whole genome, established CRISPR–Cas9 genome editing and Tol2 transgenesis techniques. We created several transgenic lines expressing calcium indicators and developed a large field of view volumetric imaging method that matches the size of the Danionella brain. We demonstrate the power of these techniques with whole-brain imaging of auditory-evoked activity in response conspecific vocalisations and heterospecific sounds. Taken together, these developments make D. translucida a promising model organism for the study of adult vertebrate brain function at single-cell resolution.

Contributed Talks

David Bierbach: Robofish – A biomimetic fish-like robot for the study of collective behavior

Live animals, especially when in a group of conspecifics, can hardly be forced to change their behaviors as intended by the experimenter. Hence, experimental manipulations are difficult and often lacking to underpin proposed theoretical assumptions on collective movement, decision-making or leadership. One elegant solution is to replace live individuals in a group by biomimetic robots. In a multidisciplinary group, we thus developed an interactive fish robot – the so-called RoboFish. Through the interactive behaviors based on real time tracking of the fish shoal, our system can be adjusted to imitate different individual characteristics of live fish and is accepted as a conspecific by live guppies (Poecilia reticulata). This allows us to use RoboFish to investigate many different questions regarding collective behavior.

Valerio Sbragaglia: Harvesting-induced evolution of collective behavior in a fish

Many fisheries around the globe preferentially capture large individuals with implications for the evolution of exploited populations. Fisheries-induced evolution may alter collective behavioral phenotypes through individual-level adaptations that affect boldness, swimming speed and tendency to follow social vs. environmental cues. Studying the behavioural mechanisms that give rise to possible changes in shoaling and other collective outputs is challenging in the wild, but first insights into whether intensive and size-selective harvesting could alter collective phenotypes and shoaling can be gathered through experiment of size-selective harvesting conducted in the laboratory. We present a multi-generation harvest selection experiment with zebrafish (Danio rerio) as a model species and demonstrate that large size-selective harvesting typical of global fisheries decreases risk-taking behavior of individuals, and surprisingly also decreases shoal cohesion. This counter-intuitive effect at the collective level is mechanistically caused by risk-averse individuals favored under large size-selective harvest paying more attention to environmental instead of social cues. Agent-based model simulations further reveal that fisheries-induced evolution of shoaling behavior is adaptive under fishing scenarios by decreasing exploitation rate. By contrast, the same collective behavior favored by size-selective harvesting is maladaptive in the presence of natural predation and increases natural mortality. The evolutionary adaptations we document may slowly, but steadily erode the natural fitness benefits offered by shoaling in many species targeted by global fisheries. Erosion of shoal cohesion can also negatively affect catchability with consequences for human well-being.

Winnie Poel: Individual and collective encoding of risk in animal groups

The need to make fast decisions under risky and uncertain conditions is a widespread problem in the natural world. While there has been extensive work on how individual organisms dynamically modify their behavior to respond appropriately to changing environmental conditions (and how this is encoded in the brain), we know remarkably little about the corresponding aspects of collective information processing in animal groups. For example, many groups appear to show increased “sensitivity” in the presence of perceived threat, as evidenced by the increased frequency and magnitude of repeated cascading waves of behavioral change often observed in fish schools and bird flocks under such circumstances. How such context-dependent changes in collective sensitivity are mediated, however, is unknown. Here we address this question using schooling fish as a model system, focusing on 2 nonexclusive hypotheses: 1) that changes in collective responsiveness result from changes in how individuals respond to social cues (i.e., changes to the properties of the “nodes” in the social network), and 2) that they result from changes made to the structural connectivity of the network itself (i.e., the computation is encoded in the “edges” of the network). We find that despite the fact that perceived risk increases the probability for individuals to initiate an alarm, the context-dependent change in collective sensitivity predominantly results not from changes in how individuals respond to social cues, but instead from how individuals modify the spatial structure, and correspondingly the topology of the network of interactions, within the group. Risk is thus encoded as a collective property, emphasizing that in group-living species individual fitness can depend strongly on coupling between scales of behavioral organization.

