Multiagent Learning through Neuroevolution

Miikkulainen, Risto; Feasley, Eliana; Johnson, Leif; Karpov, Igor; Rajagopalan, Padmini; Rawal, Aditya; Tansey, Wesley

doi:10.1007/978-3-642-30687-7_2

Cited by 13 publications

(14 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The evolution of cooperative multi-agent systems might be the next frontier in the context of evolving artificial agents, in which context not much is yet known about conditions that give rise to cooperative behavior and the complex inter-dependencies between individual and group goals [24]. For example, there might be many factors that influence whether the individuals either bow to the group or act by egoistic rules [25].…”

Section: Discussionmentioning

confidence: 99%

How cognitive and environmental constraints influence the reliability of simulated animats in groups

Fischer

Mostaghim

Albantakis

2019

Preprint

View full text Add to dashboard Cite

16Evolving in groups can either enhance or reduce an individual's task performance. Still, 17 we know little about the factors underlying group performance, which may be reduced to 18 three major dimensions: (a) the individual's ability to perform a task, (b) the dependency on 19 environmental conditions, and (c) the perception of, and the reaction to, other group 20 members. In our research, we investigated how these dimensions interrelate in simulated 21 evolution experiments using adaptive agents equipped with Markov brains ("animats"). We 22 evolved the animats to perform a spatial-navigation task under various evolutionary setups.23 The last generation of each evolution simulation was tested across modified conditions to 24 evaluate and compare the animats' reliability when faced with change. Moreover, the 25 complexity of the evolved Markov brains was assessed based on measures of information 26 integration. We found that, under the right conditions, specialized animats were as reliable 27 as animats already evolved for the modified tasks, that interaction between animats was 28 dependent on the environment and on the design of the animats, and that the task difficulty 29 influenced the correlation between the performance of the animat and its brain complexity.30 Generally, our results suggest that the interrelation between the aforementioned dimensions 31 is complex and their contribution to the group's task performance, reliability, and brain 32 complexity varies, which points to further dependencies. Still, our study reveals that 33 balancing the group size and individual cognitive abilities prevents over-specialization and 34 can help to evolve better reliability under unknown environmental situations. 35 Keywords: Collective behavior, evolutionary algorithms, cognitive science, Markov brains. PLOS Computational Biology --FOR REVIEW 3 of 35 36Author Summary 37The ability to adapt to environmental changes is an essential attribute of organisms which 38 have had evolutionary success. We designed a simulated evolution experiment to better 39 understand the relevant features of such organisms and the conditions under which they 40 evolve: First, we created diverse groups of cognitive systems by evolving simulated 41 organisms ("animats") acting in groups on a spatial-navigation task. Second, we post-42 evolutionary tested the final evolved animats in new environments-not encountered before-43 in order to test their reliability when faced with change. Our results imply that the ability to 44 generalize to environments with changing task demands can have complex dependencies on 45 the cognitive design and sensor configuration of the organism itself, as well as its social or 46 environmental conditions. 47Introduction 48Intelligence is the ability to adapt to changes. According to this prevalent perspective, 49 possessing general intelligence [1,2] not only enables one to perform a task correctly under 50 already known conditions, but also to perform well under unexpected conditions. Further, in 51 natural ...

show abstract

Section: Discussionmentioning

confidence: 99%

How cognitive and environmental constraints influence the reliability of simulated animats in groups

Fischer

Mostaghim

Albantakis

2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Similar to Olson et al [21], they also evolved swarm behavior in a predator-prey scenario, but with a learning technology based on artificial neural networks (ANN). Miikkulainen et al [17] reviewed the work around neuroevolution and discussed future research topics in this field. They concluded that cooperative multiagent systems are the next frontier of neuroevolution and that research in this field is still in an early stage.…”

Section: Related Workmentioning

confidence: 99%

“…The present results were obtained from one particular task environment. To build empirical strength, implementing more difficult and diverse tasks will be required, e.g., the predator-prey scenario used in earlier works by Olson et al and Miikkulainen et al [17,21]. Additionally, it is important to investigate variations in the animats' design to determine how the kind and number of sensors influence their behavior.…”

Section: Future Workmentioning

confidence: 99%

How swarm size during evolution impacts the behavior, generalizability, and brain complexity of animats performing a spatial navigation task

Fischer¹,

Mostaghim

Albantakis

2018

Proceedings of the Genetic and Evolutionary Computation Conference

View full text Add to dashboard Cite

While it is relatively easy to imitate and evolve natural swarm behavior in simulations, less is known about the social characteristics of simulated, evolved swarms, such as the optimal (evolutionary) group size, why individuals in a swarm perform certain actions, and how behavior would change in swarms of different sizes. To address these questions, we used a genetic algorithm to evolve animats equipped with Markov Brains in a spatial navigation task that facilitates swarm behavior. The animats' goal was to frequently cross between two rooms without colliding with other animats. Animats were evolved in swarms of various sizes. We then evaluated the task performance and social behavior of the final generation from each evolution when placed with swarms of different sizes in order to evaluate their generalizability across conditions. According to our experiments, we find that swarm size during evolution matters: animats evolved in a balanced swarm developed more flexible behavior, higher fitness across conditions, and, in addition, higher brain complexity.

show abstract

“…Artificial neural networks are in theory capable of approximating any behavior with arbitrary accuracy if the weights and topology are correctly set (Haykin, 1999). Also, evolutionary computation has proven to be a useful method for learning how to configure the weights and topologies of neural networks in order to solve many problems (Floreano and Urzelai, 2000;Yao and Liu, 1997;Miikkulainen et al, 2012). When evolutionary computation is used to evolve artificial neural networks, it is called neuroevolution.…”

Section: Neuroevolutionmentioning

confidence: 99%

Evolving multimodal behavior through modular multiobjective neuroevolution

Schrum

2015

SIGEVOlution

View full text Add to dashboard Cite

Intelligent organisms do not simply perform one task, but exhibit multiple distinct modes of behavior. For instance, humans can swim, climb, write, solve problems, and play sports. To be fully autonomous and robust, it would be advantageous for artificial agents, both in physical and virtual worlds, to exhibit a similar diversity of behaviors. Artificial evolution, in particular neuroevolution [3, 4], is known to be capable of discovering complex agent behavior. This dissertation expands on existing neuroevolution methods, specifically NEAT (Neuro-Evolution of Augmenting Topologies [7]), to make the discovery of multiple modes of behavior possible. More specifically, it proposes four extensions: (1) multiobjective evolution, (2) sensors that are split up according to context, (3) modular neural network structures, and (4) fitness-based shaping. All of these technical contributions are incorporated into the software framework of Modular Multiobjective NEAT (MM-NEAT), which can be downloaded here.

show abstract

Multiagent Learning through Neuroevolution

Cited by 13 publications

References 32 publications

How cognitive and environmental constraints influence the reliability of simulated animats in groups

How cognitive and environmental constraints influence the reliability of simulated animats in groups

How swarm size during evolution impacts the behavior, generalizability, and brain complexity of animats performing a spatial navigation task

Evolving multimodal behavior through modular multiobjective neuroevolution

Contact Info

Product

Resources

About