Cutoff with an $O(1)$ window for Potts Glauber Dynamics on lattice at High Temperature
arXiv:2608.26259
2026
Architecture
2 ideas extracted · analyzed Aug 29, 2026
What the math gives to ML
The paper develops constructive information-percolation tools for a non-monotone q-state system, combining signed ancestral histories, renewal decompositions, q-ary overlap bounds, and explicit penalties for branching histories. The most transferable asset is a way to control dependence and receptive-field growth in categorical message-passing systems without relying on monotonicity or nonnegative influences. A practical adaptation is to replace unrestricted token-to-token communication by a stochastic history graph whose branching rate is deliberately kept subcritical, while monitoring an overlap moment that predicts redundancy and interference. This could yield sparse attention or categorical graph modules with bounded dependency growth and better stability at fixed compute.
Ideas from this paper
✗ Mechanism failed
2026
Construct a sparse attention layer by sampling backward token histories as a continuous-time branching process rather than allowing every query to attend to every key. Each active ancestor either dies or branches into a bounded number of candidate ancestors, with branching probability controlled by a small parameter. The branch-out penalty predicts exponentially small probability of long, highly branching histories, providing a direct knob for receptive-field size and attention FLOPs.
Useful7/10
Difficulty5/10
Novelty7/10
Unverified
2026
Use the paper's q-ary overlap inequality as a regularizer for categorical neural networks. Two independently sampled attention, routing, or message-passing supports should rarely overlap in many locations; penalizing the moment q^{|S\cap S'|} discourages redundant histories and correlated interference between heads or experts.
Useful6/10
Difficulty3/10
Novelty6/10