The multi-modal optimization problem is considered. An automata model with improved learning schemes is proposed to solve the global optimization problem. The numerical simulation shows that the automata approach is better than the well-known gradient approach because the gradient approach is easily trapped inside the local optimal states. Theoretically, we prove that the automaton converges to the global optimum with a probability arbitrarily close to one. The simulation result also shows that our automata model converges faster than the existing models in the literature