Noisy Deep Ensemble: Accelerating Deep Ensemble Learning via Noise Injection

Neural network ensembles is a simple yet effective approach for enhancing generalization capabilities. The most common method involves independently training multiple neural networks initialized with different weights and then averaging their predictions during inference. However, this approach increases training time linearly with the number of ensemble members. To address this issue, we propose the novel ``$\mathbf{\text{Noisy}}\mathbf{\text{Deep}}\mathbf{\text{Ensemble}}$‘’ method, significantly reducing the training time required for neural network ensembles. In this method, a $\mathit{\text{parent}}\mathit{\text{model}}$ is trained until convergence, and then the weights of the $\mathit{\text{parent}}\mathit{\text{model}}$ are perturbed in various ways to construct multiple $\mathit{\text{child}}\mathit{\text{models}}$. This perturbation of the $\mathit{\text{parent}}\mathit{\text{model}}$ weights facilitates the exploration of different local minima while significantly reducing the training time for each ensemble member. We evaluated our method using diverse CNN architectures on CIFAR-10 and CIFAR-100 datasets, surpassing conventional efficient ensemble methods and achieving test accuracy comparable to standard ensembles. Code is available at \href{https://github.com/TSTB-dev/NoisyDeepEnsemble}{https://github.com/TSTB-dev/NoisyDeepEnsemble}