How iSTAR Exploits Trajectory Collapse to Speed Quantum-Inspired Solvers

Bowen Liu of eBay Inc. and Dongmei Xiao of Shanghai Jiao Tong University have developed iSTAR, a new algorithm that reduces the computational load of quantum-inspired continuous Ising solvers. The researchers demonstrate that dense interaction costs within these solvers are not inherent; during late-stage simulated bifurcation, solution trajectories collapse onto a lower-dimensional space. This collapse allows iSTAR (Ising Stable-set Tail-Aware Reduction) to identify stabilized coordinates and continue only on the active tail, incorporating their influence into an induced field on the remaining active variables. An online certified implementation on the G-set benchmark preserves the same-seed…