Benjamin Wild: Social networks through time - Individuality in a colony of honey bees

Non-negative matrix factorization is a widely used method to analyze and detect community-based structure in social networks. Unfortunately, there's no straightforward extension of this approach to temporal networks. In many model organisms, and particularly in social insects, the patterns of actions and interactions among individuals are not static but constantly evolving over time. This can be due to the emergence or demise of certain individuals, changing task allocation because of temporal polyethism and changes in the environment, or many other reasons. Understanding such temporal patterns in complex social networks remains a challenging problem. In this work, I will present a novel temporal non-negative matrix factorization algorithm that concurrently learns semantic embeddings of all individuals of a temporal social network and a functional mapping from these embeddings to factors representing the state of the individuals at a specific point in time or age. This method can be used to detect clusters of social development of individuals over time, to compare the structure of the networks at different times, or to compare the role of individuals in the social structure in a meaningful way even when they were never alive at the same time.

Fernando Wario Vazquez: Motion dynamics of pre-dance trajectories in honey-bees

Information transfer among foragers in honeybees is key for efficient allocation of work and for adaptive responses within the colony. For information to spread quickly, foragers performing a waggle dance (dancers) must reach as many other non-dancing foragers (followers) as possible. Forager bees may have radically different drives that may influence their motion pattern. For instance, dancer bees may want to widely cover the dance floor to recruit other bees, the more broadly the higher the food source profitability. Followers may instead move more erratically in the hope to meet a dance. A good mixing of individuals may be valuable to have flexibility at the level of the colony behaviour and optimally respond to changing environmental conditions.

We aim to determine differences between the motion pattern that precedes dancing and following behaviours, exploiting a data-driven computational model. To this end, real observation data (BeesBook [Wario 2015]) are used to define environmental properties (comb surface characteristics) such as dancefloor location, shape and size, and population size and density distribution, all characteristics that highly correlate with the bees walking pattern. A simulation environment is deployed to test different mobility patterns (correlated random walks [Codling 2008; Bartumeus 2005], Lévy walks [Viswanathan 2008], random waypoint models [Bettstetter 2004]) for forager bees within the hive and evaluate basic metrics such as the mean square displacement (MSD) and the interaction rate. The results show under what conditions information transfer is most efficient and open the way towards a detailed comparison between our simulation results and real-world data.

References: Bartumeus, F., Luz, M., Viswanathan, G., Catalan, J. (2005). Animal search search strategies: a quantitative random‐walk analysis Ecology 86(11), 3078 - 3087. https://dx.doi.org/10.1890/04-1806

Bettstetter, C., Hartenstein, H. & Pérez-Costa, X. Stochastic Properties of the Random Waypoint Mobility Model. Wireless Networks 10, 555–567 (2004) https://doi.org/10.1023/B:WINE.0000036458.88990.e5

Codling, E. A., Plank, M. J., & Benhamou, S. (2008). Random walk models in biology. Journal of the Royal Society Interface, 5(25), 813–834. https://doi.org/10.1098/rsif.2008.0014

Viswanathan, G., Raposo, E., Luz, M. (2008). Lévy flights and superdiffusion in the context of biological encounters and random searches Physics of Life Reviews 5(3), 133 - 150. https://dx.doi.org/10.1016/j.plrev.2008.03.002

Wario, F., Wild, B., Couvillon, M. J., Rojas, R., & Landgraf, T. (2015). Automatic methods for long-term tracking and the detection and decoding of communication dances in honeybees. Frontiers in Ecology and Evolution, 3(September), 1–14. https://doi.org/10.3389/fevo.2015.00103

Taeyeong Choi: Why Individual Tracking? Temporal Phase Prediction on Insect Society in Crisis using Global Visuals

