Local Models Label Public Data for Privacy-Preserving Global Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In cloud computing and cognitive computing, data owned by multiple parties is often private and not shared due to security and proprietary concerns, making it challenging to train models using both private and public data without compromising privacy.
Innovation Solution
Local machine learning models are trained using private data to generate labels for unlabeled public data, which are then applied to create labeled public data that can be used by a global model without revealing the original private data, thus maintaining data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If private data is shared to train global models, then model training quality improves, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the essential information (labels) from private data after training local models, rather than sharing the actual private data. This extraction process removes sensitive information while retaining useful training signals, resolving the contradiction between model quality and privacy protection
Solution Approach 2:
The patent introduces local models as intermediaries that process private data locally and generate labels without exposing the underlying private data. These labels serve as a mediator that conveys training information to global models while maintaining a privacy barrier, thus resolving the contradiction
2Object-affected harmful factors
If private data is kept local for privacy protection, then data privacy is maintained, but model training effectiveness deteriorates due to lack of data sharing
Solution Approach 1:
The patent implements a feedback mechanism where local models trained on private data generate labels that are fed back to train global models. This feedback loop allows private data to indirectly contribute to global model training without being shared, resolving the contradiction between privacy protection and training effectiveness
Solution Approach 2:
Labels generated by local models serve as intermediaries that carry training value from private data to global models without exposing the sensitive underlying data. This intermediary mechanism enables effective training while maintaining privacy
3Loss of energy
If labeled public data is created using private data, then bandwidth consumption is reduced, but the complexity of the labeling process increases
Solution Approach 1:
The patent employs local models to automatically generate labels for public data without requiring manual annotation. This self-service labeling approach reduces the need for human intervention and bandwidth-intensive data transfer, though it introduces computational complexity at the local model level
Data Source
AI summary
A computer-implemented method applies labels to unlabeled public data for use by a global model. One or more processors train one or more local machine learning models with local private data to create one or more trained models. Processor(s) generate a label for each of the local private data using the one or more trained models, where each label describes the local private data, and then apply the label to unlabeled public data to create labeled public data. One or more processors then input the labeled public data into a global model that uses the public data.