SSGD: A safe and efficient method of gradient descent
This work aims to improve the security of gradient descent for multi-node machine learning systems by preventing gradient leakage, which is an incremental improvement to existing methods.
This paper addresses the vulnerability of shared gradients in multi-node machine learning systems, where attackers can reconstruct training data from gradient information. The authors propose Super Stochastic Gradient Descent (SSGD) to prevent gradient leakage by concealing the modulus length of gradient vectors and converting them into unit vectors. Experimental results indicate that SSGD outperforms prevalent gradient descent approaches in accuracy, robustness, and adaptability to large-scale batches.
With the vigorous development of artificial intelligence technology, various engineering technology applications have been implemented one after another. The gradient descent method plays an important role in solving various optimization problems, due to its simple structure, good stability and easy implementation. In multi-node machine learning system, the gradients usually need to be shared. Shared gradients are generally unsafe. Attackers can obtain training data simply by knowing the gradient information. In this paper, to prevent gradient leakage while keeping the accuracy of model, we propose the super stochastic gradient descent approach to update parameters by concealing the modulus length of gradient vectors and converting it or them into a unit vector. Furthermore, we analyze the security of super stochastic gradient descent approach. Our algorithm can defend against attacks on the gradient. Experiment results show that our approach is obviously superior to prevalent gradient descent approaches in terms of accuracy, robustness, and adaptability to large-scale batches.