Social temperature can easily fluctuate in a group as it experiences a negative event especially at an unexpected timing. Individuals may present unusual behaviors to avoid the accompanied impacts until the society finally recovers after the crisis. Since the behavioral responses could be the most informative indicator of abnormal group state, state-of-the-art computer vision algorithms can be a useful tool to monitor and quantify the gradual temporal changes of group-level state. In this work, we take advantage of video data about Jerdon’s jumping ant, Harpegnathos saltator, as a prototype society to explore an approach that can learn a regularity between individual-level motions and colony-level temporal phases during social chaos. Our Convolutional Neural Network (CNN) based predictor does not require any additional module to either track individual ants or estimate poses but instead utilizes the optical flows as behavioral features. As a result, the trained network can classify a short video input as "Early chaos" or "Late chaos" with high accuracy and also can provide heatmaps to explain specific individuals or transient actions it refers to for prediction. In addition, latent variables extracted from an intermediate layer enable to analyze temporal state changes in a more microscopic view. Through the talk, we will show some experimental results from two separate colonies of unique members and discuss possible future applications to a different biological dataset.

Leo Epstein: Learning an interpretable representation of Physarum polycephalum's behavioural ecology

Learnability of the spatiotemporal dynamics of morphological networks in biology has been restricted by the complexity of model representations. We show how the use of a morphospace representation enables the learning of a compact and explainable model of an organism’s morphological dynamics during multispecies ecological interactions and under active exogenous control. Furthermore, use of a morphospace allows us to correlate the effect that an exogenous stimulus has on the morphogenic program of our organism of interest. Using a morphospace representation of the development of the unicellular amoeba Physarum polycephalum we may learn how its interactions with yeast are represented in its physiology. A morphospace or any other low dimensional representation of Physarums morphological network allows us to contextualize and relate Physarums well-studied behavior to each other and better understand the basal biophysical processes from which these behaviors are built. These behaviors include approximating shortest spanning trees between static points of interest, the ability to develop positive associations between negative stimuli such as salt if the stimulus is associated with food, and photoavoidance (especially blue light). Using UNET semantically segmented time-lapses of Physarum yeast ecology we can measure the physiological components of Physarums morphology shown to drive its morphogenesis. These components are related to the fluid flow that shapes Physarum morphology. Alim et al 2018. The flow is driven by coupled oscillations of contracting actin across the membrane of Physarum. With UNET We segment Physarum vasculature through which fluid flows and the actin rich membrane through which contractions propagate. There has been a great deal of interest in the field of neuroscience in using machine learning to interpret the functional morphology of neurons and how they associate with each other. We believe techniques from this area can be directly applied to understand Physarum morphogenesis during foraging as the transition between different functional morphotypes. Inspired by P.J Schubert et al. 2019 "Learning cellular morphology with neural networks" we hope to train a network with triplet loss to output the last several light states the Physarum has experienced based on consecutive sequences of semantic segmentations of Physarum physiology and its environment. With the embedding generated by the triplet loss, it may be possible to quantify Physarum’s morphological response to blue light by looking at how points with similar morphological measurements cluster together in the triplet loss embedding space and how these clusters map back to the morphospace. In summary, we believe that morphospaces and neural network embeddings can be used together to learn simple yet informative representations of Physarum's behavioral ecology.

Thejasvi Beleyur: How does jamming affect collective behaviour in echolocating bats?

Active sensing animals perceive their surroundings by emitting probes of energy and analyzing how the environment modulates these probes. However, the probes of conspecifics can jam active sensing by masking the faint echoes of interest, which should cause problems for groups of active sensing animals. This problem is called the cocktail party nightmare for echolocating bats. Despite this problem, many bats echolocate in groups and roost socially. Here, we present a biologically parametrized framework to quantify echo detection in groups. Incorporating known properties of echolocation, psychoacoustics, spatial acoustics and group flight, we quantify how well bats flying in groups can detect each other despite jamming. A focal bat in the center of a group can detect neighbors for group sizes of up to 100 bats. With increasing group size, fewer and only the closest and frontal neighbors are detected. Neighbor detection is improved for longer call intervals, shorter call durations, denser groups and more variable flight and sonar beam directions. Our results provide the first quantification of the sensory input of echolocating bats in collective group flight, such as mating swarms or emergences. Our results further generate predictions on the sensory strategies bats may use to reduce jamming in the cocktail party nightmare. Lastly, we suggest that the spatially limited sensory field of echolocators leads to limited interactions within a group, so that collective behavior is achieved by following only nearest neighbors.

