Work Streams and Activities
WS1 – Pilot studies: from Idea to Valid solutions
WS2 – From Valid Solution to Production
Activities
2.1. Automated production tool to code IMF member state time series data using ML algorithms, International Monetary Fund - Ayoub Mharzi
2.2. Deployment of a Data Lake architecture to put into production data science projects, INEGI Mexico - Abel Coronado
2.3. Design and assess a whole workflow to enable Natural Language Processing and Machine Learning methodologies to be integrated into a continuous production process, INEGI Mexico - Jael Pérez; Alejandro Ruiz
2.4. A technical platform that supports the whole machine learning process and thereby ensures the quality of that process and in the end contribute the overall quality of the statistical output, Statistics Sweden - Alexander Thorell
WS3 – Data Ethics and Governance
3.1 The establishment of a set of ethical principles to provide a clear framework to enable ethical use of Machine Learning for research and statistics, UK Statistics Authority - Lily O'Flynn; Simon Whitworth
WS4 – On The Quality of Training Data
4.1 Identifying the circumstances under which an ML model should be retrained in order to maintain the predictive power and quality of the model, Statistics Finland - Riitta Piela; Rok Platinovsek
WS5 – On The Quality Framework for Statistical Algorithms
5.1 Explore dimensions of QF4SA in a consolidated project to analyse an output based on a set of standard metrics and procedures
INEGI Mexico - Jose Jimenez; Alejandro Ruiz
ML2021 Presentations
| Date | Speaker | Presentation |
|---|---|---|
| 22 March | Sigrid van Hoek (Statistics Netherlands) | Fair algorithms project (presentation slides) |
| Lily O'Flynn and Simon Whitworth (UK ONS) | UK SA Data Ethics (presentation slides) | |
| 23 Feb | Riitta Piela and Rok Platinovsek (Statistics Finland) | Best practices in maintaining the quality of data in ML developments (presentation slides) |
| Casper Eriksen (Danish Business Authority) | Multilingual Classification of Economic Activities (presentation slides) | |
| Michael Reusens (Statistics Flanders) | WS1 Theme 5: Transferring Knowledge and Experience (presentation slides) |
How to join the Group ?
Do you have any ML topic you are interested in working together with peers? Do you have any issue (technical, strategic, organisational) you want to discuss with other NSOs? Contact ML2021 (ML2021 at ons dot gov dot uk), copying UNECE (choii at un dot org) if you want to join the Machine Learning 2021 Group!
ML 2021 Group Structure and Workstreams
If you are a member of the Global Network of Data Officers and Statisticians, you can follow us from the ML for Official Statistics group in the Network (see quick guide on how to join the Global Network).
Progress Update
January 2021
Just under under 100 members joined for the inaugural Machine Learning 2021 (ML2021) meeting on 29 January 2021.
The meeting ratified the groups governance structure and schedule for 2021. Submitted activity proposals were grouped into Workstreams in advance and confirmed with members.
In anticipation of the February meeting, members have signed up to workstreams and leads have been asked to confirm their output and objectives.
Resources
Machine learning competency in context of Big Data training and human resources (from UN Global Working Group on Big Data Task Team on Skills and it is relevant to the community)
Following the great interest in continuing the work of the HLG-MOS Machine Learning Project, UK Office of National Statistics (ONS) is launching the Machine Learning Group 2021. The new ML group will focus on developing and implementing ML for official statistics (see slides below for presentation and live survey conducted during the HLG-MOS ML Project webinar) Table 1: Work packages investigated by the ML 2019/2020 project (hyperlinks) and potential themes to be explored by the ML 2021 project (text in red). Moving from idea to valid solution (demonstration) Moving from valid solution to production (Operationalisation) Ensuring production robustness (Maintenance) Workstream 1: Support current studies towards production; welcome new studies in other processes (e.g. record linkage) and/or data sources (e.g. satellite data) Quality (accuracy, timeliness, efficiency, explainability and reproducibility) Good Training Data Skills/Competences Computing Infrastructure Interoperability / Business Process Ethics and Legal Security Workstream 2: Experiment with practices and methods on some dimensions of QF4SA (WP2); Workstream 4: How to get good training data, how to keep it up to date, when to relearn a model, what does 'good' mean, how to measure that? Workstream 5: What skills? How to learn? Where to find them? To be defined To be defined Workstream 6: Ethics handbook, regulations , etc. To be defined Organisation Sharing and Collaboration Initiatives to accelerate the integration of machine learning solutions Workstream 7 : Create/maintain a network of data science unit leaders; Update from ONS (December 15, 2020)Machine learning Group 2021
The Journey Supported by
Workstream 3: Review and improve the Framework Facilitated by
Workstream 8: Beyond 2021: How can we better prepare for next 2-5 years? What technology and data sources can we expect? What skills will we need?HLG-MOS ML Project webinar Progress status