Contagious diseases, such as COVID-19, have the potential to spread rapidly through person-to-person contact. The impact of such diseases has only been amplified by increasing levels of globalization in recent decades. Reducing the spread of a highly-infectious disease like COVID-19 requires a two-pronged approach: 1) identifying and isolating (potentially) infected individuals through methods such as contact tracing, and 2) encouraging behaviors that help prevent transmission in the first place (e.g., vaccinations, “wash-wear-wait”). Without optimal vaccination coverage, safely operating in a pandemic requires adherence to physical distancing (“wait”). While physical distancing can be (and has been) enforced via mandatory lockdowns, these are not long term solutions and they negatively impact the economy, education, and other aspects of society.
There are a number of well established methods for supporting reactive measures, with a significant amount of recent work on approaches for digital contact tracing; however, there has been far less attention paid to preventative measures. While reactionary methods can provide some insight on adherence of preventative measures, such methods leverage infection data which introduces a longer feedback loop; some diseases, such as COVID-19, have a significant delay between transmission and appearance of symptoms (and thus a positive diagnosis). While it may be possible to enable smart, data-driven policies for physical distancing through infection metrics, understanding how and where adherence to the policies is failing requires more fine-grained information. Finally, contract tracing apps provide individuals with a binary view of their risk: either they have come in contact with an infected individual or they have not. Providing users with more actionable information can enable them to make smarter decisions in terms of how they protect themselves and others.
The goal of our work is to support reduction in disease transmission by providing actionable information to enable both individuals and decision makers to be more proactive. Our insight is that measuring, summarizing, and aggregating “contact events” is the best way to provide this information. Contact events capture physical distancing between pairs of individuals irrespective of their current infection status. Given how contagious diseases spread, namely through close physical interactions between individuals, contact events serve as a metric that is directly related to the spread of the disease. By aggregating contact events across all users within a community (e.g., a university), we can generate useful statistics at both the community- and individual-level. For example, community decision makers can use these aggregate statistics to identify hotspots that may require further adjusting physical distancing policies, or for better allocation of sanitation resources. Additionally, users will be able to understand their own risk levels while also gaining valuable insight into the community as a whole.
Poirot is our system for privacy-preserving contact aggregation that meets a number of system and security requirements. First, the system needs to provide utility to both individuals and decision makers, which requires the system to compute accurate results to queries that yield actionable information. In addition, the system also needs to be performant: the client-side app needs to continuously collect contact events while running on battery-powered devices, while the server-side needs to scale to aggregating results from thousands to millions of users. In terms of security, the system needs to provide strong privacy guarantees for users such that no sensitive information is directly accessible to any principal other than the user themselves. We address a number of challenges that relate to the fundamental tension between the utility, performance, and privacy goals by leveraging strong cryptographic and privacy techniques such as multi-party computation (MPC), differential privacy (DP), and blind signatures.
Poirot: Private Contact Summary Aggregation
Yanping Zhang, Chenghong Wang, David Pujol, Johes Bater, Matthew Lentz, Ashwin Machanavajjhala, Kartik Nayak, Lavanya Vasudevan, Jun Yang
SenSys 2020 (Conference on Embedded Networked Sensor Systems)
Poirot: Private Contact Summary Aggregation
Chenghong Wang, David Pujol, Yanping Zhang, Johes Bater, Matthew Lentz, Ashwin Machanavajjhala, Kartik Nayak, Lavanya Vasudevan, Jun Yang
NeurIPS PPML 2020 (NeurIPS Workshop on Privacy-Preserving Machine Learning)
Please contact Prof. Nayak for further information.