Renaud Bastien: A Model of Movement Based on Vison

Classical phenomenological models of collective behavior often take a "birds-eye perspective," assuming that individuals have access to social information that is not directly available (e.g., the behavior of individuals outside of their field of view). Despite the explanatory success of those models, it is now thought that a better understanding needs to incorporate of the perception of the individual, i.e., how internal and external information are acquired and processed. In particular, vision has appeared to be a central feature to gather external information and influence the collective organization of the group. In this talk, I will show how models can be constructed from the perception of an individual and how virtual reality can be used to infer the relationship between the structure of the group and the individual perception of the animals. Starting with research on plants tropism, I will first discuss the design of models relating perception and movements. From this framework, a simple model can be constructed that relate directly to the visual field of each individual and the dynamics observed at the scale of the group. It is then possible to discuss how visual features can be combined to create basic interaction between individuals, as well as the existence of internal representations of space and others.

Dylan H. Morris: A Social Dilemma of Sociality and Collective Processing

Why are some animals socially gregarious while others keep to themselves? Evolutionary models of gregarious behavior typically treat benefits and costs of social interaction qualitatively. Some of these benefits and costs are contagious: for example, socializing may allow individuals to share useful information with their neighbors, but also expose them to dangerous infectious diseases. Here, we present a model the evolution of sociality in the presence of beneficial and costly social contagion processes. We characterize a socially optimal level of social interaction, and show that evolutionary dynamics produce a social dilemma: individuals maximizing their fitness drive the population to a level of sociality at which all individuals are worse off. In some cases, social behavior can disappear entirely -- even when any level of socializing would be advantageous for the species as a whole.

Madalina Vlasceanu: Collective Belief Update and Synchronization in Social Networks

Systems of beliefs organized around religion, politics, and health constitute the building blocks of human communities. Here, we study the dynamics of belief endorsement in lab-created 10-member social networks following conversational interactions. Participants first evaluated the accuracy of a set of statements (pretest), after which they were exposed to evidence either in support or against the statements. Then, participants engaged in a series of conversational interactions during which they discussed the evidence provided, following predetermined network structures (Clustered or Non-Clustered). Lastly, participants were asked to evaluate the accuracy of the initial statements again (posttest). We find that the more the evidence was endorsed in conversation, the more it was used in updating beliefs. Moreover, the belief similarity of participants within each network (i.e., pairwise correlations of belief scores) increased from pretest to posttest, suggesting conversational deliberation results in community-wide belief synchronization. Critically, we found that the network structure is essential in explaining the shift in belief similarity from pretest to posttest. These findings advance understanding of the dynamics of collective belief change and synchronization in social networks.

Tanja Kaiser: Minimal Surprise - Evolving Swarm Behaviors with an Intrisic Driver

In swarm robotics, the definition of task-specific fitness functions for the evolution of collective behaviors is challenging. In our minimal surprise approach, we make use of an intrinsic, task-independent driver which is loosely inspired by Friston's free-energy principle for natural brains. This intrinsic motivation rewards high prediction accuracy of sensor values while the robot controller emerges as a by-product of the evolutionary dynamics. We show that using minimal surprise leads to the evolution of typical swarm behaviors like flocking, dispersion or self-assembly.

Mari Kawakatsu: Evolutionary dynamics of polarization and cooperation in group-structured populations?

