Ruh, N. and Cooper, Richard P. and Mareschal, Denis (2005) A reinforcement model of sequential routine action. In: Honkela, T. and Koenoenen, V. and Poellae, M. and Simula, O. (eds.) Proceedings of the International and Interdisciplinary Conference on Adaptive Knowledge Representation and Reasoning. Helsinki, Finland: Helsinki University of Technology Press, pp. 65-70.
Abstract
In a recent paper, Botvinick & Plaut [1] claim to capture the relevant empirical data on a sequential routine action task in a single embedded SRN (simple recurrent network). In this type of model, task representations are most appropriately described as an emergent property, that is, in terms of the continuous attractor dynamics in the network’s hidden layer. Two major shortcomings of this model are identified: (a) the implausibility of the learning regime and (b) its inability to account for the goal directedness in human behavior. In this paper we suggest that reinforcement learning can overcome both of these problems. We present a first attempt to implement hierarchical routine action in a distributed reinforcement model. Employing two connectionist networks in an actor/critic architecture with TD(temporal difference)-learning it is shown that the model learns to negotiate the high-dimensional state space of a simplified routine action task. This is possible because of the delicate interplay of several principles, each of which simplifies the learning of the task at hand in its own fashion. The implications of the emergent representations in this type of model are discussed.
Metadata
| Item Type: | Book Section |
|---|---|
| School: | Birkbeck Faculties and Schools > Faculty of Science > School of Psychological Sciences |
| Depositing User: | Sarah Hall |
| Date Deposited: | 16 Sep 2019 15:06 |
| Last Modified: | 02 Aug 2023 17:54 |
| URI: | https://eprints.bbk.ac.uk/id/eprint/28959 |
Statistics
Additional statistics are available via IRStats2.
Archive Staff Only (login required)
