Task-Specific Hash Keys for Privacy Across Neural Tasks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep neural networks are susceptible to deep embedding attacks when accessed by attackers, compromising personal data, and current methods to mitigate this, such as adding noise or using API interfaces, are not always effective, while systems with multiple models are less efficient and accurate.
Innovation Solution
Implementing a system that trains an encoder to generate task-specific hash keys, allowing access to only necessary information, thereby preventing unauthorized data access and maintaining privacy across multiple tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single pre-trained neural network is reused for multiple tasks, then efficiency and accuracy are improved, but the system becomes susceptible to deep embedding attacks that compromise personal data
Solution Approach 1:
The patent segments the neural network model into two distinct components: a shared pre-trained encoder that extracts features from input data, and task-specific hash keys that control access to task-specific information. This segmentation allows the encoder to be reused across multiple tasks for efficiency, while the hash keys provide security by limiting access to specific task data. The encoder processes input observations and generates embeddings, but without the corresponding hash key, an attacker cannot access or infer task-specific personal data.
Solution Approach 2:
The patent introduces hash keys as an intermediary mechanism between the shared encoder and task-specific data. The hash keys act as a gatekeeper that mediates access control - they are required to properly query the encoder for task-specific information. This intermediary layer prevents direct access to personal data while allowing legitimate task execution, thus resolving the contradiction between sharing the model for efficiency and protecting data security.
2Reliability
If noise is added to the input space or API interfaces are employed, then some protection against attacks is achieved, but these techniques are not always effective and may reduce system performance
Solution Approach 1:
The patent extracts the security function from the input data and API interface layer and places it at the model parameter layer through hash keys. Instead of adding noise to inputs or restricting API access, the system extracts and protects the critical task-specific information using hash keys that selectively enable or disable access to specific task data. This approach provides robust attack resistance without degrading system performance because the encoder itself remains fully functional and accurate.
3Object-affected harmful factors
If multiple separate models are employed to protect data, then attack on a single model does not compromise all data, but the system becomes less efficient and accurate
Solution Approach 1:
The patent merges multiple task-specific functionalities into a single shared encoder model, improving efficiency and accuracy. The encoder is trained on multiple tasks and learns shared representations that benefit all tasks. Security is maintained not by separating models, but by combining task-specific hash keys with the shared encoder. This allows the system to enjoy the benefits of model sharing while maintaining security through the hash key mechanism that controls access to task-specific data.
4Ease of operation
If full control of the pre-trained encoder is given to developers, then they can access all information, but this exposes all data to potential attacks
Solution Approach 1:
The patent applies local quality by providing different levels of access control to different parts of the model. The shared encoder has universal access to process various types of input data, but task-specific hash keys provide localized access control to task-specific information. Developers can operate on specific tasks using only the hash keys relevant to those tasks, gaining sufficient access to perform their functions while being restricted from accessing unrelated task data. This resolves the contradiction between ease of operation and data exposure.
Data Source
AI summary
A system for using hash keys to preserve privacy across multiple tasks is disclosed. The system may provide training batch(es) of input observations each having a customer request and stored task to an encoder, and assign a hash key(s) to each of the stored tasks. The system may provide a new batch of input observations with a new customer request and new task to the encoder. The encoder may generate a new hash key assigned to the new customer request and determine whether any existing hash key corresponds with the new hash key. If so, the system may associate the new batch of input observations with the corresponding hash key and update the corresponding hash key such that it is also configured to provide access to the new batch of input observations. If not, the system may generate a new stored task and assign the new hash key to it.