Abduction as Incomplete Parameter Estimation
Keywords: Autoregressive model, Parameter Estimation, Observation, Cauchy distribution, Power law
AbstractAbduction is a kind of logical inference, and has been studied in computer science and artificial intelligence (Fin- lay and Dix 1996). Recently, Sawa and Gunji (2010) introduced a diagram to represent three types of inference: i.e. deduc- tion, induction, and abduction, which are articulated by C.S.Peirce. Sawa-Gunji’s representation provides a new approach to a numerical aspect of abduction. In the present paper, we show that Sawa-Gunji's representation of abduction is consistent with Finlay-Dix's one, and integrate the two representations. Both parameter estimation and abduction occupy a similar position on the integrated representation, although they are not completely corresponding. We present "incomplete" pa- rameter estimation as a sort of "simulated abduction", which is a numerical aspect of abduction. It is applied to a first-order autoregressive (AR(1)) model. As a result of numerical analyses on AR(1), the incompletely estimated parameter (IEP) follows a Cauchy distribution, which has a power law of the slope -2 in the tail, although conventionally estimated parameter is normally distributed. It is shown that the Cauchy distribution of the IEP is based on structure of ratio distribution of normal random variables generated from the AR(1). This research suggests that the distribution of the IEP is not based on a mech- anism of system itself, but on relationship between data structure on the given system (i.e. the given AR(1) process) and one on the system observer (i.e. the estimator of the AR(1) parameter).
Special Issue: Towards a New Science of Information
tripleC is a peer-reviewed, open-access journal (ISSN: 1726-670X). All journal content, except where otherwise noted, is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Austria License.