The Cyclic-SOM has a high recognition rate for facial features, as shown in Figure 5. itsIts accuracy was 96.4179 % for Faceface recognition through pose and illumination. Before adjusting the learning rate to cyclic learning, we experimented on the SOM by changing only the way of finding the propinquity between the input and the feature map, or what's known as “the way of finding the winner neuron”. Its accuracy reached 95.9398 %. It used a learning rate of 0.95, which decreased over time in a fixed pattern until it reached 0 at the end of the training period, as in the regular SOM. This decrease equals 0.95 by the number of training periods. When we used the cyclic learning with the SOM which used MAD, we used a maximum learning rate equal to 0.95 and a minimum learning rate equal to 0.57 with length steps equal to 4 to reach this accuracy with the facesfaces' database. Adding cyclic learning after the MAD improves the accuracy and speed of the proposed technique, as shown in figureFigure 6. Adjustments in the learning rate and propinquity calculation enable the method to achieve full accuracy after 113 training periods. However, in the case of other techniques, a longer training duration of 400 times is needed, as opposed to the regular SOM, which has only 89 % accuracy in this state of the experiments. The identification rate was 59.7% when using the PCA, 83.54 % when using the MLP, and 60.1 % when using the SVM. Based on the previous results, we can conclude that the proposed technique responds to the complexities of feature data in faces under various illuminationilluminations and poses with robust performance when used for object recognition.