SecDD: Efficient and Secure Method for Remotely Training Neural Networks
This addresses security concerns for remotely deployed neural networks, but appears incremental as it repurposes existing vulnerabilities rather than introducing a fundamentally new approach.
The paper tackles the problem of securely training neural networks over unsecured channels by leveraging typically negative aspects of deep learning, such as high computational cost and vulnerability to adversarial perturbations, to develop an efficient method.
We leverage what are typically considered the worst qualities of deep learning algorithms - high computational cost, requirement for large data, no explainability, high dependence on hyper-parameter choice, overfitting, and vulnerability to adversarial perturbations - in order to create a method for the secure and efficient training of remotely deployed neural networks over unsecured channels.