DMT-HI: MOE-based Hyperbolic Interpretable Deep Manifold Transformation for Unspervised Dimensionality Reduction
Zelin Zang,Yuhao Wang,Jinlin Wu,Hong Liu,Yue Shen,Stan.Z Li,Zhen Lei
2024-10-25
Abstract:Dimensionality reduction (DR) plays a crucial role in various fields, including data engineering and visualization, by simplifying complex datasets while retaining essential information. However, the challenge of balancing DR accuracy and interpretability remains crucial, particularly for users dealing with high-dimensional data. Traditional DR methods often face a trade-off between precision and transparency, where optimizing for performance can lead to reduced interpretability, and vice versa. This limitation is especially prominent in real-world applications such as image, tabular, and text data analysis, where both accuracy and interpretability are critical. To address these challenges, this work introduces the MOE-based Hyperbolic Interpretable Deep Manifold Transformation (DMT-HI). The proposed approach combines hyperbolic embeddings, which effectively capture complex hierarchical structures, with Mixture of Experts (MOE) models, which dynamically allocate tasks based on input features. DMT-HI enhances DR accuracy by leveraging hyperbolic embeddings to represent the hierarchical nature of data, while also improving interpretability by explicitly linking input data, embedding outcomes, and key features through the MOE structure. Extensive experiments demonstrate that DMT-HI consistently achieves superior performance in both DR accuracy and model interpretability, making it a robust solution for complex data analysis. The code is available at \url{<a class="link-external link-https" href="https://github.com/zangzelin/code_dmthi" rel="external noopener nofollow">this https URL</a>}.
Machine Learning,Artificial Intelligence