https://aaai.org/ocs/index.php/WS/AAAIW18/paper/viewFile/16578/15542
对于文本分类,需要深度卷积网络吗?
讨论的问题:
1. 层次的深度
2. 粒度对于NLP。认为词向量可以分层的学习到,从词到n阶、短语,句子的特征
结论:对比图像的连续性,NLP任务中token的离散特征,在深度卷积网络中效果提升并不明显。
https://arxiv.org/abs/2012.15699
Better Robustness by More Coverage:Adversarial Training with Mixup Augmentation for Robust Fine-tuning