MODELING RECOGNITION OF SIGNALS USING ARTIFICIAL NEURAL NETWORKS

Neural networks are powerful tools that overcome constraints inherent in conventional computational methods, and its application extends into the sciences, medicine, engineering and finance. Within the field of animal behaviour, they have been used to model evolutionary processes, classify animal signals, and investigate stimulus-response relationships. Within its application to animal communication, neural network input has comprised artificial stimuli, or if based on real animal signals, examples from bioacoustics. Modelling of movement-based visual signals has received much less attention. We examined the feasibility of applying neural networks to understanding the design of movement-based visual signals, using static input summed across time (static feed-forward networks), as well as dynamic networks where the temporal component of the signal and noise is preserved (dynamic feed-forward and recurrent networks). Results indicated that such models can be trained to distinguish the movement characteristics of a Jacky dragon tail-flick and wind-blown plants, and that they can generalise successfully to unseen exemplars.

1. Concepts in neural network design

• Network units
• Network architecture
• Training networks

2. Neural Network modelling in animal behaviour

• Modelling evolution
• Classification of signals
• Predicting response to animal signals

3. Modelling recognition of tail-flicks by the Jacky dragon

4. References