Fig. 3.
From: Honeybee communication during collective defence is shaped by predation
Theoretical model of one defensive event. We model the colony as an ensemble of N=100 identical bees, which are artificial learning agents that decide whether to sting (“Sting” and S) or not (“Chill” and ∙) based on their sensory perception. These percepts include the alarm pheromone concentration (binned logarithmically from 0 to 8) and a visual signal that the predator is leaving, vESC. The 100 bees act sequentially, so there are 100 time steps, and each stinging bee releases one unit of alarm pheromone (small ticks on middle panel’s y axis), so that the sensory environment of a bee is defined by the behaviour of previous bees. A predator attacks the colony and kills k bee per time step from the time it reaches the colony, tatt, until it receives a certain number of stings sth. At this point, it stops killing, but is still in the vicinity for Δtv time steps before truly escaping, modelled as the activation of vESC for the remaining bees after time Δtv. Once every bee has made a decision, the outcome of the defensive event is evaluated and the individual decision process is updated based on the colony performance (reward factor R, proportional to the number of remaining live bees) for each percept (glow matrix g), with some forgetting (γ). The upper panels show the internal structure of bees and the predator’s parameters, the middle panel the time course of the bees’ perception and the bottom panel the behaviour of both bees and predator during an example trial