And now for something entirely different!
Trusted Research Environments
Is social science reinventing bioinformatics?
Yes.
Working Paper Principal Consultant
Boring? Really?
97
Million
Americans
👉 Meta (but really any re-use of observational data)
Making Computational Social Science Possible...
By balancing privacy and utility
(with trusted research environments, obvi.)
Incredible opportunity for new science
Substance
- Frameworks
- Examples
- Brain Candy on Privacy-Enhancing Technology
Basic Frameworks
What am I talking about when I say Trusted Research Environment?
And why do we need to think about them at all?
Five Safes
- Safe people
- Safe projects
- Safe settings
- Safe data
- Safe outputs
Trusted Research Environments
- Computational analysis of data
- Within a contained system (e.g. not a data download portal)
- That supports governance and auditability
Auditability is most important
Solving Problems by demonstrating compliance
Legal Requirements
- Technical Safeguards
- Organizational Safeguards
Safeguards like TREs are more important...
when you can't use traditional protections like consent
Advanced Frameworks
Spectrums of TREs
Risk Calibration Along the Spectrum
Inputs and outputs matter
Real World Needs
Meta Content Library
The Digital Services Act:
An Opportunity
Privacy Enhancing Technologies
SWIFT Federated Learning
How do you share data ... without sharing any data?
Sharing network fraud signals
No one PET does it all
- Federated learning
- Differential privacy
- Fully homomorphic encryption
- Hardware trusted execution environments
Federated Learning
Differential privacy
On model weights
Fully Homomorphic Encryption
On global model updates
Just
20 percent slower!
TEEs
Hardware root of trust
What does ... delete mean?
What could we do with Trusted Research Environments that use this tech?
Reproducibility for API-based science
Just save it all. Forever...
Right?
Encapsulation, Libraries, and Meshes
.png)