The emergence of stable social structures has long fascinated researchers in fields ranging from evolutionary biology to political science. Political theorists as early as James Madison posited that a diversity of interests should inhibit the formation of factions: when individuals have many political interests, they must cooperate and form alliances across disagreements to achieve common goals. But this idea runs counter to a result established by evolutionary models of cooperation with dynamic population structures: cooperation often promotes increased connectivity, which in turn makes the system vulnerable to invasion, leading to a fragmented population. Under what conditions, then, can cooperation and social cohesion evolve together? We develop a framework to study this question in the context of multiple, overlapping groups that represent various issues of interest. Individuals belong to an interest group if they have opinions about the corresponding issue, and they adopt strategies that are conditional on the opinions of others. Individuals interact through an evolutionary game if they share a common interest, with both strategies and group memberships subject to evolutionary updating. We first ask what factors promote cooperation and cohesion and then investigate how these factors shape local population structure and the diversity of strategies in the population. Our work contributes to a growing body of literature on how heterogeneity in individual attributes shapes social structure.

Pascal Klamser: Collective Predator Response: Group vs. Individual Optima and Criticality

Collective behavior inspired and statistical physics models share commonalities as a transition region from an ordered to a disorder state. These transition regions are promising candidates for the final state of a self-organized system because statistical physics models characterize them as most responsive to external perturbations (diverging susceptibility). Here we test this hypothesis in an agent-based model which self-organizes by evolving under predatory pressure. Interestingly we find strong effects on the fitness function at the transition region; however, the final-state is linked to a trade-off between individual and social information and is not restricted to the transition region.

Wataru Toyokawa: Collective rescue: social reinforcement learning promotes adaptive behavioural-shift in risky decision-making

Objective Gambling tasks, where high magnitudes of reward are associated with high risks of failure, elicit suboptimal risk aversion by myopic reinforcement learning. Here, our aim was to investigate whether and, if so, how social learning can promote behavioural-shifts toward more adaptive risk seeking in such decision problems. Previous studies have suggested that biases in individual risk-taking behaviour may be amplified in collectives. One example is group polarisation in description-based gambling tasks, where conformist copying amplifies the initial distribution of risk preferences. Our research explores whether recursive feedback processes between learning and collective performance can rescue individuals trapped in the suboptimal risk aversion. Methods We simulated groups of agents’ behaviour in ‘two-armed bandits’ consisting of one safe and one risky option. Especially, we focused on the situation under which individual reinforcement learning tends to converge on the suboptimal safer option. Agents in a group played the same task simultaneously and each could observe other agents’ behaviour, allowing them to use the frequency-based social learning strategies. Results and conclusions Although individuals were potentially risk-aversive, moderate use of social information promoted more adaptive risk seeking. The pattern was robust over different implementations of social learning strategies. Interestingly, when the risky option’s mean payoff was lower than the safe option’s so that the risk aversion was adaptive, social learning instead promoted risk-aversion. Our results demonstrated an overlooked adaptive benefit of social learning.

Daniel Cooney: PDE Models of Multilevel Selection and the Evolution of Cooperation

In this presentation, we will discuss PDE models for multilevel selection, with an emphasis on studying the evolution of cooperation when there is reproductive competition both between individuals and between groups. We focus on the derivation and analysis of the long-time behavior of the replicator dynamics for multilevel selection. When interactions consist of the Prisoners’ Dilemma, we show that whether the within-group advantage of defectors or the between-group advantage of groups with many cooperators wins out in the long run depends on the relative selection strength at the two levels. A notable finding is that lower-level selection casts a long shadow: if groups are best off with a mix of cooperators and defectors, then there will always be fewer cooperators than optimal at steady state, even in the limit of infinitely strong selection strength at the group level.


Eneko Aspillaga: Unveiling the behavioural variability and social structure of wild fish populations using high resolution acoustic telemetry

Recent developments in acoustic telemetry techniques have revolutionized our ability to study aquatic organisms in the wild. Thanks to the continuous miniaturization of transmitters and the new signal encoding and positioning systems, now it is technically possible to simultaneously monitor from hundreds to thousands of individuals, including different species and different life-stages, obtaining high-resolution movement data. Recently, we have carried out an unprecedent high-resolution tracking experiment focused on our model species, the pearly razorfish (Xyrichtys novacula), in which more than 300 individuals were simultaneously tracked in the marine protected area of Palma Bay (Mallorca, Balearic Islands). We will use the obtained tracks to characterize the variability of the main behavioural traits of individuals in the population (e.g. chronotypes, behavioural modes), as well as their social characteristics and collective movements. In the future, we will combine this information with genetic studies and personality tests carried out in the laboratory, with the main objective of disentangling the causes the observed behavioural variability, its consequences on the social structure of the population, and the possible evolutionary effects of the fishing pressure.

