US 12,072,987 B1
Private counting from anonymous messages: near-optimal accuracy with vanishing communication overhead
Pasin Manurangsi, Mountain View, CA (US); Badih Ghazi, San Jose, CA (US); Shanmugasundaram Ravikumar, Piedmont, CA (US); Rasmus Pagh, Hjarup (SE); and Amer Sinha, San Francisco, CA (US)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Apr. 12, 2021, as Appl. No. 17/227,532.
Claims priority of provisional application 63/008,332, filed on Apr. 10, 2020.
Int. Cl. G06F 21/60 (2013.01); G06N 20/00 (2019.01)
CPC G06F 21/60 (2013.01) [G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method to enable privacy-preserving aggregation of data, the method comprising:
obtaining, by one or more computing devices, private data comprising a private value associated with a client device;
sampling, by the one or more computing devices, a first random value from a first sampling distribution associated with the client device, wherein the first sampling distribution is a base distribution of a first infinitely divisible distribution;
sampling, by the one or more computing devices, a second random value from a second sampling distribution, wherein the second sampling distribution is a base distribution of a second infinitely divisible distribution;
sampling, by the one or more computing devices, a third random value from a third sampling distribution, wherein the third sampling distribution is a base distribution of a third infinitely divisible distribution;
generating, by the one or more computing devices, a multiset of messages for the client device based at least in part on the private value, the first random value, the second random value, and the third random value, wherein the multiset of messages comprises a first number of increment messages equal to a sum of the private value, the first random value, and the third random value, and wherein the multiset of messages comprises a second number of decrement messages equal to a sum of the second random value and the third random value; and
transmitting, by the one or more computing devices over a network, the multiset of messages for differentially private aggregation with a plurality of additional multisets of messages respectively generated for a plurality of additional private values respectively associated with a plurality of additional client devices;
wherein the first infinitely divisible distribution is shared among the plurality of additional client devices.