Pattern recognition and active vision in bees.

Abstract

Bees represent a common model for studying learning and memory. Previous research on their cognition has primarily focused on similarities between humans and bees, but such results only answer the dichotomous question “are bees able to do X or not ?”. The mechanisms underlying these abilities remain to be investigated. Bees possess a miniature brain and a visual system very different from our own; a large visual field, fixed eyes and limited stereopsy aas well as a spatial resolution ~100 times lower than ours, but a higher temporal resolution. This enables them to process visual information more quickly, seeing many more “images” per unit time than humans. These characteristics suggest that the strategies they use during the acquisition, storing and use of visual information might fundamentally differ from those of humans. In this thesis, I test a hypothesis concerning how bees acquire and store visual information. Due to the specificities of their vision and the relatively limited information storage and processing capacity, I propose that bees rely on active sampling of their surroundings, using stereotypical body movements when scanning object edges to develop a sensory-motor memory. My hypothesis was supported by the results of of chapter 2, 4 and 6 especially where I showed that bees would scan specific parts of patterns depending on the task and these active scanning movements appeared to represent consistent translations of the presented stimuli, and bees developed strategies during training to discriminate rewarding patterns from non-rewarding. They concentrated their efforts on visually salient features (e.g. chapter 2 and 4), reducing the number of movements required to discriminate between objects. This represents an elegant, simple solution to seemingly complex problems, and explains how invertebrates with miniature brains achieve performances comparable to those of mammals. These findings may help inform the fields of robotics and machine vision, particularly when exploring and developing powerful, dynamic neural networks to process visual information.