A Platform for Privacy-Preserving Statistical Analysis of Smartphone Data
Almost everyone carries a smartphone today. Smartphones can be used to collect data to help control pandemics. However, such data collection is obviously at odds with concerns about user privacy. Consequently, it is essential to examine this utility-privacy trade-off and develop techniques to store, share and use smartphone data in a privacy-preserving manner. A number of recent proposals address this trade-off for the limited use case of contact tracing, where smartphone encounter information is used to track contacts of confirmed infected people. However, these proposals do not consider the equally important use of smartphone data for scientific analysis of disease parameters, e.g., the efficacy of social distancing measures, the disease’s basic reproduction number (R0) or geographic regions of excessive disease activity. Such scientific analysis informs policy decisions for pandemic control and is, therefore, extremely important for society.
In our work, we are designing an open platform that allows data uploaders (smartphone owners) and data users (scientific organizations) to interact with each other, balancing utility for users with privacy for uploaders. Specifically, our platform: (a) allows smartphone users to voluntarily upload one or more kinds of disease-relevant data from their smartphones in a secure and privacy-preserving manner, (b) stores this data securely against a very strong threat model, and (c) allows multiple scientific organizations to issue privacy-preserving aggregate queries, limiting access to the minimum amount of data needed for each query, and subject to user preferences. In doing so, we are combining several ideas ranging from classic encryption, key management and access control to modern privacy-preserving techniques like CPU-backed trusted execution environments (TEEs), homomorphic encryption and differential privacy.
Our current design supports many kinds of smartphone data such as bluetooth encounters with nearby phones, location and smart-sensor readings. We plan to support queries that feed into statistical and stochastic models of epidemic spread, measurement of effectiveness of control measures like social distancing, and estimation of fundamental disease parameters like the basic reproduction number.
In our work, we are designing an open platform that allows data uploaders (smartphone owners) and data users (scientific organizations) to interact with each other, balancing utility for users with privacy for uploaders. Specifically, our platform: (a) allows smartphone users to voluntarily upload one or more kinds of disease-relevant data from their smartphones in a secure and privacy-preserving manner, (b) stores this data securely against a very strong threat model, and (c) allows multiple scientific organizations to issue privacy-preserving aggregate queries, limiting access to the minimum amount of data needed for each query, and subject to user preferences. In doing so, we are combining several ideas ranging from classic encryption, key management and access control to modern privacy-preserving techniques like CPU-backed trusted execution environments (TEEs), homomorphic encryption and differential privacy.
Our current design supports many kinds of smartphone data such as bluetooth encounters with nearby phones, location and smart-sensor readings. We plan to support queries that feed into statistical and stochastic models of epidemic spread, measurement of effectiveness of control measures like social distancing, and estimation of fundamental disease parameters like the basic reproduction number.
Important Links
Coming soon