Reproduce “Contrastive Network 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
Zihan Dong, Xin Zhou, Ryumei Nakada, Lexin Li, Linjun Zhang
Adaptive Contrastive Edge Representation Learning (ACERL) provides a statistically robust framework for generating network edge embeddings directly from network structure, without relying on node or edge attributes. The method achieves minimax optimal convergence rates for sparse and heterogeneous networks by employing a novel adaptive masking strategy, leading to improved classification and prediction accuracy in brain connectivity analyses.
Key insights
- Adaptive random masking, which dynamically adjusts augmentation based on edge signal strength, is crucial for handling heterogeneous noise and leads to sharper error rates in contrastive learning.
- Conventional methods like Sparse Principal Components Analysis (sPCA) fundamentally deteriorate under heterogeneous noise, exhibiting an O(sqrt(r)) estimation error lower bound.
- ACERL achieves the minimax optimal convergence rate for edge representation learning in sparse and heterogeneous network settings, providing a strong theoretical foundation for its performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.