正文
Noise is an emerging and popular keyword in recent years. The noise-based models have attracted more and more attention in the pattern recognition community, including but not limited to random forest, dropout in neural networks (a kind of structural beneficial noise), generative adversarial networks, adversarial training, noisy augmentation, noisy label, positive-incentive noise, diffusion models, and flow matching models. Although most of these models don’t explicitly claim that they aim to learn noise, they actually utilize the beneficial noise implicitly. In many current studies, it is pointed out that noise can also be beneficial to large models. Therefore, noise should not be simply regarded as a harmful component any more. The positivity of noise deserves more systematic studies.
Noise is an emerging and popular keyword in recent years. The noise-based models have attracted more and more attention in the pattern recognition community, including but not limited to random forest, dropout in neural networks (a kind of structural beneficial noise), generative adversarial networks, adversarial training, noisy augmentation, noisy label, positive-incentive noise, diffusion models, and flow matching models. Although most of these models don’t explicitly claim that they aim to learn noise, they actually utilize the beneficial noise implicitly. In many current studies, it is pointed out that noise can also be beneficial to large models. Therefore, noise should not be simply regarded as a harmful component any more. The positivity of noise deserves more systematic studies.
Guest editors: Xuelong Li, Hongyuan Zhang, Murat Sensoy, Enze Xie
软件工程
Journal of Systems and Software
Evaluation and Assessment in Industrial Software Engineering
