Reproduce “Random Forest Autoencoders for Guided Representation Learning” 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
Adrien Aumon, Shuang Ni, Myriam Lizotte, Guy Wolf, Kevin R. Moon, Jake S. Rhodes
Random Forest Autoencoders (RF-AE) provide a framework for creating supervised, interpretable out-of-sample data visualizations by mapping Random Forest-derived proximities to a low-dimensional space using an autoencoder with geometric regularization. The method achieves high classification accuracy while robustly preserving the structural importance of task-relevant features.
Key insights
- Random Forest-derived proximities, specifically the extended RF-GAP kernel, effectively capture supervised neighborhood information critical for guiding representation learning.
- Combining proximity reconstruction with geometric regularization in an autoencoder architecture allows for learning a parametric mapping that preserves complex, classification-relevant manifold structures.
- Class-wise prototype selection not only significantly reduces computational costs and accelerates OOS projection but can also act as an implicit regularizer, potentially improving embedding quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.