Some Aspects of Design and Accounting for a Priori Information when using Cluster Regression to Solve Geological and Geophysical Problems
Event name: Geomodel 2019
Session: Современные методы решения прямых и обратных задач и машинного обучения
Publication date: 17 September 2019
Info: Extended abstract, PDF ( 898.91Kb )
Price: € 20
The gradual complication of geological and geophysical tasks supplied by practice leads to an increase in the specialization of practitioners in the application of geological and geophysical software. At the same time, specialized geophysical software is not always developed by specialists who have an adequate understanding of the complexity and specifics of the tasks being solved, and geophysics practitioners are often not sure about the limitations inherent in the algorithms and the ability to set the algorithm in practice. The situation is aggravated on the one hand by the cooperation of professionals who are insufficiently versed in geophysics (often both programmers and non-specialized managers) and the complexity of the tools used (which do not always understand and explain the degree of likelihood or randomness of the results obtained by the programs). Such a direction of development follows the general trend of simplification and McDonaldization of many technologies, but in the case of geology and geophysics the level of uniqueness and specificity of most geological and geophysical objects limits this trend. In the advanced version of McDonaldization, it is accompanied by an increase in quality control and testing. However, in geophysics, sets of free tests and polygons for solving specific problems are not widely used. As a result, there is a tendency when the software and its developers, instead of the role of assistant specialist, begin to claim the role of a host, whose slave professionals are. One of the answers to the above trends is the formalization and use of a priori geological and geophysical information. The report discusses the features of the use of such information on the example of algorithms such as cluster regression.