Transfer Learning for Private Data Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning approaches struggle to facilitate shared learning across entities without requiring the sharing of sensitive data, as they often necessitate direct access to each other's data, violating privacy constraints and differing data labeling practices.
Innovation Solution
The implementation of a system comprising an initial model trainer, iterative model trainers, a model training broker, and a model prediction evaluation broker, which allows entities to train classifiers using shared scores or embeddings that do not reveal private features, enabling collaborative learning without data sharing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If entities share sensitive data for machine learning, then learning effectiveness and pattern recognition improve, but data privacy and security are compromised
Solution Approach 1:
The patent introduces an intermediary system that enables collaborative learning between entities without direct data sharing. The intermediary coordinates the exchange of model updates, scores, and embeddings while preventing access to raw private data, thus resolving the contradiction between learning effectiveness and data privacy protection
Solution Approach 2:
The patent uses model copies, score copies, and embedding copies as surrogates for the actual private data. Entities train and share these copies instead of the original sensitive data, allowing pattern recognition to improve while maintaining data privacy through the use of data representations that do not reveal the underlying private information
2Reliability
If entities use different data labeling practices, then each entity maintains its own standards and quality control, but shared learning and model collaboration become difficult
Solution Approach 1:
The patent transforms the labeling problem by changing from sharing labeled data to sharing unlabeled data with model scores. Entities can maintain their own labeling standards and quality control while the model learns from the aggregated scores and embeddings, enabling cross-entity learning compatibility without compromising individual labeling quality
3Productivity
If entities directly access each other's data for training, then model training is simplified and faster, but privacy constraints and security requirements are violated
Solution Approach 1:
The intermediary system streamlines the training process by automating the coordination of data exchanges, model updates, and score aggregations. This maintains training efficiency by providing a structured workflow while ensuring privacy constraints are met through controlled, indirect information exchange between entities
Solution Approach 2:
The patent enables efficient training by working with copies of data representations (scores, embeddings, model parameters) instead of raw data. These copies can be exchanged and processed quickly while maintaining privacy, thus preserving training productivity without violating privacy constraints
Data Source
AI summary
Embodiments herein use transfer learning paradigms to facilitate classification across entities without requiring the entities access to the other party's sensitive data. In one or more embodiments, one entity may train a model using its own data (which may include at least some non-shared data) and shares either the scores (or an intermediate representation of the scores). One or more other parties may use the scores as a feature in its own model. The scores may be considered to act as an embedding of the features but do not reveal the features. In other embodiments, parties may be used to train part of a model or participate in generating one or more nodes of a decision tree without revealing all its features. The trained models or decision trees may then be used for classifying unlabeled events or items.