通过匹配数据与隐空间的成对距离,提升自编码器的结构保持能力。
Manifold-Matching Autoencoders
- 用成对距离误差最小化实现隐空间与输入空间的结构对齐。
- 在近邻保持和持久同调度量上优于同类方法。
- 可扩展地近似多维缩放,适合高维数据降维与可视化。
我们研究了一种简单的无监督正则化方案——流形匹配自编码器(Manifold-Matching Autoencoders, MMAE):通过最小化均方误差,使隐空间中的成对距离与输入数据空间中的成对距离对齐。由于对齐基于距离而非坐标,该方法可推广至数据的低维表示,增强灵活性。实验表明,MMAE 在基于最近邻距离保持和持久同调度量的指标上优于现有方法。此外,MMAE 可作为多维缩放(Multi-Dimensional Scaling, MDS)的可扩展近似,适用于大规模数据的结构保持降维。
原文摘要 · Abstract (English)
We study a simple unsupervised regularization scheme for autoencoders called Manifold-Matching (MMAE): we align the pairwise distances in the latent space to those of the input data space by minimizing mean squared error. Because alignment occurs on pairwise distances rather than coordinates, it can also be extended to a lower-dimensional representation of the data, adding flexibility to the method. We find that this regularization outperforms similar methods on metrics based on preservation of nearest-neighbor distances and persistent homology-based measures. We also observe that MMAE provides a scalable approximation of Multi-Dimensional Scaling (MDS).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。