The Promises and Perils of Trusted Research Environments
Gavidia-Calderon, Carlos; Rangel Smith, Camila
- Publisher
- Zenodo
- Language
- en
Abstract
Data Science and machine learning are powered by data, and there is plenty of data today: many human activities are powered by software that generates large quantities of it.While large datasets are constantly generated, putting them in the hands of the engineers and researchers, who can better exploit them, is not always easy. In domains like finance or health, data sensitivity and user privacy require a limited and controlled distribution of data assets.Arguably, the gold standard for solving this data availability problem is placing datasets within trusted research environments (TREs). TREs are closed computational environments equipped with technical and process controls that enable secure data access to vetted researchers. The Research Engineering Team at the Turing uses a TRE software, our in-house developed Data Safe Haven, alongside our trusted research process to support academic projects of diverse sizes and domains.We learned that while software and processes are pivotal to the success of a TRE initiative, there are other external critical factors that, if ignored, can derail an academic project. For example, placing data in a TRE comes with regulatory and legal obligations that need to be addressed way before onboarding the researchers into the system. Or dealing with researcher frustration when facing the necessary security controls imposed by platform and process. In this talk, we will rely on years of experience to describe these external factors, their influence on the success of a TRE initiative, and how we approach them at Turing.We will discuss:Enabling data-driven research with TREs.Software and process: The essential factors of every TRE initiative.External critical factors: Legal, financial, data management, usability, and others.The Turing way of dealing with these factors, to ensure a successful TRE initiative.Acknowledgements This work is supported by the UK Engineering and Physical Sciences Research Council (EPSRC) Grant EP/Z531297/1.A recording of this session is available on YouTube: https://youtu.be/8yqIg17DU9Y
Full text
The Promises and Perils of Trusted Research Environments Carlos Gavidia-Calderon and Camila Rangel-Smith https://commons.wikimedia.org/wiki/File:The_Lover_and_Dame_Oyseuse_outside_a_walled_garden__Roman_de_la_Rose_(c.1490-1500),_f.12v_-_BL_Harley_MS_4425_(cropped).jpg
Agenda 1. Sensitive Data and Trusted Research Environments. 2. The Promises and Perils. –Secure Data Access. –Regulatory Compliance. –Infrastructure Management. –Scalable Data Analysis. 3. Conclusion. From:https://commons.wikimedia.org/wiki/File:Treasure-Island-map.jpg
Sensitive Data ”Research organisations working with personal data, including data related to health and social care, must ensure that data protection legislation is adhered to when accessing and processing data. …These legal basis are met through accessing sensitive data within safe environments…” From: “Common governance model: a way to avoid data segregation between existing trusted research environment” by Torabi et al. (2023)
Trusted Research Environments (TRE) “Secure data store that hosts … sensitive data for research... Access is restricted, monitored, and often based on clear data use agreements.” From: “Local Data Spaces: Leveraging trusted research environments for secure location-based policy research in the age of coronavirus disease-2019” by Macdonald et al. (2023)
Agenda 1. Sensitive Data and Trusted Research Environments. 2. The Promises and Perils. –Secure Data Access. –Regulatory Compliance. –Infrastructure Management. –Scalable Data Analysis. 3. Conclusion. From: https://www.rawpixel.com/image/13968297/image-cartoon-face-people
Technical Controls on the Turing DSH
The Peril of Disengagement ”We’ve always recognized that our security proposals come with certain costs in terms of usability. Traditionally, that’s the compromise we make to be secure.” From: “Security Usability” by Gutmann and Grigg (2005). Source: https://www.lookandlearn.com/history-images/YW019743V/A-shaving-machine-powered-by-steam
Mitigating with Incremental Security Controls Technical Control Tier 1 Tier 2 Tier 3 Inbound connections Any IP address. IP addresses from institutional networks. IP addresses from restricted networks. Outbound connections Internet access permitted. Internet access blocked. Internet access blocked. User devices Own devices. Own devices. Managed devices. Data transfer from user devices. Copy -and-paste enabled. Copy -and-paste disabled. Copy -and-paste disabled. Python/R package availability Any package repository. Only CRAN or PyPi . Pre -agreed packages from CRAN or PyPi.
Higher Tier Personal data, even it’s not obvious. Special category data, like race, politics or religion. Commercially-sensitive data, like clients or employees. Lower Tier Data already available without restrictions, even if it’s medical. Rigorously pseudonymized data, even if it’s medical. From: https://alan-turing-institute.github.io/trusted-research/guidance/determine_security_tier.html
Agenda 1. Sensitive Data and Trusted Research Environments. 2. The Promises and Perils. –Secure Data Access. –Regulatory Compliance. –Infrastructure Management. –Scalable Data Analysis. 3. Conclusion. From: https://commons.wikimedia.org/wiki/File:Philippe_de_champaigne,_mos%C3%A8_presenta_le_tavole_della_legge,_1648_ca._01.jpg
Managing Infrastructure “In some cases, owning is cost prohibitive due to people, hardware, and software costs. In contrast, the cloud provides a more pay-as-you-go feel, with less need to manage infrastructure and pay for individual software licenses.” From: ”Genomics in the Azure Cloud” by Ford (2022).
The Perils of Managing Everything Else 1. Developing new features and fixing bugs. 2. Installing research specific software. 3. Evaluating data ingress and egress. 4. Running the production environment. Source: https://commons.wikimedia.org/wiki/File:Aerger.jpg
Mitigating with a TRE Operations Team Day-to-day user support. Ingress and egress management. Running the production environment. Feature development.
Agenda 1. Sensitive Data and Trusted Research Environments. 2. The Promises and Perils. –Secure Data Access. –Regulatory Compliance. –Infrastructure Management. –Scalable Data Analysis. 3. Conclusion. From: From: https://.wikimedia.org/wiki/File:LZB_in_Flensburg_-_Niederdeutsche_Lutherbibel_von_1574-1580,_Bild_11B. commonsJPG
Scaling Data Analysis “Azure provides various highperformance computing resources, such as H-series virtual machines for memory bound applications, Nseries virtual machines for CUDA/OpenCL based application From: ”Exam Ref AZ-304: Microsoft Azure Architect Design” by Agrawal et al. (2022).
Scaling Compute and Storage
The Peril of Bankruptcy Using a virtual machine with 32 vCPUs and 64 GB of RAM for 6 months costs around 6000 GBP. With that money, you can buy a similar workstation. Source: https://www.flickr.com/photos/mazanto/14248772486
Mitigating with Cost Management
Conclusion Information security, or cyber security, is not a digital problem only. Humans have been termed “the weakest link” … From: “Socio-Technical Security Metrics” by Golman et al. (2014) From:https://commons.wikimedia.org/wiki/File:Stamp_of_Peru_-_1969_-_Colnect_386666_-_Peasant_breaking_Chains.jpeg
