Engagement in the group continues to be very high. Over 20 projects were proposed, either test ML techniques or the share knowledge and experience.

A sprint was held on May 13 (see objectives and agenda). It was attended by 12 participants from 9 countries, and hosted by the ONS.  Good progress on Work packages 1 and 2 was made:

• WP1 - Pilot Study on Coding and Classification: Building on US Bureau of Labor Statistics' successful implementation of machine learning to autocode injuries and illnesses, the code and practices has been shared and will be tested on different types of data in Serbia, Poland and Belgium. The types of data include data collected on the Web, notably to measure sentiment. Other applications may join this pilot study.
• WP1 - Pilot Study on Edit and Imputation:  A sub-group will investigate the potential of ML in automating the editing process and determine how the statistical foundations of ML and traditional techniques differ, where ML techniques can add value and in what context this value added can be the most beneficial. Examples from the UK and Italy will be used to conduct these investigations. Knowledge and experiences in imputation will be shared. Communication with the Statistical Data Editing expert group will be assured by members sitting on each group, as well as a member of the SDE on the ML project's distribution list.
• WP1 - Pilot Study on Imagery: The scope of this pilot study has not been finalized because some key project members were not in attendance.  At the sprint, the UK presented a successful application of ML to use street images to produce relevant statistics at a relatively local scale (two cities). It is relevant to the needs and interests of Belgium and Netherlands. We discussed an idea to produce a document to assist new users of imagery data by describing the processing pipeline (that calls on ML) in the use of such data, and its accompanying high-level ML-questions and aspects to consider, as well as proposed ML solutions/applications or places to find them. The pilot study may look at using satellite data to measure population (density, change). It will likely not look at satellite data for land use, as this topic, including its ML aspects, has been extensively covered a UN Task Team a couple of years ago.
• WP2 - Quality: It will identify quality indicators and performance indicators in two contexts: ML applied to carry out traditional processes on traditional data, and ML applied on non-traditional data sources. To do this it will consider quality features (definitions, dimensions, indicators) from the official statistics and ML communities. It will seek to apply some of these indications in one or two of the WP1 pilot studies. One of the challenges will be to remain focused on ML issues and not the broader issues that come with the various data sources. These are covered more extensively by the ESSnet Big Data II WPK on Methodology and Quality. Communication with this group will be assured by two members sitting on each group and our respective wiki spaces. 
• WP 3 - Lessons learned: This WP was not discussed directly. One of the intentions is to combine the experiences of organisations who have implemented or are close to implementing ML techniques with the experiences of organisations who will make advancements in implementing them through the WP1 pilot studies into lessons learned on topics such as: facilitators, obstacles, importance and costs of creating and maintaining learning datasets, role of manual operations, etc. 

The project manager and participants at the sprint wish to thank the ONS and the UN Data Science Campus for their hospitality, facilities and overall accompaniment throughout the three days. 

The documents produced at the sprint will be posted on the wiki and shared with all project members.