Reproduce “Knowledge-Embedded Latent Projection for Robust 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
Weijing Tang, Ming Yuan, Zongqi Xia, Tianxi Cai
The Knowledge-Embedded Latent Projection (KELP) model enhances representation learning for high-dimensional, imbalanced, and sparse discrete data by fusing internal observations with external semantic information. It learns more robust and accurate latent embeddings, especially in scenarios with limited sample sizes, and demonstrates improved performance in tasks like knowledge graph reconstruction and patient phenotyping in Electronic Health Records.
Key insights
- Formally integrating external semantic embeddings through an RKHS mapping significantly reduces the effective degrees of freedom in latent space models, improving robustness in imbalanced data settings.
- Using KPCA for subspace construction allows KELP to leverage the RKHS framework for large feature spaces while maintaining computational efficiency.
- The data-driven kernel selection procedure is critical for ensuring adaptive knowledge fusion, preventing negative transfer of information and enhancing overall model reliability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.