Skip to Main Content
 

Global Search Box

 
 
 

ETD Abstract Container

Abstract Header

Abductive Meta Hypothesis Plausibility Estimation and Selection Policies

Fadnis, Kshitij Prakash
http://rave.ohiolink.edu/etdc/view?acc_num=osu1374064363

Abstract Details

2013, Master of Science, Ohio State University, Computer Science and Engineering.
Abduction, or `inference to the best explanation,’ is a form of logical inference which derives explanations for observations. This study addresses abduction problems which are sequential in nature, i.e., sets of observations arrive sequentially in time and abductive inferencing is used to explain those observations, after each set arrives, in terms of a changing world state, or updated world estimate. The abduction algorithm used for this study tentatively accepts its best explanation, after processing each set of observations, and uses that explanation to inform the generation and evaluation of explanatory hypotheses in subsequent reasoning steps. However, the best explanation at any stage may be partially or completely incorrect, since, typically, alternative explanations cannot be completely ruled out. A mistake might be symptomless, and never caught, or it might cause some subsequent processing difficulty, that can be detected, and used to stimulate an attempt to identify and repair the mistake. One such kind of difficulty is a reasoning state wherein no hypotheses are available to explain one or more observations. Such unexplainable observations are referred to as anomalous observations or No-Explainer anomalies. Previous research by Bharathan and others has shown that meta-reasoning, or reasoning about one’s own reasoning, can sometimes be used to detect, diagnose, and repair reasoning mistakes indicated by anomalous observations. The present study examines the possible causes of anomalous observations, and the methods for attempting to repair mistakes, in sequential abductive reasoning, as they were defined recently by Eckroth. The main contribution of this thesis is to provide computationally inexpensive means to assess the plausibility of each possible cause, and to represent it as the confidence score associated with the corresponding meta-hypothesis, for an abductive meta-reasoning algorithm. Plausibility estimation functions were designed for each type of meta-hypothesis that take into account details of the anomalous observations and information available in the reasoning state. This way of estimating plausibility contrasts with Eckroth’s approach, in which each meta-hypothesis is assessed by evaluating the results of attempted repair, which may be computationally expensive. It is shown experimentally, in two different domains, that these plausibility estimation functions are effective, with only minor losses in correctness compared with Eckroth’s method, and with gains in computational efficiency.
John Josephson (Advisor)
Balakrishnan Chandrasekaran (Committee Member)
66 p.
Computer Science
Abductive meta-reasoning, meta-hypotheses, meta-operations, plausibility estimation functions

Recommended Citations

Citations

  • Fadnis, K. P. (2013). Abductive Meta Hypothesis Plausibility Estimation and Selection Policies [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1374064363

    APA Style (7th edition)

  • Fadnis, Kshitij. Abductive Meta Hypothesis Plausibility Estimation and Selection Policies. 2013. Ohio State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1374064363.

    MLA Style (8th edition)

  • Fadnis, Kshitij. "Abductive Meta Hypothesis Plausibility Estimation and Selection Policies." Master's thesis, Ohio State University, 2013. http://rave.ohiolink.edu/etdc/view?acc_num=osu1374064363

    Chicago Manual of Style (17th edition)

osu1374064363
351
© 2013, some rights reserved.Creative Commons License Abductive Meta Hypothesis Plausibility Estimation and Selection Policies by Kshitij Prakash Fadnis is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by The Ohio State University and OhioLINK.