Delegation Verifier for Privacy-Controlled ML Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The training of machine learning models using video and audio data from monitored individuals raises privacy concerns, as existing solutions do not adequately allow individuals to control when their data is used for training.
Innovation Solution
A method and system that utilize a chain of delegations to manage access to media data for training machine learning models, where a data structure comprising a chain of delegations is used to request and obtain decryption keys, ensuring that data usage is controlled and privacy is protected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If video and audio data from monitored individuals is used for training machine learning models, then the performance and adaptability of the models improve, but privacy concerns arise and individuals lose control over their data
Solution Approach 1:
A delegation verifier acts as an intermediary between the media capturing device and the training data provider. The verifier receives delegation information from the device, verifies its authenticity, and issues decryption keys to authorized providers. This intermediary mechanism enables model training while protecting individual privacy control, as the verifier ensures only authorized entities can access encrypted media data for training purposes.
2Reliability
If continuous monitoring is implemented to improve quality of life for elderly or disabled individuals, then care and safety are enhanced, but privacy issues arise as individuals feel continuously monitored
Solution Approach 1:
The patent extracts and separates the training data provision function from continuous monitoring. Media capturing devices continuously monitor individuals for care and safety, but the training data provider only receives authorized, encrypted data specifically for model training purposes. This extraction allows continuous monitoring benefits while addressing privacy concerns by limiting training data access to authorized scenarios only.
3Measurement precision
If machine learning models are trained using monitored data to enable continuous improvement, then model accuracy and reliability increase, but individuals cannot control when their data is used for training
Solution Approach 1:
The delegation verifier serves as a mediator that implements individual control over training data usage. Individuals can specify conditions under which their data may be used for training, and the verifier enforces these conditions by checking delegation information before issuing decryption keys. This maintains both model accuracy through authorized training and individual control through condition-based access.
Solution Approach 2:
The system dynamically adjusts data access based on delegation conditions. The training data provider receives encrypted media data and decryption keys only when specific conditions are met, as verified by the delegation verifier. This dynamic control mechanism allows individuals to specify when their data can be used for training, balancing model accuracy improvement with individual autonomy.
Data Source
AI summary
It is provided a method for providing data for training a machine learning model. The method is performed in a training data provider (1) and comprises the steps of: obtaining (40) a data structure comprising a chain of delegations, the chain of delegations covering a delegation path from a media capturing device (3) to the training data provider (1) such that, in the chain of delegations, each delegation is a delegation from a delegator to a receiver; sending (42) a key request to a delegation verifier (2), the key request comprising the data structure; receiving (44) a decryption key from the delegation verifier (2); obtaining (46) encrypted media data captured by the media capturing device (3); decrypting (48) the encrypted media data, resulting in decrypted media data; and providing (50) the decrypted media data for training the machine learning model.