Federated Random Forest Regression Secure Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current federated learning protocols lack methods for training random forest regressors in a horizontal federated learning setting while ensuring data privacy and enhancing model transparency, particularly in applications like job candidate screening where fairness and diversity are crucial.
Innovation Solution
A secure aggregation protocol is implemented to compute purity measures for decision trees across edge nodes, allowing for the construction of a common regression decision tree and random forest regressor in a private manner, using techniques like pairwise masking and Shamir secret sharing to maintain privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If federated learning protocols are used to train random forest regressors, then model transparency and fairness are improved, but data privacy is compromised
Solution Approach 1:
The patent segments the data into local datasets at different edge nodes while maintaining a global model structure. Each edge node trains local decision trees on its own data without sharing the actual data, only sharing aggregated purity measures and split information. This segmentation allows model transparency through global decision trees while preserving data privacy at the edge nodes.
Solution Approach 2:
The patent introduces an intermediary aggregation mechanism that computes global purity measures and split decisions without accessing individual edge node data. The intermediary layer processes local statistics (cardinality, sum, sum of squares) to determine global splits, acting as a mediator between local data and global model construction, thus maintaining both transparency and privacy.
2Measurement precision
If individual edge node data is revealed to construct decision trees, then model accuracy is improved, but data privacy is compromised
Solution Approach 1:
The patent merges local purity measures from multiple edge nodes to construct a global decision tree. Instead of using data from a single node, the system combines statistics (cardinality, sum, sum of squares) from multiple nodes to compute global splits. This merging approach improves model accuracy by incorporating diverse data perspectives while maintaining privacy through secure aggregation protocols.
Solution Approach 2:
The patent transforms raw data into aggregated parameters (cardinality, sum, sum of squares) that can be combined to compute global purity measures. By changing the representation from individual data points to statistical parameters, the system achieves accurate global models without exposing individual edge node data, thus maintaining both accuracy and privacy.
3Object-affected harmful factors
If secure aggregation protocols are implemented, then data privacy is improved, but computational complexity is increased
Solution Approach 1:
The patent implements secure aggregation by computing and sharing only the necessary partial statistics (cardinality, sum, sum of squares) required for purity measure calculation, rather than requiring full data exposure or complex cryptographic operations. This partial action approach provides privacy protection through aggregation while minimizing computational complexity to only what is essential for the task.
Data Source
AI summary
A horizontal federated random forest regressor with secure aggregation is disclosed. When constructing a node of a decision tree, multiple potential splits are performed at each of multiple edge nodes using local data. Federated variance data, which includes sums, is generated and transmitted to a central node. Using the sums, a global variance can be determined for each of the splits without requiring the individual nodes to share the specific samples. The split with the lowest global variance is selected by the central node and implemented for the node of the decision tree at each of the edge nodes. A random forest regressor can be constructed and trained such that each of the edge nodes includes the same random forest regressor with the same splits for the features at the nodes of the decision trees that constitute the random forest regressor.