Startups leveraging machine learning to improve exploration targeting
Can we use machine learning (ML) to improve our discovery rate, to help us identify and define targets faster and better? Should the question be how rather than can? Regardless of which side of the fence you sit on, it is interesting to see, particularly in the startup space, who is working on what and which areas they are focusing on. I am also intentionally leaving out the term Artificial/Augmented Intelligence (AI) from this discussion. As a personal preference, I think the term is too broad for this discussion and can lead to unfounded expectations on what can reasonably be achieved.
As we move through the hype curve of ML (in terms of recent applications), I believe we now are able to have educated discussions on how approaches can compliment our workflows. Here, I provide an overview of the startups and smaller explorers and consultancies working on pushing the boundaries in the use of ML, specifically in exploration targeting. It has been my pleasure to learn from a lot of these founders directly, and I would encourage you to reach out and do the same.
Founded in 2018, Brisbane-based Orefox is the commercialisation of research undertaken by Quantum Geology. The Orefox system uses artificial intelligence in the forms of machine learning and cognitive analysis for rapidly generating quality exploration targets for minerals. The toolkit compares the data of thousands of known deposits against the data of areas to be explored, and looks for obvious correlations in data, hidden patterns, clusters and relationships, across massive data sets.
Lead by dynamic duo Warwick Anderson and Sheree Burdinat, both exploration geologists by background, Orefox really focus on bringing geologists and their clients along on their journey. They make sure the ML results are meaningful and useable, and ensure their analytics are accessible to anyone regardless of their background. The Orefox team can always be found supporting the local startup and geological communities in Queensland, often speaking at events, and promoting the use of ML and open data for exploration.
EarthAI was founded in 2016 by Roman Teslyuk after he realised through his PhD in geochemistry, the huge opportunity to improve exploration with more efficient use of data, particularly publicly available data. EarthAI graduated from the Cicada Innovations incubator in Sydney and received seed funding from two Australian venture capital firms: Airtree and Blackbird Ventures.
With a young and energetic team with a range of experience and backgrounds, EarthAI is really not like any exploration company that has come before them. Breaking away from traditional methods, they are setting new industry standards for efficiency, diversity and innovation. Their first innovation was a machine learning algorithm that processes millions of data points from all over the world to predict mineral enriched areas, even in unexplored remote greenfield areas of Australia. This has enabled fast, low impact field work on precisely targeted exploration areas.
More recently, EarthAI are bringing different tech into their process, including a certified drone pilot team to conduct high-precision magnetic surveys of prospects and an automated diamond drilling rig, which is currently drilling their Northern Territory projects. It is fair to say the EARTH AI team are bullish in the applications of tech providing no limit in improving mineral exploration efficiency.
Based in Melbourne, Solve Geosolutions is a team of geoscientists, with mathematical and programming backgrounds, who use data science to solve geological problems. Solve think data science is something more geologists should have in their toolkit and have a strong focus on educating and upskilling their clients.
In the exploration space Solve Geosolutions specialises in the application of ML based prospectivity and target vectoring at regional to deposit scale. In the last few years they have specialised in image analysis, particularly focused on Corescan hyperspectral imagery, core photography, remote sensing and geophysics. These datasets lend themselves to the application of Deep Learning techniques which are evolving very quickly and are adding significant value to these often underutilised datasets. In the mine environment Solve also commonly works on geometallurgical and processing optimisation problems such as building geological classification models and trying to predict the behaviours of ore during processing.
Koan Analytics expertise is in applied mathematics, machine learning and computer graphics. As math and IT experts, they first came across the opportunities in the mining industry through the Unearthed Toronto hackathon in 2017 with Barrick, which they won. Then working in pharmaceuticals, following the hackathon in 2018, Koan changed direction and partnered with Barrick to develop a data aggregation, predictive analytics and visualisation solution designed to capitalise on the vast quantity of static and unstructured data.
Their solution geo-spatially integrates unstructured data (maps, documents, images, tables, etc.) and provides customised algorithms to perform predictive analytics across a large corpus of information (proprietary, public and third party). The analytics interface graphically enables users to interact with, and make meaning behind, an enormous volume of information. The solution is intended to complement not compete with or replace other GIS or modelling systems, and is architected to pass structured data back into standard open protocol systems (Arc-GIS, Voyager, Leapfrog, etc.). Koan retained ownership of all intellectual property and have completed an MVP solution. Now, they are beginning the task of generating industry awareness.
KoBold is the first exploration startup drawing significant interest and investment from major players in Silicon Valley. Based in Berkley, California, and founded in 2018, the company has raised money from venture capital firm Andreessen Horowitz and Breakthrough Energy Ventures, a fund backed by Bill Gates and other tycoons including Jeff Bezos, Ray Dalio and Michael Bloomberg.
