返回
Stochastic subsampling with average pooling
DOI:10.1016/j.asoc.2026.114611.png)
摘要
En 中文
• We propose a new module called stochastic average pooling (SAP). • SAP attains improved theoretical properties of consistency compared to Dropout. • SAP can be seamlessly integrated into existing architectures. • Experiments demonstrate that SAP improves performance across a variety of tasks. • Ablation studies were conducted to validate the module design of SAP.
期刊
IF:
6.6
论文数:
1.4W
被引数:
4.8W
机构
引用论文
Regularizing deep neural networks for medical image analysis with augmented batch normalization用于医学图像分析的正则化深度神经网络与增强批归一化
Efficient token pruning in Vision Transformers using an attention-based Multilayer Network基于注意力机制的多层网络在视觉Transformer中的高效token剪枝
没有更多内容

