Modelling homeostatic regulation in multi-objective decision-making
Year: 2018
Type: dissertation
Abstract: This thesis attempts to model homeostatic regulation, a behavioural phenomenon ubiquitous in animals, in the domain of reinforcement learning. We specifically look at multi-objective reinforcement learning that can facilitate multi-variate regulation. When multiple objectives are to be handled, the current framework of Multi-objective Reinforcement Learning proves to be unsuitable without information on some preference over the objectives. We therefore model homeostatic regulation as a motivational process, that selectively activates some objectives over others, and implements cognitive control. In doing so, we utilize cognitive control not as behavioural principle, but as a control mechanism that arises as a natural necessity for homeostatic regulation. We utilize a recent framework for drive reduction theory of reinforcement learning, and attempt to provide a normative account of arbitration of objectives from drives. We show that a purely reactive agent can face difficulties in achieving this regulation, and would require a persistence-flexibility mechanism. This could be handled effectively in our model by incorporating a progress metric. We attempt to build this model with the intention of acting as a natural extension to the current reinforcement learning framework, while also showing appropriate behavioural properties.
Author: Naresh Balaji Ravichandran
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2891130756