| 摘要 |
A method and apparatus for providing learning capability to processing device, such as a computer game, is provided. One of a plurality of computer actions to be performed on the computer-based device is selected. In the case of a computer game, the computer actions can take the form of moves taken by a computer-manipulated object. A user input indicative of a user action, such as a move by a user-manipulated object, is received. An outcome value of the selected computer action is determined based on the user action. For example, in the case of a computer game, an intersection between the computer-manipulated object and the user-manipulated object may generate an outcome value indicative of a failure, whereas the non-intersection therebetween may generate an outcome value indicative of a success. An action probability distribution that includes probability values corresponding to said plurality of computer actions is updated based on the determined outcome value. The next computer action will be selected based on this updated action probability distribution. For example, the probability value of the last computer action taken can be increased if the outcome value represents a success, thereby increasing the chance that such computer action will be selected in the future. In contrast, the probability value of the last computer action taken can be decreased if the outcome value represents a failure, thereby decreasing the chance that such computer action will be selected in the future. In this manner, the computer-based device learns the strategy of the user. This learning is directed to achieve one or more objectives of the processing device. For example, in the case of a computer game, the objective may be to match the skill level of the player with that of the game. |
