Update to UNECE HLG MOS EB meeting  - November / December 2021 

Overview 

The Group ended a successful year by presenting highlights of its research to the global official statistics community at its annual webinar on 19 November on the sidelines of the UNECE HLG MOS workshop. Over 279 attendees from 48 different countries for presentations of the group’s research highlights and discussion with national data science leads of the way forward for ML in official statistics involving data science leads from Canada, Sweden and the UK. A post-event survey showed a 96% satisfaction rate and that 74% of attendees were now interested in taking part in the ML Group activities next year 

Research and knowledge sharing. In the workstreams activity leads have been writing up their research results and preparing final reports. These are being published on the new public page for the 2021 activities here. The last all group meeting held at the end of October some interesting presentations including Supervised Text Classification with Leveled Homomorphic Encryption from Saeid Molladavoudi, Statistics Canada, and MLOps in the Australian Taxation Office, from James Beck, Australian Taxation Office. A discussion panel on model retraining saw representatives of different NSOs (US, UK, Norway, Canada) present their perspectives on the challenges and possible approaches to model retraining. 

Evaluation and strategy. The Coordination Team ran a survey evaluating the activities of the group in 2021 and seeking to understand priorities for next year. Feedback showed high levels of satisfaction with the group’s activities and the benefits of connecting with others to exchanging ideas, information and experience when many NSOs are still in the early stages of their machine learning journey. As one respondent commented “The project has created the opportunity to spread the knowledge of ML among other members of the institution, benefiting the institution as a whole.” The survey flagged interest in more training activities such as the Coffee and Coding sessions, and looking at the challenges of moving ML models into production. It also highlighted some obstacles to the group’s plans to expand next year, notably that many members found it difficult to contribute to activities in a sustained way due to the pressures of their day-to-day work commitments.  

ML 2022: The Team has developed detailed plans to deliver a bolder, bigger and more collaborative project next year. It will launch the next stage with a series of training sessions, presentations and workshops at the UN CEBD / UAE Big Data conference at Expo 202 in Dubai in January 2022. We will issue a call for proposals for activities next week and would encourage NSOs to promote the call and put forward their ideas for research and knowledge sharing activities they could contribute.