Rigidity of expanders and pseudorandom graphs
arXiv:2608.21058
2026
Architecture
2 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper supplies a graph-design principle for making pairwise distance constraints identify a high-dimensional configuration: sufficiently jumbled graphs contain a large induced subgraph that is generically rigid in dimension proportional to edge density. This can transfer to geometric neural networks by replacing arbitrary sparse interaction masks with pseudorandom masks whose edge distances constrain latent node coordinates up to global translation and rotation. The most promising uses are rigidity-certified latent-coordinate modules and sparse attention masks with uniform global coverage. The theorem's stated constants are extremely conservative, so practical experiments should estimate rigidity numerically rather than apply the constants literally.
Ideas from this paper
Unverified
2026
Use a pseudorandom sparse interaction graph as the measurement pattern for latent node coordinates. Add a loss on edgewise latent distances and train on an automatically selected large induced subset, so that coordinates are constrained by many distributed measurements rather than local neighborhoods alone. The target is to eliminate non-global geometric ambiguities and reduce drift in geometric GNN or transformer representations.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Design sparse attention masks using a graph discrepancy criterion rather than selecting only local or nearest-neighbor edges. A mask with approximately uniform edge counts between every pair of token subsets spreads information globally, while the rigidity consequence provides a principled way to preserve enough independent pairwise constraints for latent geometric features.
Useful5/10
Difficulty4/10
Novelty6/10