Reproduce “Differentiable Lifting for Topological Neural Networks” 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
Jorge Luiz Franco, Gabriel Duarte, Alexander Nikitin, Moacir Ponti, Diego Mesquita, Amauri H. Souza
The paper introduces ∂lift (DiffLift), a differentiable framework that learns graph liftings to higher-order topological domains, integrating this process into an end-to-end training pipeline for Topological Neural Networks (TNNs). This approach consistently improves TNN performance on graph and node classification tasks, outperforming static lifting methods and other differentiable approaches by up to approximately 45% in accuracy.
Key insights
- Making the graph lifting process learnable and task-informed significantly enhances the expressive power and overall performance of Topological Neural Networks, surpassing fixed, unsupervised approaches.
- The quality of the initial node embeddings generated by the GNN backbone directly influences the effectiveness of learned higher-order structures and, consequently, the TNN's final performance.
- Using probabilistic sampling for accepting or rejecting candidate cells generally yields better performance than deterministic thresholding, suggesting benefits from exploring structural variations during training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.