Influence maximization under the linear threshold model on a CMOS Ising solver

Influence Maximization (IM) is a fundamental problem in network science with applications in viral marketing, information dissemination, cybersecurity, and epidemiology. Classical IM solvers often trade off solution quality for runtime efficiency due to the NP-hardness of typical models such as Linear Threshold, Independent Cascade, and Triggering. Among these, the Linear Threshold model stands out by avoiding costly Monte Carlo simulations, enabling a more tractable Ising formulation. In this study, we propose a novel workflow for solving the Linear Threshold IM problem on directed acyclic graphs using a CMOS Ising solver through an Ising formulation. Our approach combines an integer linear programming-based Ising formulation with hardware-aware decomposition and preprocessing steps that adapt the model to the constraints of the hardware solver. We evaluate the efficiency of our approach under various coefficients and identify the optimized configuration. Experimental results show that our approach achieves superior solvability over some state-of-the-art IM solvers while maintaining competitive runtime and orders-of-magnitude lower energy consumption in both randomly generated and real-world benchmarks. These results demonstrate the potential of Ising solvers for energy-efficient IM applications.