Deep Neural Network models have been developed in different fields bringing many advances in several tasks. However, they have also started to be incorporated into tasks with critical risk. That worries researchers who have been interested in studying possible attacks on these models, discovering a long list of threats from which every model should be defended.

The weights modification attack is presented and discussed among researchers who have presented several versions and analyses about such a threat. It focuses on detecting the vulnerable weight to modify them, misclassifying the desired input data. Therefore, analyzing the different approaches of this attack can help to understand more precisely how to defend such vulnerabilities.
In this work, a new version of the weight modification attack is presented. That approach is based on three processes: input data clusterization, weight selection, and the modification of the weights. The data clusterization allows attacking the model more precisely. The weight selection uses the gradient given by the input data to know the desired parameters. The modification is incorporated little by little via reduced noise.