In this paper, we introduce a novel articial neural network to solve the portfolio optimiza- tion problem. The proposed neural network is called the Mixed Tabu Machine since its structure is similar to the Tabu Machine, but includes both discrete and continues variables. Similar to the Hopeld network, the state of the Mixed Tabu Machine is updated to nd the global minimum energy state. To escape from local minimum states of the energy in the Mixed Tabu Machine, the state transition mechanism is controlled by a tabu search in both discrete and continues search spaces. The experimental results for ve standard benchmark data sets show that the Mixed Tabu Machine can clearly obtain better solutions in less CPU time than the recently proposed Hopeld network.

Keywords