| Abstract No.: | C-G3184 |
| Country: | Canada |
| | |
| Title: | AN ECOLOGICAL MODEL OF BEHAVIOUR SELECTION |
| | |
| Authors/Affiliations: | 1 Ignasi Cos-Aguilera*; 2 Gillian Hayes; 3 Lola Caņamero;
1 Universite de Montreal, QC, Canada; 2 University of Edinburgh; 3 University of Hertfordshire, Edinburgh, United Kingdom
|
| | |
| Content: | A model to learn potentialities of action is proposed and integrated in an actor-critic reinforcement learning algorithm. This is aimed has been integrated in a situated agent, and simulated to provide behavioural predictions on the arbitration among several courses of action. The model is based on the core principles of motivation and reinforcement driving behaviour selection and learning in biological beings. Its working principle is based on the capacity of perceiving changes in the environment via internal hormonal responses and of modifying the agent's sensorimotor neural associations accordingly. These changes are based on reward delivery, constraint by the requirement of physiological stability introduced by Ashby. The definition of reward used for learning is defined in the agent's internal physiological space, where the effect of interacting with the environment can be qualified in an ethologically consistent manner.
The model is inspired after evolutionary biology mechanisms, used for adaptation to a variety of environments. A central element of this model is the consideration of the agent as the central point of reference. The agent and the environment have to be considered as two parts of a single entity, related by a dynamics of interaction. As it occurs with a biological being, it only perceives what it perceives, either from the environment or from its own internal physiology. Within this framework, behaviour selection has been commonly related to theories of motivation, and learning has been bound to theories of reinforcement. Our model provides a possible manner to integrate both processes from an ecological perspective. If the selection of a particular behaviour is an expression of the agent's motivations, the feedback from the environment can also be viewed as part of the same process, since it also influences the agent's internal motivations and the learning processes via reinforcement. This model naturally constraints the combination of stimuli in an ecological fashion, leading to adaptive behavioural patterns. The patterns obtained confirm behavioural patterns obtained via additive of multiplicative formula combining inner and outer stimuli in classical ethology. This suggests that reward and motivation are probably supported by the same neural substrate, hence relating the elicitation of behaviour to the perception of reward resulting from its execution. |
| | |
| Back |
|
