Tuesday, April 17, 2018
4:00-5:00 PM EDT | Show in my time zone
The rate of mineral deposit discovery has fallen significantly in the past decade in part due to lack of “easy targets”, which begs the question as to whether the industry’s exploration data processing needs to be reviewed. The last years have seen a major shift towards the application of the rapidly evolving science of Machine Learning to provide new interpretations of data sets. Most of mining companies possess large amount of data which could hold clues to the understanding and interpretation of mineralized systems. Our ability to harness the predictive capabilities of these data sets does not need to be limited to JUST the power of the human mind, which lack the multidimensional correlation capabilities.
However, one should beware of the limiting factors associated with Machine Learning, of which, domain adaptation and learning bias are the most significant for the application to geological data. Careful considerations must be taken when building a predictive model from data associated to a mine or near mine domain and applying it to reginal domains.
Input data and learning sets are key to building a good performing predictive model. However, one must be able to judge on the performance and results from the algorithm, because a result will always be produced. The geological expertise is paramount to generate credible results when using Machine Learning on geological data sets.
Jean-Philippe Paiement graduated from Université du Québec à Montréal with a B.Sc. in Resources Geology and from Université Laval with a M.Sc. in Metallogeny and Geochemistry. His main fields of interest are centered on the application of new technologies for...Read More