To set the scene for the workshop ahead, we start with a brief overview of Privacy Enhancing Technologies and their impact on Official Statistics.

In this session, we will get hands-on with privacy-enhancing technology (PETs). This 2-part session will demonstrate various approaches to privacy-enhancing technologies including the use of differential privacy, secure enclaves and multiparty computation.

This session will introduce the work of the UN PPT Task Team. This will be followed by four presentations of use-cases that leverage PETs and 10 minutes time for questions at the end before proceeding to the Round table.

The presentation describes a logical framework to define in a detailed way the scenarios of input privacy that are relevant for Official Statistics. The framework is a result of the Input Privacy Preserving Techniques project, which is an on-going project under the supervision of the UNECE High-level Group on Modernization of Official Statistics (HLG-MOS).

We report on our recent project using PETs to reproduce a commonplace statistical application that requires inter-agency sharing of sensitive personal data within the US Department of Education. The prototype reproduces a portion of the annual US 2015–16 National Postsecondary Student Aid Study (NPSAS:16), including statistics on average federal Title IV financial aid received by undergraduate students. Our approach relies on computing the necessary statistics while the data remains encrypted. Our experiments produce accurate results, provide strong cryptographic security, and incur resource costs reasonable in comparison to costs of the typical (non-privacy preserving) methods used to produce the same statistics.

In this pilot project, we use distributed Machine Learning (ML) algorithms and other privacy preserving techniques to build a simulation environment for collaborative ML training and inference tasks among multiple National Statistical Organizations (NSO) in a low trust environment while allowing them to mutually benefit from the outcome. The set up involves ML human activity recognition using accelerometer and sensor data collected from personal and smart devices. The project is part of the Input Privacy Preserving Project within the High-Level Group for Modernization of Official Statistics that is supported by UNECE.

Privacy-preserving computation technologies promise to enable "processing without disclosing" of personal or otherwise confidential data. Such technologies become increasingly appealing in scenarios where multiple organisations are involved in the computation process and when input data are held by different entities.

However, while these technologies are maturing quickly and commercial products already exist, the rate of adoption in production environment is still slow, especially in the public sector. The round-table will gather representatives from producers/developers and potential users/adopters of privacy-preserving computation technologies, to discuss what are the challenges and obstacles towards a more widespread adoption of these technologies in data science and statistical applications.