Differentially Private Distributed Aggregation via Iterative Message Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current protocols for secure and distributed aggregation in the shuffled model face challenges in scalability and privacy, as they often result in communication or error that grows polynomially with the number of users.
Innovation Solution
The proposed method involves a computer-implemented process where private data is encoded into a series of messages, with at least one message value being randomly selected, to approximate the private value. These messages are then aggregated with additional messages from other private values, and the process is designed to ensure differential privacy and reduce communication costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current protocols for secure aggregation are used, then privacy is preserved, but communication cost and error grow polynomially with the number of users
Solution Approach 1:
The patent segments the aggregation process into multiple rounds where users iteratively update their messages. Instead of transmitting all private data at once, users send incremental updates that collectively reveal the sum while maintaining individual privacy. This segmentation reduces the polynomial communication overhead by distributing the information flow across multiple smaller transactions.
Solution Approach 2:
The patent implements periodic action through iterative message updating where users repeatedly send refined approximations of their private values. Each iteration brings the aggregate sum closer to the true value while maintaining privacy guarantees. This periodic refinement process achieves accurate aggregation with sub-linear communication costs compared to single-shot protocols.
2Reliability
If current protocols for secure aggregation are used, then privacy is preserved, but computational resources and error increase polynomially with the number of users
Solution Approach 1:
The patent segments the computational burden by dividing the aggregation task into multiple simple iterative steps. Each user performs lightweight local computations to generate message updates rather than executing complex cryptographic protocols. This segmentation reduces device complexity from polynomial to near-linear scaling with the number of users.
Solution Approach 2:
The patent uses copying by having users transmit simplified message representations rather than their complete private data. Each user maintains a copy of their private value locally and only transmits processed message copies that contribute to the aggregate sum. This copying approach minimizes computational overhead and memory requirements on user devices.
3Quantity of substance
If approximate aggregation is used to reduce communication, then communication cost decreases, but measurement precision of the aggregate sum deteriorates
Solution Approach 1:
The patent applies periodic action through iterative refinement where users repeatedly send updated messages that progressively improve the accuracy of the aggregate sum. Each iteration reduces the approximation error while maintaining communication efficiency. This periodic refinement achieves high measurement precision with sub-linear communication costs by continuously converging toward the true sum.
Solution Approach 2:
The patent implements feedback mechanisms where the aggregate sum from previous iterations is used to guide subsequent message updates. Users adjust their messages based on the current approximation quality, sending corrections that reduce error in the next iteration. This feedback loop ensures that communication resources are efficiently allocated to achieve the desired precision level.
Data Source
AI summary
An encoding process performed by a computing device (e.g., a user's private device) can include obtaining private data that includes a private value. According to an aspect of the present disclosure, the computing device can produce a plurality of messages that respectively comprise a plurality of message values, where a total sum of the plurality of message values approximates the private value, and where at least one of the plurality of message values is randomly selected. The device can provide the plurality of messages for aggregation with a plurality of additional messages respectively generated for a plurality of additional private values. For example, the messages can be transmitted to a shuffler model configured to shuffle the plurality of messages with the plurality of additional messages.