Neural networks reveal emergent properties of collective learning in democratic but not despotic groups

Morford, Joe; Lewin, Patrick; Biro, Dora; Guilford, Tim; Padget, Oliver; Collet, Julien

doi:10.1016/j.anbehav.2022.09.020

Cited by 4 publications

(17 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, while in this case what the individual shoal members had learnt within the collective seemed similar to what single individuals learnt, this may not necessarily be the case. Theoretical simulations using groups of machine-learning agents have indeed shown that members of a group can learn very different solutions to a task than identical machine-learning agents solving the same task alone [ 23 , 24 ], as we will see in more detail later (§3c).…”

Section: Experimental Examples Of Collective Learningmentioning

confidence: 99%

“…While it is generally assumed in many collective movement models that the rules governing these direct interactions are fixed (e.g. [ 23 , 24 , 52 ]), in theory they could be influenced by the learning capacities of members. In associative learning terms, individuals could form an association between what they perceive about others and how best to respond (i.e.…”

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

“…Yet to our knowledge this latter avenue has seldom been explored by modellers: even when learning rules are implemented in biologically inspired models (e.g. [ 23 , 24 , 52 ]), all models we know of implement direct interactions as a fixed rather than as a learnable and tuneable rule: learning is directed towards other components of the behaviour. It should be noted that this is not the case in Artificial Intelligence (AI) studies [ 72 ], which may thus offer up useful insights for biologists.…”

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

“…A later example, perhaps more intuitive, is provided by a different modelling simulation [ 24 ]. There, neural network agents were paired to collectively navigate.…”

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

“…The magnitude of the difference between this collective direction and the ‘correct’ direction (arbitrarily chosen but constant across trials) was used to modulate the reinforcement of each neural network of the pair in its individually proposed direction. Across trials, the collective direction of most of these pairs converged towards the correct direction (collective learning) [ 24 ]. However, in most of these pairs, none of the individual neural network members converged to individually propose the correct direction.…”

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

See 4 more Smart Citations

Mechanisms of collective learning: how can animal groups improve collective performance when repeating a task?

Collet

Morford

Lewin

et al. 2023

Phil. Trans. R. Soc. B

Self Cite

View full text Add to dashboard Cite

Learning is ubiquitous in animals: individuals can use their experience to fine-tune behaviour and thus to better adapt to the environment during their lifetime. Observations have accumulated that, at the collective level, groups can also use their experience to improve collective performance. Yet, despite apparent simplicity, the links between individual learning capacities and a collective's performance can be extremely complex. Here we propose a centralized and broadly applicable framework to begin classifying this complexity. Focusing principally on groups with stable composition, we first identify three distinct ways through which groups can improve their collective performance when repeating a task: each member learning to better solve the task on its own, members learning about each other to better respond to one another and members learning to improve their complementarity. We show through selected empirical examples, simulations and theoretical treatments that these three categories identify distinct mechanisms with distinct consequences and predictions. These mechanisms extend well beyond current social learning and collective decision-making theories in explaining collective learning. Finally, our approach, definitions and categories help generate new empirical and theoretical research avenues, including charting the expected distribution of collective learning capacities across taxa and its links to social stability and evolution. This article is part of a discussion meeting issue ‘Collective behaviour through time’.

show abstract

Section: Experimental Examples Of Collective Learningmentioning

confidence: 99%

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

“…A later example, perhaps more intuitive, is provided by a different modelling simulation [ 24 ]. There, neural network agents were paired to collectively navigate.…”

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

Section: Mechanisms Of Collective Learning: a Classificationmentioning

confidence: 99%

See 3 more Smart Citations

Mechanisms of collective learning: how can animal groups improve collective performance when repeating a task?

Collet

Morford

Lewin

et al. 2023

Phil. Trans. R. Soc. B

Self Cite

View full text Add to dashboard Cite

show abstract

Lévy movements and a slowly decaying memory allow efficient collective learning in groups of interacting foragers

Falcón-Cortés

Boyer

Aldana

et al. 2023

Preprint

View full text Add to dashboard Cite

Many animal species benefit from spatial learning to adapt their foraging movements to the distribution of resources. Learning involves the collection, storage and retrieval of information, and depends on both the random search strategies employed and the memory capacities of the individual. For animals living in social groups, spatial learning can be further enhanced by information transfer among group members. However, how individual behavior affects the emergence of collective states of learning is still poorly understood. Here, with the help of a spatially explicit agent-based model where individuals transfer information to their peers, we analyze the effects on the use of resources of varying memory capacities in combination with different exploration strategies, such as ordinary random walks and Lévy flights. We find that individual Lévy displacements associated with a slow memory decay lead to a very rapid collective response, a high group cohesion and to an optimal exploitation of the best resource patches in static but complex environments, even when the interaction rate among individuals is low.

show abstract

Automatic Multi-Class Collective Motion Recognition Using a Decision Forest Extracted from Neural Networks

Abpeikar,

Kasmarik,

Garratt

2023

2023 IEEE Region 10 Symposium (TENSYMP)

View full text Add to dashboard Cite

Neural networks reveal emergent properties of collective learning in democratic but not despotic groups

Cited by 4 publications

References 32 publications

Mechanisms of collective learning: how can animal groups improve collective performance when repeating a task?

Mechanisms of collective learning: how can animal groups improve collective performance when repeating a task?

Lévy movements and a slowly decaying memory allow efficient collective learning in groups of interacting foragers

Automatic Multi-Class Collective Motion Recognition Using a Decision Forest Extracted from Neural Networks

Contact Info

Product

Resources

About