With a focus on a search for ethically sourced cobalt as a key need for future green technologies, KoBold is certainly able to draw interest from outside of the traditional investment community. This is an interesting shift in the type of investor typically interested in exploration. Is the driver to source more ethical materials, and/or the belief that the application of ML is providing significantly more reward, and less risk for investors? KoBold’s team has a balance of data science PhDs and highly experienced geologists. The first public release around the projects they are working on included an acquisition of a large amount of ground in Saskatoon.
With a vision to massively disrupt the exploration sector, Goldspot is probably the most well known company in this space. Particularly since they became the first ML focused exploration company to publicly list in February 2019. Personally, I am hoping that this listing provides more public awareness of where ML can currently be used effectively in exploration, and its role for the future.
Goldspot is particularly focused on changing the investment decision model by using data science to stake acreage, acquire projects and royalties, and invest in public vehicles to ultimately create a portfolio of assets with favorable reward to risk ratio. Goldspot started to rise into the spotlight when they won 2nd place in the $1million Integra Gold Rush competition and then went on to be finalists in Disrupt Mining. Prior to listing on the TSX, they received early support and funding from the likes of Hochschild Mining and Sprott Mining.
Born out of the Israeli high-tech community in 2013, Quantum Discovery (nee Quantum Pacific Exploration) is a privately funded exploration company focused, initially at least, on using ML to identify porphyry copper deposits in Chile. There is not much publicly available information on the outcomes of their approach, but arguably they are the most well funded and advanced company working in this space. Some not so quiet whispers indicate that QD have been part of some deals with major miners looking for large copper deposits.
Vancouver based Minerva Intelligence applies semantic AI to problems in geoscience, as well as data-driven machine learning solutions. Semantic AI is the process of generating meaning and context from natural language. Minerva can therefore use both human knowledge models and outputs from supervised and unsupervised learning techniques to provide explanations and advice to stakeholders and end users. Earlier this year, Minerva showcased thousands of new exploration targets in the Yukon. The targets were identified using Minerva’s Matcher® technology and are freely available to the public. I strongly suggest you check this out, there is no other site in the world, that I know of, where you have access to this type of tech for free. It is also crazy that the Yukon survey or government are not promoting this.
The semantic approach separates Minerva from many other players in this space and is revolutionising how businesses store, use and access documents. Minerva’s tools and techniques can identify mineralisation zones across the entire spectrum of assayed elements in drilling datasets. Economic geology has yet to see a modern classification of mineral deposits. Minerva is using Aristotelian principles, machine reasoners and world class mineral deposit expertise to create a new classification system accessible to AI products.
Based in a global hub for AI, Quebec, Albert Mining is a project generator which uses AI and data mining. Albert uses its proprietary CARDS (Computer Aided Resources Detection System) to help mineral exploration professionals identify areas with a high statistical probability of similarity to known areas of mineralisation. In combination with modern exploration techniques, CARDS is a useful tool to save both money and time by limiting target areas for exploration.
Toronto based Exiro Minerals is an exploration company with an interesting business model and application for machine learning. Exiro is taking advantage of the large amount of historical paper records, which are largely overlooked, being too difficult to use and process. Exiro creates digital copies of this data, and in exchange for a copy themselves, takes on the cost of the work. This provides value back to the data owner, but also allows Exiro to build up a large database of proprietary information to be used for exploration.
While not a startup per se, Descartes Labs are doing some interesting work in facilitating the use of ML for exploration targeting. DL is a commercial spin-out of Los Alamos National Laboratory that developed a unique data fusion and modeling platform. The platform is a cloud-based data refinery that continually amasses geospatial and geoscience data to create a digital model of the Earth that extends into the subsurface. The platform enables disparate data to be rapidly cleaned, conditioned, normalised, prepped for analysis, and stored in a single modeling environment.
The platform was originally designed to fuse disparate satellite, aerial and drone data collected across spectrum, modalities and time to power machine learning models for the US defense department , but has since evolved to include over 40 different types of data, including geoscience data. Users can load proprietary hyperspectral aerial, gravity and magnetics, seismic, geochemistry, well logs and production history to create clean geoscience data library. These data sets can also be merged with planetary scale satellite data archives to develop prospectivity workflows.
Azimut Exploration is a publicly listed Canadian explorer that has been taking advantage of using advanced analytics on large datasets since 2003. President Jean-Marc Lulin presented some results from their country-wide ‘big-data’ approach at PDAC in 2016.
Have I missed someone out? I would love to hear about your company or companies that you are working with in this space. Drop me an email: holly@unearthed.solutions.
