AI Blogs and Articles

Mc+qubo Achieves Improved Reinforcement Learning with Quadratic Unconstrained Binary Optimisation

Researchers introduced MC+QUBO, a method that recasts Monte Carlo episode selection as a Quadratic Unconstrained Binary Optimisation problem and solves it with quantum‑inspired samplers. By using Simulated Quantum Annealing and Simulated Bifurcation, the approach filters trajectories to maximise reward and promote exploration. Experiments on GridWorld environments from 3×3 to 20×20 show faster convergence and higher returns than standard Monte Carlo, especially in larger, sparse‑reward grids. Solver computation remains under 100 ms for problems up to 200 binary variables, with cloud latency being the primary overhead.