The Use of Electronic Techniques in the Construction of Machines Which Can Learn

  • J.F. Young

    Student thesis: Master's ThesisMaster of Science (by Research)

    Abstract

    Four different approaches to the design and
    construction of electronic machines capable of displaying
    Pavlovian conditioning by association have
    been investigated theoretically and practically.
    Conflicts in the basic requirements for such a machine
    might be resolved by the introduction of Probability,
    Inhibition and Forgetting. The methods adopted
    should not make it impossible to’ introduce these
    features as and when required later.

    The methods of association information storage
    considered have been :-
    1. Use of recordings of sinusoidal signals resulting
    from beats between sinusoidal input signels. This
    @
    introduced problems of linearity in the recording
    process, and of selection of the input frequencies.
    2. Use of rectangular waves of various frequencies.
    While this method could eliminate linearity problems,
    it introduced problems of addition of new information
    to previously-recorded information.
    36 Use of electrical pulse signals recirculating .
    dynamically. While use of this method could eliminate
    many of the earlier problems, the problem remained of
    noise augmentation in a recirculating system. It is v
    desirable to reduce this if possible.
    4. Use of pulse signals recirculating non-dynamically This fourth. method proved to be the most
    promising, and a machine was constructed using
    this approach. The organisation of this machine
    was based on the earlier work. Methods for further
    simplification are suggested in the thesis.

    Work on this fourth method involved the
    following :-
    1. Consideration of methods of avoidance of
    Spurious response and of possible alternative
    methods of storage of information.
    2. Development of various forms of prime-number
    counter. The method finally adopted used integrated
    circuits.
    3. Investigation of a number of circuits displaying
    majority logic action.
    4. Relationship to other work using multi-dimensional
    theory, and the simplification of the visualisation
    of associations.
    5. Theoretical investigation of magnetic pulse |
    recording waveshapes.
    6. Development of methods of construction of the
    machine.
    Date of Award1969
    Original languageEnglish

    Keywords

    • electronic techniques
    • machine learning
    • electrical engineering

    Cite this

    '