Scissorhands is a novel machine unlearning approach that effectively removes the influence of data from trained models. The method first identifies the most relevant parameters in the model relative to the data to be forgotten using connection sensitivity. These parameters are then reinitialized to reduce the model's memory of the forgotten data. Subsequently, the model is fine-tuned using a gradient projection-based approach to preserve information on the remaining data while discarding information related to the forgotten data. Experimental results on image classification and generation tasks show that Scissorhands performs competitively compared to existing methods. The approach is effective in erasing the influence of random samples, discrete classes, and sensitive content such as nudity. The method addresses the challenge of unlearning by balancing the need to remove the influence of forgotten data while maintaining the model's utility on the remaining data. Scissorhands uses a gradient projection-based optimization algorithm to achieve this balance, ensuring that the model retains its performance on the remaining data while effectively forgetting the influence of the forgotten data. The algorithm is evaluated on various datasets, including SVHN, CIFAR-10, CIFAR-100, and CelebAMask-HQ, demonstrating its effectiveness in both classification and generation tasks. The method also shows promising results in mitigating the generation of inappropriate content in diffusion models. Scissorhands is a practical and effective machine unlearning algorithm that can be applied to a wide range of tasks, including image classification and generation. The approach is designed to handle the challenges of unlearning in different domains, including image generation, and provides a balance between erasing the influence of forgotten data and maintaining the model's utility on the remaining data. The algorithm is evaluated on various datasets and shows superior performance compared to existing methods. The method is also effective in reducing the amount of nudity content generated by diffusion models. The approach is supported by extensive experiments and evaluations, demonstrating its effectiveness in various scenarios. The algorithm is designed to handle the challenges of unlearning in different domains, including image generation, and provides a balance between erasing the influence of forgotten data and maintaining the model's utility on the remaining data. The method is evaluated on various datasets and shows superior performance compared to existing methods.Scissorhands is a novel machine unlearning approach that effectively removes the influence of data from trained models. The method first identifies the most relevant parameters in the model relative to the data to be forgotten using connection sensitivity. These parameters are then reinitialized to reduce the model's memory of the forgotten data. Subsequently, the model is fine-tuned using a gradient projection-based approach to preserve information on the remaining data while discarding information related to the forgotten data. Experimental results on image classification and generation tasks show that Scissorhands performs competitively compared to existing methods. The approach is effective in erasing the influence of random samples, discrete classes, and sensitive content such as nudity. The method addresses the challenge of unlearning by balancing the need to remove the influence of forgotten data while maintaining the model's utility on the remaining data. Scissorhands uses a gradient projection-based optimization algorithm to achieve this balance, ensuring that the model retains its performance on the remaining data while effectively forgetting the influence of the forgotten data. The algorithm is evaluated on various datasets, including SVHN, CIFAR-10, CIFAR-100, and CelebAMask-HQ, demonstrating its effectiveness in both classification and generation tasks. The method also shows promising results in mitigating the generation of inappropriate content in diffusion models. Scissorhands is a practical and effective machine unlearning algorithm that can be applied to a wide range of tasks, including image classification and generation. The approach is designed to handle the challenges of unlearning in different domains, including image generation, and provides a balance between erasing the influence of forgotten data and maintaining the model's utility on the remaining data. The algorithm is evaluated on various datasets and shows superior performance compared to existing methods. The method is also effective in reducing the amount of nudity content generated by diffusion models. The approach is supported by extensive experiments and evaluations, demonstrating its effectiveness in various scenarios. The algorithm is designed to handle the challenges of unlearning in different domains, including image generation, and provides a balance between erasing the influence of forgotten data and maintaining the model's utility on the remaining data. The method is evaluated on various datasets and shows superior performance compared to existing methods.