Abstract
Photonic Ising machines are leading key advancements in solving large combinatorial problems, leveraging large-scale platforms with parallel computing capabilities. A well-known bottleneck of complex problems is the appearance of multiple minima in the energetic landscape that attract Metropolis-based iterations in suboptimal solutions, thus hindering the performance of standard optical solvers in large systems. By introducing a double single-pixel detection scheme based on intensity and field averages in an optical-based Ising machine, we effectively implement local and nonlocal nonlinear Hamiltonians, representing a complex and simple state, respectively. Transitioning from nonlocal to local nonlinear detection enables to adiabatically morph the energetic landscape, enhancing the success rate of finding the optimal solution compared to standard isothermal approaches.