Workshop on Machine Learning on Images of Natural History Collections


An interdisciplinary team of 19 researchers from seven European countries discussed machine learning solutions to extract information contained in the world's natural history collections. Machine learning will allow us to collect data faster, but may also allow the collection of new types of data. The workshop was framed by the Action COST Mobilise (Mobilising Data, Experts and Policies in Scientific Collections) and took place in Coimbra, between 8-10 September 2021 and was hosted by the Herbarium of the University of Coimbra, Department of Life Sciences.

On the first day there were 11 presentations, four remotely, on use cases and practical applications. There was a wide range of topics from workflows and accelerating collection digitisation, to metadata extraction, training datasets and trait recognition. On the second day there was a brainstorm for a SWOT analysis of a machine learning strategy. New ideas gathered during the workshop were incorporated into the skeleton of a paper on an infrastructure to support image analysis across collections.

After a long period of lockdown, it was an enriching experience to be able to see each other eye to eye, exchange ideas, and share laughter at the social events.