The total knowledge contained within a collective supersedes the knowledge of even its most intelligent member. Yet the collective knowledge will remain inaccessible to us unless we are able to find efficient knowledge aggregation methods that produce reliable decisions based on the behavior or opinions of the collective's members. It is often stated that simple averaging of a pool of opinions is a good and in many cases the optimal way to extract knowledge from a crowd. The method of averaging has been applied to analysis of decision-making in very different fields, such as forecasting, collective animal behavior, individual psychology, and machine learning. Two mathematical theorems, Condorcet's theorem and Jensen's inequality, provide a general theoretical justification for the averaging procedure. Yet the necessary conditions which guarantee the applicability of these theorems are often not met in practice. Under such circumstances, averaging can lead to suboptimal and sometimes very poor performance. Practitioners in many different fields have independently developed procedures to counteract the failures of averaging. We review such knowledge aggregation procedures and interpret the methods in the light of a statistical decision theory framework to explain when their application is justified. Our analysis indicates that in the ideal case, there should be a matching between the aggregation procedure and the nature of the knowledge distribution, correlations, and associated error costs. This leads us to explore how machine learning techniques can be used to extract near-optimal decision rules in a data-driven manner. We end with a discussion of open frontiers in the domain of knowledge aggregation and collective intelligence in general.

Traveling waves (from action potential propagation to swimming body motions or intestinal peristalsis) are ubiquitous phenomena in biological systems and yet are diverse in form, function, and mechanism. An interesting such phenomenon occurs in cephalopod skin, in the form of moving pigmentation patterns called "passing clouds". These dynamic pigmentation patterns result from the coordinated activation of large chromatophore arrays. Here, we introduce a new model system for the study of passing clouds, Metasepia tullbergi, in which wave displays are very frequent and thus amenable to laboratory investigations. The mantle of Metasepia contains four main regions of wave travel, each supporting a different propagation direction. The four regions are not always active simultaneously, but those that are show synchronized activity and maintain a constant wavelength and a period-independent duty cycle, despite a large range of possible periods (from 1.5 s to 10 s). The wave patterns can be superposed on a variety of other ongoing textural and chromatic patterns of the skin. Finally, a traveling wave can even disappear transiently and reappear in a different position ("blink"), revealing ongoing but invisible propagation. Our findings provide useful clues about classes of likely mechanisms for the generation and propagation of these traveling waves. They rule out wave propagation mechanisms based on delayed excitation from a pacemaker but are consistent with two other alternatives, such as coupled arrays of central pattern generators and dynamic attractors on a network with circular topology.

Animals moving in groups coordinate their motion to remain cohesive. A large amount of data and analysis of movement coordination has been obtained in several species, but we are lacking theoretical frameworks that can derive the form of coordination rules. Here, we examine whether optimal control theory can predict the rules underlying social interactions from first principles. We find that a control rule which is designed to minimize the time it would take a pair of schooling fish to form a cohesively moving unit correctly predicts the characteristics of social interactions in fish. Our methodology explains why social attraction is negatively modulated by self-motion velocity and positively modulated by partner motion velocity, and how the biomechanics of fish swimming can shape the form of social forces. Crucially, the values of all parameters in our model can be estimated from independent experiments that need not relate to measurement of social interactions. We test our theory by showing a good match with experimentally observed social interaction rules in zebrafish. In addition to providing a theoretical rationale for observed decision rules, we suggest that this framework opens new questions about tuning problems and learnability of collective behaviours.

No abstract

Most animals fight by repeating complex stereotypic behaviours, yet the internal structure of these behaviours has rarely been dissected in detail. We characterized the internal structure of fighting behaviours by developing a machine learning pipeline that measures and classifies the behaviour of individual unmarked animals on a sub-second time scale. This allowed us to quantify several previously hidden features of zebrafish fighting strategies. We found strong correlations between the velocity of the attacker and the defender, indicating a dynamic matching of approach and avoidance efforts. While velocity matching was ubiquitous, the spatial dynamics of attacks showed phase-specific differences. Contest-phase attacks were characterized by a paradoxical sideways attraction of the retreating animal towards the attacker, suggesting that the defender combines avoidance manoeuvres with display-like manoeuvres. Post-resolution attacks lacked display-like features and the defender was avoidance focused. From the perspective of the winner, game-theory modelling further suggested that highly energetically costly post-resolution attacks occurred because the winner was trying to increase its relative dominance over the loser. Overall, the rich structure of zebrafish motor coordination during fighting indicates a greater complexity and layering of strategies than has previously been recognized.

Theoretical studies of ecosystem models have generally concluded that large numbers of species will not stably coexist if the species are all competing for the same limited set of resources. Here, we describe a simple multi-trait model of competition where the presence of N resources will lead to the stable coexistence of up to 2N species. Our model also predicts that the long-term dynamics of the population will lie on a neutral attractor hyperplane. When the population shifts within the hyperplane, its dynamics will behave neutrally, while shifts which occur perpendicular to the hyperplane will be subject to restoring forces. This provides a potential explanation of why complex ecosystems might exhibit both niche-like and neutral responses to perturbations. Like the neutral theory of biodiversity, our model generates good fits to species abundance distributions in several datasets but does so without needing to evoke inter-generational stochastic effects, continuous species creation or immigration dynamics. Additionally, our model is able to explain species abundance correlations between independent but similar ecosystems separated by more than 1400 km inside the Amazonian forests.

Abstract:11 Animals often assess each other by paying special attention to signals, which 12 help to communicate the quality of each individual. When there is a conflict 13 of interest between the signaler and the receiver, then the signaler has an 14 incentive to cheat by producing signals which exaggerate its apparent quality. 15One opportunity for cheating might be to rely on sensory illusions, but it has

We propose a new readout architecture for echo state networks where multiple linear readout modules are activated at distinct time points to varying degrees by a separate controller module. The controller module, like the reservoir of the echo state network, can be initialized randomly. All linear readout modules are trained through simple linear regression, which is the only adaptive step in the modified algorithm. The resulting architecture provides modest improvements on a variety of time series processing tasks (between 5 to 50% in performance metric depending on the task studied). The novel architecture is guaranteed to perform at least as accurately as a conventional linear readout. It can be utilized as a general purpose readout method when augmentations to performance relative to the standard method is needed.