Quantum Masked Autoencoders Outperform Classical Models in Image Learning
Researchers propose quantum masked autoencoders (QMAEs), a novel architecture that extends classical masked autoencoders to quantum states. Unlike classical autoencoders that learn features from masked input data, QMAEs operate on quantum embeddings, enabling them to reconstruct missing image features with higher visual fidelity. In experiments on MNIST-family datasets, QMAEs achieved a 12.86% average improvement in classification accuracy over state-of-the-art quantum models. The work, detailed in arXiv:2511.17372, marks the first design and implementation of masked autoencoders in the quantum computing domain.
Key facts
- QMAEs are the first quantum masked autoencoders.
- They operate on quantum states instead of classical embeddings.
- Tested on MNIST-family images.
- 12.86% average improvement in classification accuracy over state-of-the-art quantum models.
- Paper available on arXiv with ID 2511.17372.
- Classical masked autoencoders inspired the architecture.
- QMAEs reconstruct masked input images with improved visual fidelity.
- The work was announced as a replace-cross type on arXiv.
Entities
Institutions
- arXiv