Surveying Techniques for Machine Unlearning in Deep Neural Networks and Federated Learning Environments

With over 90% of the world's data having been generated in the last two years alone, data privacy and regulatory compliance concerns have increased. Selective forgetting of training data by machine unlearning has emerged as a primary solution for data regulation like GDPR's "Right to be Forgotten." This survey mentions prominent deep learning techniques and federated systems based on more than 75 recent publications across various domains. We categorize exact and approximate unlearning methods, their time complexity, and their application to federated environments, where over 60% of today's deployments already occur. We further address existing benchmarks, privacy solutions, and open research questions, charting the path to future progress in crafting explainable, auditability, and responsible AI systems.