Get Started
Home
Topics
Search
Library
Research questionHow can categorical graph generators preserve coherent, permutation-invariant structure without artificial category relationships?Categorical graph generation must choose node and edge types that form coherent structures while remaining independent of how nodes are ordered. Fixed one-hot encodings can impose an artificial geometry that treats categories as equally separated.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.Embedded Graph Flows for Categorical Graph GenerationThe evidence concerns Embedded Graph Flows, which learn continuous embeddings for node and unordered-edge categories and use a permutation-equivariant graph transformer before decoding them back to discrete categories. Results are reported on molecular QM9 and ZINC250k benchmarks against DiGress and GruM, using FCD, MMD, and NSPDK-based evaluation; broader graph-domain performance is not established by the input.research paper · Sep 4, 2026
Related questions
How can graph topology inference from node observations be unified with generating new graph instances?How can graph processors preserve relational intermediate states when learning algorithmic computations?How can graph neural networks capture long-range interactions without oversmoothing or oversquashing on diverse large graphs?How can attribute hypergraphs refine frozen graph-clustering assignments without labels or altering the model or original graph?