University of Sussex
Browse

File(s) not publicly available

Conditioning and Concept Formation in Embodied Agents

presentation
posted on 2023-06-08, 06:04 authored by Terry Stewart, Sharon Wood
Learning algorithms typically model the acquisition of con- ceptual knowledge from some start state to some fixed learned end state. Natural associative learning demonstrates a more comprehensive range of processes which complement this static view of learning. An experimental regimen is pre- sented for evaluating learning algorithms against this wider remit. This approach provides a general basis for analysing performance and measuring concept formation. We use it here to examine the Distributed Adaptive Control (DAC2) model.

History

Publication status

  • Published

Page range

pp 75-79

Pages

5.0

Presentation Type

  • paper

Event name

AAAI Spring Symposium in Learning Grounded Representations

Event location

Stanford, USA

Event type

conference

ISBN

1577351398

Department affiliated with

  • Informatics Publications

Notes

Originality: Previous work discusses performance in an incomplete way; this work uses a methodology to identify gaps and weaknesses in performance in order to provide a more complete view of the phenomena investigated. Discusses requirements for the achievement of `conditional learning in autonomous/robotic systems based on existing work of others towards this goal and the limitations and omissions identified therein. Rigour: The work uses a systematic methodology to provide a careful analysis of evidence to support claims in the literature to have achieved `conditional learning in an autonomous system, in relation to psychological theory. Significance: Demonstrates ability to identify weaknesses and deficits in performance through application of the methodology presented. This is used to systematically establish the extent to which the criteria for, specifically, `classical conditioning and, non-specifically, other forms of (associative) learning have been met. The use of this methodology is novel in application to autonomous systems, to evaluate performance in relation to all predictions within the paradigm, rather than a select few (for which the system may have been designed). Impact: First meeting in a burgeoning area.

Full text available

  • No

Peer reviewed?

  • Yes

Legacy Posted Date

2012-02-06

Usage metrics

    University of Sussex (Publications)

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC