Avian magneto-receptive home-point navigation: A 2-Dimensional representation and analysis

A neural network project modelling avian magnetoreception systems.

Abstract

This paper examines avian magnetoreception sensory mechanisms and their implications for navigational behaviors focusing on migratory birds’ ability to navigate back to a “home point” using geomagnetic cues. We integrate biophysical models of magnetoreception into a neural network designed to simulate avian navigation by downscaling the task onto a 2-dimensional plane. Two primary receptors are explored: the “compass” receptor in the retina, which operates via a radical pair mechanism sensitive to specific light and magnetic field conditions, as well as the magnetite-based receptor in the upper beak, functioning as a magnetometer to provide magnetic field strength cues. The neural network created here employs these inputs to determine the return heading from a randomly generated location on the plane towards a predesignated home point. Methods include the generation of a synthetic 2-dimensional magnetic field, a winner-take-all synapse action model demonstrating efficacy of reorientational behaviour in the presense of adequate light, and finally a trained neural network model that functions to provide the ‘direct to home’ heading from the simulated avian sensory inputs. Results indicate that the model can effectively learn and mimic a simple navigation strategy, pointing to significant potential for further research into more biologically accurate neural models and their applications in understanding complex navigational systems in birds.

See the full paper here and the relevant modelling notebook here.