Shuffled Model Privacy Aggregation via Discrete Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current protocols for secure and differential privacy-preserving distributed aggregation in the shuffled model face challenges with high communication costs and error scaling polynomially with the number of users, making them inefficient for large-scale applications.
Innovation Solution
A computer-implemented method that introduces noise messages based on probabilistic distributions, such as Bernoulli and truncated discrete Laplacian distributions, to enable privacy-preserving aggregation, reducing communication costs to polylogarithmic and achieving constant error, by using a randomizer algorithm to produce multisets that are shuffled and aggregated without revealing individual user data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current protocols for secure aggregation in the shuffled model are used, then differential privacy guarantees are achieved, but the expected aggregation error or communication per user scales polynomially with the number of users
Solution Approach 1:
The patent changes the parameter of noise distribution from traditional Laplace to a discretization of continuous unimodal distribution (such as truncated discrete Laplacian), which maintains privacy guarantees while reducing communication overhead. This parameter change enables the system to achieve both differential privacy and improved communication efficiency by optimizing the noise addition mechanism.
Solution Approach 2:
The patent segments the noise distribution into discrete values that can be efficiently communicated. By discretizing the continuous unimodal distribution into a finite set of possible noise values, the system reduces the communication burden while maintaining the privacy guarantees required by differential privacy.
2Reliability
If noise messages based on continuous unimodal distribution are used, then privacy preservation is improved, but communication costs increase due to the complexity of representing continuous distributions
Solution Approach 1:
The patent uses discrete noise values that can be represented with limited precision, effectively creating simplified versions of the continuous noise distribution. These discrete noise messages serve as approximate substitutes that maintain privacy guarantees while being much easier to communicate and process, reducing the complexity of the communication protocol.
Solution Approach 2:
The patent transforms the continuous distribution parameter space into a discrete parameter space by selecting a finite set of noise values from the continuous unimodal distribution. This parameter discretization maintains the essential privacy-preserving properties while enabling efficient communication with reduced protocol complexity.
Data Source
AI summary
An encoding method for enabling privacy-preserving aggregation of private data can include obtaining private data including a private value, determining a probabilistic status defining one of a first condition and a second condition, producing a multiset including a plurality of multiset values, and providing the multiset for aggregation with a plurality of additional multisets respectively generated for a plurality of additional private values. In response to the probabilistic status having the first condition, the plurality of multiset values is based at least in part on the private value, and in response to the probabilistic status having the second condition, the plurality of multiset values is a noise message. The noise message is produced based at least in part on a noise distribution that comprises a discretization of a continuous unimodal distribution supported on a range from zero to a number of multiset values included in the plurality of multiset values.