MER-DG: Modality-Entropy Regularization for Multimodal Domain Generalization
📄 MER-DG: Modality-Entropy Regularization for Multimodal Domain Generalization 5.4/10 | 创新 1/2 | 严谨 1.2/1.5 | 实验 1/1.5 | 清晰 0.8/1 | 影响 0.6/1.5 | 开源 0/1.5 | 复现 0.3/0.5 | 工程 0.5/1.5 📝 5.4/10 | 后50% | #音视频理解 | #多模态模型 | arxiv 👥 作者与机构 第一作者:Yavuz Yarici(Georgia Institute of Technology, OLIVES at the Center for Signal and Information Processing, School of Electrical and Computer Engineering) 通讯作者:Yavuz Yarici(Georgia Institute of Technology, OLIVES at the Center for Signal and Information Processing, School of Electrical and Computer Engineering) 作者列表:Yavuz Yarici(Georgia Institute of Technology, OLIVES at the Center for Signal and Information Processing, School of Electrical and Computer Engineering)、Ghassan AlRegib(Georgia Institute of Technology, OLIVES at the Center for Signal and Information Processing, School of Electrical and Computer Engineering) 💡 毒舌点评 这项工作精准诊断了多模态域泛化中的一个关键失败模式——“Fusion Overfitting”,并通过熵正则化这一手段实现了一致性的性能提升,融合训练导致编码器退化的问题诊断和验证体系较为完整。然而,方法的创新性本质上是将已知的信息最大化技术(Log-Determinant熵估计)拆解后嫁接到多模态编码器上,理论贡献有限;实验仅在两个来自同一社区的小规模数据集上完成,且基线覆盖不全,泛化性存疑;缺乏代码与模型开源进一步降低了其实际影响力。 ...