Image (above): Machine-generated ontology for structural geology.
Big Data, Analytics, Machine Learning, Deep Learning … the hype is everywhere! Yes, these ‘disruptive technologies’ (more hype) are already changing our world. And yes, they do have enormous potential to derive new insight from vast volumes of data. But the hype-ometer is already off the scale, and it’s safe to say that ‘not everyone’s expectations are going to be met’ (i.e. a lot of managers are going to throw a lot of money at this, without having a clue what they’re doing, and some are going to be very disappointed)! Our CEO (Richard) has had hands-on experience in artificial intelligence for nearly 30 years, and has already experienced the technology hype-cycle several times, so it was with direct interest (and a healthy dose of scepticism) that we joined the 2018 Janet Watson conference, “A Data Explosion: The Impact of Big Data in Geoscience”, at the Geological Society, London.
It was a great meeting! There was a very broad range of very interesting technical content, and this more than compensated for the occasional navel-gazing on just how to define ‘Big Data’, or the odd talk that was heavy on management-speak and light on substance. Most of the presentations are now available online – so if you missed the conference, you can catch-up here. Make sure you watch Jon Thurmond’s keynote talk (“When failure (a lot of failure) becomes an option”), which includes both an eloquent caution regarding hyped expectation vs. reality, and a great overview of the machine learning (ML) strategies that Jon and his co-workers have tried.
Most of the recent focus in Big Data has concentrated on various neural network approaches to analyse large multivariate datasets in order to understand non-linear relationships (using data that is numeric, or can readily be made so via digitisation). We’ve long been interested in these approaches for analysing satellite imagery and other geospatial datasets. However, we also have a long legacy in other aspects of artificial intelligence, particularly the use of natural language processing (NLP) to help analyse unstructured (non-numeric) text data, since very much of the knowledge value that we use as geologists resides in scientific articles, books, company reports and technical memos. In our daily work in GRL, we spend a lot of time searching for relevant information, so we’re very interested in looking at ways in which NLP can help us to find, sort, classify and access the knowledge content in these types of data. At the “Big Data” conference we summarised our current status in tackling this challenge in a presentation on “Accessing Knowledge in Geoscience Text using Natural Language Processing”.
In parallel with all the presentations on ML and NLP, the “Big Data” conference also highlighted recent progress in visualisation of large outcrop geology datasets, with multiple sessions showing live virtual fieldtrips. This is another theme that lies close to the core of GRL’s business, and we enjoyed presenting whistle-stop tours of some Zagros and Zechstein outcrop geology. More on this to follow soon …