BACON
收藏资源简介:
BACON数据集是由北京航空航天大学电子信息工程学院和利物浦大学计算机科学系联合开发的,旨在通过Bayesian理论框架优化数据集精简过程。该数据集通过将大型数据集的知识提炼成更紧凑的形式,同时保持测试集的性能,从而降低存储成本和训练费用。BACON数据集的应用领域包括持续学习、联邦学习、知识蒸馏和对抗学习等,旨在解决现有数据集精简方法中存在的计算强度大和性能不理想的问题。
The BACON dataset was jointly developed by the School of Electronic and Information Engineering of Beihang University and the Department of Computer Science of the University of Liverpool, with the goal of optimizing the dataset pruning process through a Bayesian theoretical framework. This dataset distills the knowledge contained in large-scale datasets into a more compact form, while preserving the performance on the test set, thereby lowering both storage costs and training expenses. The application fields of the BACON dataset include continual learning, federated learning, knowledge distillation, adversarial learning and other related domains, and it aims to resolve the problems of high computational intensity and unsatisfactory performance in existing dataset pruning methods.

- 1BACON: Bayesian Optimal Condensation Framework for Dataset Distillation北京航空航天大学电子信息工程学院 · 2024年



