The evolution of reaction-diffusion controllers for minimally cognitive agents.

This article describes work carried out to investigate whether a classic reaction-diffusion (RD) system could be used to control a minimally cognitive animat. The RD system chosen was that first described by Gray and Scott, and the minimally cognitive behaviors were those used by Beer et al. involving the fixation and discrimination of diamond and circle shapes by a whiskered animat. A further task was added, which required the RD controllers to maintain and use a chemical memory. The parameters of these controllers were evolved using an evolutionary, or genetic, algorithm.

Weak and strong priming cues in bumblebee contextual learning.

Bees have the flexibility to modulate their response to a visual stimulus according to the context in which the visual stimulus is seen. They readily learn that in one context a yellow target, but not a blue one, should be approached to reach sucrose and that in another context blue, but not yellow, leads to sucrose. Here we contrast the bumblebee's ability to use two types of contextual or priming cue in deciding which of two coloured targets to approach. Bumblebees could perform this task well when the pairs of colours to be discriminated were in two different places, so that the cues associated with each place indicated whether the bees should select a blue or a yellow target. In this case the priming cues were presented concurrently with the rewarded stimuli. Priming cues, which occur a little earlier than a rewarded stimulus, seem less powerful in their ability to bias a bee's choice of rewarded stimulus. We tried with a variety of methods to train bees to use a priming colour to indicate which of two colours should be approached a few seconds later. Our only success with such sequential priming cues was when each pair of rewarded and unrewarded colours could be distinguished by additional spatial cues. Bees were trained to choose a blue-black checkerboard over a yellow-black checkerboard, after viewing a yellow priming cue, and to choose a uniform yellow target over a uniform blue one, after viewing a blue priming cue. They performed this task almost without error. To see whether bees had associated each rewarded stimulus with the relevant sequential priming cue, bees were tested with a choice between the two rewarded stimuli (the yellow target and the blue-black checkerboard). The bees' choice was biased towards the blue-black checkerboard, when the preceding priming cue was yellow, and towards the yellow target, when the priming cue was blue. We suppose that the experiment works because the presence or absence of the checkerboard provides an additional distinguishing spatial cue that can be linked to and reinforce the sequential one. Under natural conditions, as when bees follow routes, there will normally be such a synergy between spatial and sequential cues.

Learning speed and contextual isolation in bumblebees.

Bumblebees will learn to approach one of a pair of patterns (a 45 degrees grating) and to avoid the other (a 135 degrees grating) to reach a feeder, and to do the opposite to reach their nest (approach a 135 degrees grating and avoid a 45 degrees grating). These two potentially competing visuo-motor associations are insulated from each other because they are set in different contexts. We investigated what training conditions allow the two sets of associations to be acquired without mutual interference. If the discrimination at the feeder has already been learnt, then the discrimination at the nest can be readily acquired without disrupting the bees' performance at the feeder. But, if the two are learnt simultaneously, there is mutual interference. Prior experience of the two contexts before the discriminations are learnt does not prevent interference. We conclude that visual patterns and contextual cues must already be associated with each other for a visuo-motor association to be isolated from the interfering effects of a competing association that is acquired in a separate context. This pattern of results was mimicked in a simple neural network with Hebbian synapses, in which local and contextual cues were bound together into a configural unit.