Reproduce “Discovering Structure in High-Dimensional Data Through Correlation Explanation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Greg Ver Steeg, Aram Galstyan
Researchers Greg Ver Steeg and Aram Galstyan from USC's Information Sciences Institute developed Correlation Explanation (CorEx), an unsupervised learning method that uncovers hierarchical representations in high-dimensional data by maximizing the total correlation explained by latent factors. CorEx successfully recovered the "Big Five" personality traits from survey data and extracted multi-level demographic structures from human DNA, demonstrating superior performance over conventional methods in discovering latent tree models on synthetic data.
Key insights
- The total correlation within a dataset can be effectively explained and reduced by a set of latent variables, providing a principled measure for uncovering underlying common causes.
- A complex, non-convex information-theoretic optimization problem can be solved efficiently and practically for high-dimensional data through a scalable alternating optimization scheme.
- Hierarchical structures naturally emerge by recursively applying the correlation explanation principle, leading to richer and more abstract representations of complex systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.