Superior resilience to poisoning and amenability to unlearning in quantum machine learning

The remarkable success of artificial intelligence (AI) has placed it as a transformative force in science and society [1-3], yet the reliability of these powerful models is fundamentally contingent on the integrity of their training data. In real-world applications, datasets are often compromised by corruption, such as mislabeled examples or malicious poisoning attacks, which can severely impair a model's generalization, introduce dangerous biases, and create significant security vulnerabilities [4-7]. A closely related and increasingly urgent challenge is that of machine unlearning: the need to efficiently remove the influence of specific data from a trained model to comply with privacy regulations or to correct for erroneous information [8-14]. While retraining a model from scratch on a sanitized dataset offers a definitive solution, its prohibitive computational cost for large-scale systems makes it impractical, driving the search for more efficient alternatives. As the field of quantum computing evolves rapidly [15-17], quantum machine learning (QML) has emerged as a promising new paradigm with the potential to solve problems intractable for classical computers [18-29]. These models, often implemented as quantum neural networks (QNNs), leverage principles like superposition and entanglement to navigate vast computational spaces [30-47]. However, despite rapid theoretical and experimental progress, the behavior of quantum models in the face of real-world data imperfections remains a critical and largely unexplored frontier.