Obfuscation Network Training with Information Reduction for Edge Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing adversarial representation learning (ARL) methods require significant computing resources and fail to function properly on edge devices with limited resources, leading to privacy leaks and inefficient processing.
Innovation Solution
A learning method for training an obfuscation network that minimizes computing resources by using an encoding network and information reduction module, combined with a proxy adversarial network to preserve task utility and protect privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If previous ARL methods are used to train the obfuscator to generate obfuscated representation with high task utility, then task utility is improved, but computing resources and processing time are excessively consumed
Solution Approach 1:
The obfuscation network is segmented into two distinct components: an encoding network that extracts features and generates data representation, and an information reduction module that transforms the representation to remove sensitive attributes. This segmentation allows each component to be optimized independently, reducing overall computational burden while maintaining task utility.
Solution Approach 2:
The information reduction module specifically extracts and removes sensitive attributes from the data representation while preserving task-relevant information. By taking out only the necessary privacy-protecting transformations rather than comprehensive obfuscation, the system reduces computing resources while maintaining both privacy and utility.
2Reliability
If the trained obfuscator requiring significant computations is installed on edge devices with limited computing resources, then privacy protection is improved, but the obfuscator fails to function properly and consumes excessive time
Solution Approach 1:
The information reduction module uses computationally lightweight operations such as random projection, downsampling, and simple transformations that can be executed efficiently on resource-constrained edge devices. These simplified operations provide adequate privacy protection without requiring the significant computational resources of previous ARL methods.
3Loss of information
If the obfuscator removes all information unrelated to the utility task, then privacy leakage is reduced, but the processing time and computational load increase significantly
Solution Approach 1:
The information reduction module applies partial obfuscation by removing only the most critical sensitive attributes through targeted transformations rather than comprehensively removing all unrelated information. This partial action approach reduces processing time while still achieving adequate privacy protection for the most sensitive information.
Data Source
AI summary
A learning method for training an obfuscation network, including steps of: (a) inputting a training data into the obfuscation network to (i) extract features and thus generate a data representation by performing a learning operation on the training data and (ii) transform the data representation and thus generate an anonymized data representation, and (b) inputting the anonymized data representation into a task learning network to (i) perform a task by using the anonymized data representation and thus output a task result, (ii) generate a task loss by referring to the task result and its corresponding ground truth, (iii) train the task learning network through a first backpropagation of the task loss such that the task loss is minimized, and (iv) train the obfuscation network through a second backpropagation of the task loss such that the task loss is minimized.