Juliane Lukas: Waving at the enemy: the adaptive anti-predator behavior of two extremophile poeciliids

Animal groups such as fish swarms or bird flocks often respond to attacking predators through wave-like evasion, as information spreads through the group. In Mexico, two poeciliids (genera Poecilia and Gambusia) inhabit sulphide-rich and severely hypoxic springs, where they form several-thousand-strong shoals at the water surface during aquatic surface respiration. During this time, they are particularly vulnerable to avian predation. Following an attack, these fish produce a series of synchronized collective waves by repeatedly diving down in a cascade-like manner. We hypothesize that these repeated waves may function as a distraction and/or deterrent to attacking birds. By analyzing over 500 attacks, we found that birds with pre-attack wave-exposure actively avoid waving areas and show prolonged inter-attack intervals compared to birds that did not experience waves. This is the first study that describes repeated waves as a form of collective information transfer among schooling prey in the wild.

Adrian Palacios Munoz: Dissection of social interactions during male-male courtship behavior in D. melanogaster

In the last decades, advances in the computational tools available have allowed scientists to investigate the rules that govern the organization of animal groups and their emergent properties. Today it is possible to collect data from many experiments involving many animals, track their movements and poses, and quantitatively analyse their behavior. Combining these tools with the genetic and experimental toolkit available for Drosophila melanogaster provides us with the opportunity to manipulate the conditions of their interactions, and consequently their collective behavior. This will provide a causal understanding of the behavioral rules and the neural mechanisms underlying behavior in groups. We use the so-called male-male chaining behavior - a form of homosexual courtship - in which male flies organize themselves into long chains and even circles. As a first step towards understanding the behavioral rules and neural mechanisms by which these group level patterns arise, we used generalized linear models (GLMs) to determine the contribution of different flies in a chain to the behavior of a focal fly. Using this analysis, we provide first insights into how flies integrate information from others in the chain - e.g. if they only attend to the nearest fly or evaluate information from multiple flies in the chain. We tracked the position and pose of flies from 1h video recordings of groups of 10 male flies and detected chaining events based on the distance, the relative position and the orientation of the flies. We considered triads of flies in a chain and trained GLM models to predict the velocity of a focal fly within the triad using absolute and relative kinematic variables of all members of the triad. The GLMs showed good performance (r2 > 0.7). For very short time delays into the future (< 50 ms) self motion is most predictive, but social information becomes more informative at longer delays. To better understand how the information is flowing through the individuals in a chain and their individual contributions to the interaction, we proceed to compare the different models we trained. Our results show that flies in front of the focal fly are more predictive than those behind, suggesting that chaining behavior is mainly driven by chasing the frontal and not escaping the rearward fly. While flies next in the chain are most predictive, flies further in front still contribute independent information to the focal fly's behavior. Additionally, pooling information from multiple frontal flies did not decrease performance, suggesting that flies do not assess social information individual-by-individual, but rather react to the collective behavior of others in the chain. Future directions of the project will include a more complete description of the chaining behavior including orientation, turning and wing extension. Another extension for the models will be the incorporation of the environment’s information, such as the wall of the chamber or of other flies that are not members of the chain. Using optogenetics during behavior and calcium imaging in social virtual reality, we aim to dissect the neural circuits controlling the male-male courtship behavior.

