BIROn - Birkbeck Institutional Research Online

    A hidden knowledge discovering approach for past tense and plural problems to language cognition

    Yang, J. and Thomas, Michael S.C. (2016) A hidden knowledge discovering approach for past tense and plural problems to language cognition. In: UNSPECIFIED (ed.) 2016 12th International Conference on Semantics, Knowledge and Grids (SKG). New York, U.S.: IEEE Computer Society, pp. 47-53. ISBN 9781509047956.

    Full text not available from this repository.

    Abstract

    One purpose of using computational models in cognitive related researches is to use them as a tool to understand human's learning processes, especially focusing on understanding the processes of language development. Although connectionist computational models can simulate the learning process in a black-box way, understanding the inside rational relations is as important as the correlations between input and output. Therefore, this problem can be transferred to a machine learning problem: to discover the hidden knowledge(rules) for a non-classification problem domain. Unfortunately, most available rule extraction techniques are focusing on classification problems, few of them can solve non-classification problems. In this paper, we proposed a hidden knowledge exploration model for the Auto-encoder based computational models which implements the following two functions: (1) simulate the language learning process, (2) reveal the correlations between prior knowledge and the new knowledge through re-building the inherent rules. Experiments prove its efficiency by comparing its fidelity performances with the well performed black-box learning technique(Multi Layer Perceptron Artificial Neural Network).

    Metadata

    Item Type: Book Section
    Depositing User: Administrator
    Date Deposited: 07 Mar 2017 12:59
    Last Modified: 07 Mar 2017 12:59
    URI: http://eprints.bbk.ac.uk/id/eprint/18293

    Statistics

    Downloads
    Activity Overview
    0Downloads
    224Hits

    Additional statistics are available via IRStats2.

    Archive Staff Only (login required)

    Edit/View Item Edit/View Item