Statistical Set Updating for Secure Decentralized Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In decentralized distributed training of machine learning models, the convergence rate of statistical sets is limited by the processing power of individual nodes, and the global accessibility of these sets makes them vulnerable to unauthorized access by third parties.
Innovation Solution
Implementing a system that updates statistical sets by generating new sets from parent sets without sharing the underlying data, ensuring security and compliance by only sharing derived statistical sets among nodes, thereby preventing unauthorized access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If decentralized distributed training is implemented, then processing power is distributed across nodes, but convergence rate is limited by individual node processing power
Solution Approach 1:
The patent merges statistical sets from multiple decentralized nodes through averaging operations to create a unified global model. By combining the computational results from distributed nodes without centralizing the underlying data, the system achieves faster convergence than individual nodes could achieve alone, while maintaining the benefits of distributed processing power.
Solution Approach 2:
The patent introduces statistical sets as intermediary representations that capture the essence of training results without exposing underlying data. These statistical sets act as mediators between decentralized nodes, enabling information exchange and model convergence while preserving data locality and security at each node.
2Reliability
If statistical sets are made globally accessible for training, then model convergence is enabled, but vulnerability to unauthorized access by third parties increases
Solution Approach 1:
The patent extracts only the necessary statistical information from underlying data, separating the training model from the sensitive data. By transmitting and sharing only statistical sets (averages, gradients, or other aggregated metrics) rather than raw data, the system enables model convergence while removing the vulnerability associated with exposing underlying data to unauthorized access.
Solution Approach 2:
The patent uses statistical sets as simplified copies or representations of the underlying training data. These copies contain the essential information needed for model training and convergence but lack the detailed structure and sensitivity of the original data, thereby enabling collaborative training while maintaining data security and compliance.
3Measurement precision
If underlying data is shared among nodes for training, then training accuracy improves, but data security and compliance are compromised
Solution Approach 1:
The patent segments the training process into separate operations: local statistical set generation at each node, followed by centralized averaging of these statistical sets. This segmentation allows training accuracy to improve through aggregation of distributed knowledge while maintaining data security by never requiring the movement or sharing of underlying sensitive data between nodes.
Data Source
AI summary
Systems, computer-implemented methods, and/or computer program products to facilitate updating, such as averaging and/or training, of one or more statistical sets are provided. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can include a computing component that updates a first statistical set with an additional statistical set from an additional system. The additional statistical set can have been generated from a parent statistical set that is based on underlying data. To update the first statistical set, the additional statistical set can be obtained by the system without obtaining the parent statistical set and without obtaining the underlying data. According to an embodiment, the first statistical set can be a model parameter set generated from a first parent statistical set that is an analytical model.