LMEraser: A Novel Machine Unlearning Method for Large Models Ensuring Privacy and Efficiency - MarkTechPost
Large models like BERT, GPT-3, and T5 boast billions of parameters and extensive training data, enabling them to discern intricate patterns and yield high accuracy. However, their widespread use raises privacy concerns regarding the unauthorized exposure of sensitive user information. Machine unlearning emerges as a solution, allowing for removing specific data from trained models without complete retraining. Yet, existing unlearning methods designed for smaller models need help with the complexities of larger models, facing challenges in pinpointing data influence, coping with computational demands, and maintaining overall performance amid data removal. IEEE researchers have developed LMEraser, an efficient unlearning method for
