Post-doctoral research Big Data Mineral Exploration

Post-doc position: Machine Learning & Value of Information for Battery Metals Exploration

Principal Investigator
Jef Caers, Professor of Geological Sciences, Stanford University

Director, Stanford Center for Earth Resource Forecasting

Sponsoring Company
KoBold Metals

We are seeking a postdoctoral researcher for a 2-year position to research the application of machine-learning driven Value-of-Information (VOI) to mineral exploration in the near-mine environment. The candidate will collaborate directly with KoBold Metals (KoBold), a San Francisco Bay Area company that is developing artificial intelligence to improve the efficacy and efficiency of mineral exploration. The collaboration with KoBold will focus on the near-mine environment, and it will provide data from existing mines, as well as undeveloped deposits, to enable the application of machine learning to resource expansion exploration. This research will investigate how the VOI decision-theoretic can optimize and guide mineral exploration, in the near-mine environment, by rigorously determining how new data collection will improve predictive power.

This research will require a disciplinary background in data science, including experience with geospatial data; further background/training in the broader geosciences will be useful. The candidate will also need experience with databases (SQL, etc.) and Python scripting. In order of importance, we are looking for candidates with:

1.  Excellence in research as demonstrated through publications in international journals

2.  Demonstrated computer science expertise in data science programming, big data manipulation, and cloud computing.

3.  Having a dual data science / geoscience background in research or application

4.  Background in geostatistics and geophysics

5.  Having worked on real data with real practical impact

Summary of research project

Data-driven exploration relies on geophysical, geochemical, and geological data to develop mineral potential maps. The historic focus of this type of work has been on developing such maps, rather than on how such maps could be used to guide exploration activities. Further, these prospectivity mapping exercises have a poor track record for a variety of reasons. First, uncertainty is seldom, if ever, properly carried through the assessment, and as such, it is not possible to quantify a map’s reliability.  Second, even if reliability were rigorously quantified, what should be done next?  Should more data be collected? If so, where? And, what kind?

In brown field exploration, when a significant resource has already been identified through drilling and high-resolution data collection, the key question becomes: can the resource be cost-effectively enlarge by near field exploration? Again, what type of data should be collected and where? Is the statistically-valid expected value of the incremental resource greater than the expected cost of collecting the data? Is that right trade off?

Send a CV, statement and a list of three referees to jcaers@stanford.edu