Machine Learning

Progress

• Most of the project's outputs were released on November 13 (see: HLG-MOS Machine Learning Project). Congratulations and thank you to all who have contributed in any manner.

• New information added since the release are highlighted on the New information page. It currently includes:

  • In addition to the pilot studies, other developments or uses of ML were presented to project members. They are shared to further highlight the interest in advancing the use of ML, please see Other applications of Machine Learning. It includes the poster sessions presented at the 2020-10-(13 & 14) Virtual sessions.

  • Several initiatives taken by statistical offices to further facilitate and accelerate the advancement of in the use of machine learning and data science, more generally, are provided at Initiatives to accelerate the integration of machine learning solutions. (added on 2020-11-27)

• Talks delivered by project participants at organized events, e.g. conferences, workshops, have been added at the bottom of the  WP1 - Pilot Studies page.

• The project outputs were shared at the ML webinar held on November 16 and 17 (see agenda and presentations at  HLG-MOS ML Project webinar). The webinar was attended by 203 people from 33 countries, and 60 national and international organisations. Congratulations and thank to all who participated at the webinar.

• The project was presented at the 2020 Workshop on the Modernisation of Official Statistics (presentation and spear notes at Presentation to HLG-MOS workshop November 2020 v2.pptx). 

• Some of the work of the ML project was presented at the BigSurv20 event. (see bottom of WP1 - Pilot Studies page for the videos).

• WP leaders and InKyung Choi delivered a panel presentation at a meeting of the Israel Central Bureau of Statistics Data Science and Machine Learning Forum on December 15.

Input Privacy-preserving Techniques 

Progress

The work of the project was presented in the November workshop. One of the feedback points was that the work of the project is very technical in nature. At the same time it was indicated that there may be overlap with the project of the synthetic data.

The first five use cases have been collected, documented according to the established classification and included in the list of use cases. The classified use cases are discussed in the project meeting. In the original WP1, the final step was to establish criteria for assessing maturity and readiness of techniques that wish to be tested in WP2 and WP3. During the discussion how to establish criteria, with the feedback from the November workshop on the technical nature of the project, we decided to chose for a new approach. The new approach is a generalization based on the used cases that have been submitted in order to draw a few reference scenarios. We want to provide these scenarios with benchmark data in Wp2 and then approach external technology partners to implement and test these scenarios. In the last work package the experiences are bundled and provided with recommendations. We believe that the chance of results will be greater by engaging external parties. At the same time, we are breaking free from technology (a recommendation from the November workshop), and it is easier to link up with the other initiatives. The proposal for the new work packages will be presented to the HLG-MOS November-December meeting.

Next Steps

New participants will bring new use-cases that will be added to the inventory and possibly generalized. A subset of most prominent use-cases will be identified. Reference scenarios that are relevant for different types of source data and/or different NSOs will be drafted and prioritized for follow-up action in WP2.

Risks and Issues