The reliability of artificial intelligence hinges on the integrity of its training data, a foundation often compromised by noise and corruption. Here, through a comparative study of classical and quantum neural networks on both classical and quantum data, we reveal a fundamental difference in their response to data corruption. We find that classical models exhibit brittle memorization, leading to a failure in generalization. In contrast, quantum models demonstrate superior resilience, underscored by a phase transition-like response to increasing noise, revealing a critical point beyond which the model’s performance changes qualitatively. We further introduce a framework of quantum machine unlearning, the process of efficiently forcing a trained model to forget bad influences. We show that classical models form rigid, stubborn memories of erroneous data, while the quantum model is significantly more amenable to efficient forgetting with approximate unlearning methods. Our findings establish that quantum machine learning possesses the dual advantage of intrinsic resilience and efficient adaptability, providing a promising paradigm for the trustworthy and reliable artificial intelligence of the future. This study demonstrates that quantum machine learning models exhibit inherent resilience to data poisoning and a unique capacity for efficient unlearning compared to classical machine learning.

Nature is a scientific journal publishing research articles, reviews, and commentary across a wide range of scientific disciplines. Readers can find articles on topics such as biology, chemistry, physics, and environmental science. Additionally, they can learn about  research, scientific discoveries, and advances in understanding the natural world.

Nature

A comparative study of classical and quantum neural networks reveals that quantum models are fundamentally more resilient to data poisoning attacks than classical models. Classical models exhibit brittle memorization and fail to generalize under data corruption, while quantum models show a phase transition-like response with a critical noise threshold. The research also introduces a quantum machine unlearning framework, demonstrating that quantum models can efficiently forget corrupted training data using approximate unlearning methods, whereas classical models form rigid memories that are hard to erase. These findings suggest quantum machine learning offers dual advantages of intrinsic resilience and efficient adaptability for trustworthy AI.

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