Eduardo Sampaio: Disentangling multi-partner effects on decision-making and cognition in interspecific collaborative hunting between octopus (Octopus cyanea) and fishes

In collective behaviour, inspired by the “self-propelled particle” concept, coordination and group decision-making can be defined from localized interaction rules. However biological (individual or species level) traits can alter the “weight” of individual decision within groups. How collective movement predicted by physics is modulated by heterogeneous morphology, behaviour and cognition-based processes is severely understudied. Parallelly, brain evolution is often associated with the cognitive demands and challenges of social life, since cooperative partners rely on being able to understand non-immediate rewards, constantly re-evaluate costs and benefits, and correctly use partner control mechanisms. Notwithstanding their solitary lifestyle, pairwise hunting between octopus and fish occurs through recently-described complex communicative mechanisms. Howerver, in multi-partner contexts, a network of interactions occurs and the consequent existence of multiple sources of information has the potential to shape cognition and decision-making, by eliciting the emergence of alternative (non-)cooperative strategies (e.g. dynamic leader-follower status, information manipulation). We are analyzing footage of interspecific cooperative hunting events between O. cyanea and multiple partners (i.e. various fish species with different hunting stragies) in Eilat (Israel) and El Quseir (Egypt). To describe the underlying dynamics of these events, multi-level quantitative and qualitative approaches are being used, to model, track, and analyze: 1) Motor patterns of O. cyanea and fish partners, 2) Leadership during collective movements, 3) Relative orientation of interacting partners (e.g. facing towards octopus, towards fish, towards prey), 4) Aggressive interactions (fish biting octopus, octopus chasing fish), 5) Kleptoparasitism (i.e. stealing prey from one another), and 6) Habitat factors for group movement. These events are seemingly highly social, communicative, with great interdependence and member specialization. Understanding the mechanisms underpinning interspecific collaborative interactions can potentially further our understanding of cephalopod cognition, interspecific communication, and collective decision-making in nature.

Claudia Winklmayr: Wisdom Of Stalemates

Most models of collective decision-making assume that groups reach a consensus during a decision-making bout, often through simple majority rule. In many natural and sociological systems, however, groups may fail to reach consensus,resulting in stalemates. Here, we build on opinion dynamics and collective wisdom models to examine how stalemates may affect the wisdom of crowds. We find that stalemates can improve collective accuracy in both simple and complex information environments. We identify network properties that tune the system between consensus and accuracy, providing mechanisms by which animals, or evolution, could dynamically adjust the collective decision-making process in response to the reward structure of the possible outcomes. Overall, these results highlight the adaptive potential of stalemate filtering for improving the decision-making abilities of group-living animals.

Benjamin Wild: Social networks through time - Individuality in a colony of honey bees

Non-negative matrix factorization is a widely used method to analyze and detect community-based structure in social networks. Unfortunately, there's no straightforward extension of this approach to temporal networks. In many model organisms, and particularly in social insects, the patterns of actions and interactions among individuals are not static but constantly evolving over time. This can be due to the emergence or demise of certain individuals, changing task allocation because of temporal polyethism and changes in the environment, or many other reasons. Understanding such temporal patterns in complex social networks remains a challenging problem. In this work, I will present a novel temporal non-negative matrix factorization algorithm that concurrently learns semantic embeddings of all individuals of a temporal social network and a functional mapping from these embeddings to factors representing the state of the individuals at a specific point in time or age. This method can be used to detect clusters of social development of individuals over time, to compare the structure of the networks at different times, or to compare the role of individuals in the social structure in a meaningful way even when they were never alive at the same time.

Yinong Zhao: The full phase diagram of a simple alignment model and the validity of Toner-Tu Theory

Self-propelled particle models are widely used for understanding collective motions, among which the alignment-based models are mostly studied. We implement a minimal, continuous-time SPP model with alignment interaction, and explore its phase diagram for a wide range of parameters. We observe different regimes with disorder-order and gas-liquid phase transition. The Toner-Tu (TT) theory has been proposed to provide a general theoretical description of the large-scale behavior of the spatially homogeneous flocking state. We compare our numerical results with predictions of the TT-theory for parameter regimes where homogeneous states can be observed, and discuss the source of potential deviations from TT theory.