Encrypted Tree Ensemble Inference via Node Path Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Tree-based ensemble models face significant performance degradation and increased computational costs when operating in an encrypted domain, particularly in untrusted environments, due to the need to evaluate all paths and nodes.
Innovation Solution
The method involves optimizing a tree-based ensemble model by identifying unique combinations of features and thresholds, assigning distinct identifiers to nodes, and representing paths using these identifiers. This optimized model is then encrypted using a fully homomorphic encryption algorithm, allowing secure inference in an untrusted domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If tree-based ensemble models are deployed in encrypted domain for secure inference, then data security is improved, but inference speed and computational efficiency deteriorate
Solution Approach 1:
The patent segments the tree-based ensemble model into multiple decision trees, each with multiple paths. By representing paths using distinct node identifiers and evaluating only relevant paths rather than all possible paths, the system reduces the computational burden in the encrypted domain while maintaining security through homomorphic encryption.
2Measurement precision
If all nodes and paths in tree-based ensemble model are evaluated in encrypted domain, then inference accuracy is maintained, but computational cost increases significantly
Solution Approach 1:
The patent extracts and represents only the necessary paths through the decision trees using distinct node identifiers. By identifying and evaluating only the relevant paths that contribute to the final prediction, rather than evaluating all possible paths, the system maintains inference accuracy while significantly reducing computational cost in the encrypted domain.
3Productivity
If tree-based ensemble model is optimized by identifying unique feature-threshold combinations, then model efficiency is improved, but model complexity increases
Solution Approach 1:
The patent performs preliminary optimization by identifying unique feature-threshold combinations and assigning distinct node identifiers to represent paths before deployment. This preprocessing step consolidates redundant nodes and paths, creating an optimized model structure that improves efficiency during inference while the added complexity of node identification is incurred only once during model preparation.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The optimized approach reduces the number of nodes and paths evaluated, leading to faster and more efficient inference operations while maintaining data security by protecting the information about the tree-based ensemble model.
Implementation Method 1
receiving data of one or more examples, the data encrypted by a fully homomorphic encryption algorithm
Data Source
AI summary
Techniques are provided for inference by tree-based ensemble models on encrypted data. One method includes identifying a plurality of nodes included in a plurality of decision trees of a tree-based ensemble model, and determining, from the plurality of nodes, a first set of nodes where each node represents a unique combination of a feature and a threshold. The method further includes assigning distinct identifiers to the nodes of the first set, identifying a second set of paths included in the plurality of decision trees, and generating an optimized model, where each path of the second set is represented using the distinct identifiers that correspond to the respective nodes along the path, and branch directions taken from the respective nodes.