Low-Dimensional Embeddings and Dimensionality Reduction

The field of low-dimensional embeddings and dimensionality reduction is rapidly advancing, driven by the increasing demand for effective methods to analyze and visualize high-dimensional data. Recent developments have focused on improving the preservation of local and global structures in the data, as well as enhancing the scalability and stability of existing techniques. Notably, researchers are exploring new approaches to capture complex relationships between data points, such as heterogeneous co-occurrence embedding and linear cost mutual information estimation. These innovations have the potential to significantly improve the reliability and accuracy of visual analytics and data exploration. Noteworthy papers include: UMATO, which introduces a two-phase optimization technique to effectively capture local and global structures in high-dimensional data. Heterogeneous co-occurrence embedding, which enables the visualization of asymmetric relationships between heterogeneous domains. Linear cost mutual information estimation, which provides a practical alternative to HSIC with higher dependence sensitivity and linear computational complexity.

Sources

Low-dimensional embeddings of high-dimensional data

UMATO: Bridging Local and Global Structures for Reliable Visual Analytics with Dimensionality Reduction

Heterogeneous co-occurrence embedding for visual information exploration

Linear cost mutual information estimation and independence test of similar performance as HSIC

High-Dimensional Quasi-Monte Carlo via Combinatorial Discrepancy

Embedding Font Impression Word Tags Based on Co-occurrence

Dimension Agnostic Testing of Survey Data Credibility through the Lens of Regression

Built with on top of