Effective Dimension Governs Generalization in Quantum Kernel Vision Models
Researchers have identified a key metric, the effective dimension ($d_{\rm eff}$), that governs generalization in quantum kernel vision models. This metric explains why models with more entanglement or even added quantum noise can sometimes achieve better test accuracy. The study shows that by manipulating $d_{\rm eff}$, entanglement structure and quantum noise can be controlled to act as regularization, improving model performance. AI
IMPACT Introduces a unifying principle for designing quantum vision models by controlling effective dimension.
- arXiv
- Amplitude damping channel
- Quantum Kernel Vision Models
- Quantum Vision Transformers
- Quantum Convolutional Networks
- effective dimension
- Quantum Feature Kernel
- Quantum Feature Map
- Kernel classifier construction using orthogonal forward selection and boosting with Fisher ratio class separability measure
- Depolarized Kernel
- Kernel Machine Capacity Bound
- Capacity/Alignment Risk Decomposition