Shared Embedding Space Generation Across Data Silos
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches fail to generate embeddings in a shared embedding space that can be meaningfully combined across multiple separate data silos due to random rotations, resulting in embedding spaces with different semantic structures.
Innovation Solution
A central computing system generates synthetic data examples from aggregated data distributions across silo computing systems, which are then used to train embedding generation models or directly generate embeddings within a shared embedding space, ensuring consistent semantic structure across silos.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is stored in separate data silos, then data privacy and operational independence are improved, but embedding space consistency and semantic interpretability across silos deteriorate
Solution Approach 1:
A central computing system acts as an intermediary that receives data distributions from multiple silo computing systems, generates synthetic data examples in a unified feature space, and provides these synthetic examples back to the silos. This mediator enables embedding space consistency without direct data sharing between silos, thus preserving privacy while achieving semantic interpretability across silos.
2Stability of the object's composition
If synthetic data examples are generated from aggregated data distributions, then embedding space consistency is improved, but computational complexity and data processing requirements increase
Solution Approach 1:
The system segments the data processing task by having each silo computing system first determine and transmit only the data distribution characteristics (e.g., statistical parameters) of its local data to the central system. The central system then aggregates these distributed statistics to generate synthetic data examples. This segmentation reduces the computational burden on individual silos while enabling consistent embedding space generation across all silos.
Data Source
AI summary
Provided are systems and methods for privacy-preserving learning and analytics of a shared embedding space for data split across multiple separate data silos. A central computing system can generate a plurality of synthetic data examples having respective feature data within an aggregate feature-space that represents an aggregation of different component feature-spaces associated with the multiple separate data silos. The synthetic data examples can be used by different computing systems associated with the data silos to generate embeddings within a shared embedding space. Once the embeddings have been generated in the shared embedding space, multiple different types of analytics can be performed on the shared embedding space. As one example, the multiple data silos can correspond to multiple separate entity domains and an analysis of embeddings generated in the shared embedding space can be used to facilitate identification or classification of malicious actors across the multiple separate entity domains.