GeneratOre
GeneratOre is a proposed Discovery Machine based on Probabilistic Inference which will be used as a prospecting tool and as an automatic means for mineral property evaluation. The function of the tool is to find and delineate deposits of gold ore on several different scales including:
• District-wide location and evaluation of potential properties
• Property scale prospecting, drill hole location and reserve refinement
• Mine scale ore modeling and reserves estimation
A Discovery Machine is a type of computer learning system designed to solve physical problems, particularly ones in which evidence might be associated with more than one physical event. For example, clay alteration might be an indicator of oxydation state at the time of gold ore deposition, it might be due to a secondary event, or it might be an indicator of structure and faulting. Discovery Machines are made of two integrated parts, the Machine itself which is used to evaluate the probability of a hypothetical physical system and an Inference Engine which drives the Machine in order to answer specific questions.
GeneratOre is designed to simulate every factor involved in a problem domain. In this case the domain is geology and gold ore genesis. Factors are linked using physical laws, statistics or both. Like any real machine, it is best described by a schematic diagram in which individual components of the Machine are linked to the other components which directly affect them. An example diagram for the GeneratOre Machine is given in Figure 1. In addition to being able to use GeneratOre to answer questions, the advantages of building it this way are:
--Relationships, knowledge and uncertainty are explicitly encoded.
--A rigorous statistical framework is provided for evaluating and storing historical data.
--Insight is provided into the processes that geologists use to draw conclusions from varied bits of evidence.
--Subsets of GeneratOre can be run independently for other tasks such as formulating standards for geophysical or other data collection.
--The graphical nature of GeneratOre makes it simple to explain to executives, investors and potential partners.
--Unlike a Neural Network or other black box algorithm which computes outputs from a fixed set of inputs, the GeneratOre machine is transparent and flexible.
--Discovery Machines are patentable.
GeneratOre does not simply learn to compute outputs based on pattern recognition. It seeks to explain data in terms of fundamental processes. In order to do this correctly, as much domain expertise (that is, knowledge from geoscientists) must be built into the Machine as possible. Not only physical knowledge, but intuition and experience need to be translated into explicit mathematics and compared to historical data by experienced statisticians. Making GeneratOre work optimally will require a strong partnership between Chatelet Resources and the client group which will provide the necessary domain knowledge and data. This tool would most valuably developed as a Joint Venture Partnership.