Diagonal Observables Reduce Quantum Neural Network Complexity
Researchers propose Diagonal Adaptive Non-local Observables (DANOs) for quantum neural networks, reducing parameter and measurement complexity while maintaining full expressivity. Standard ANOs expand the function space of Variational Quantum Algorithms by making observables dynamic, but increase classical optimization cost. DANOs restrict observables to diagonal form, cutting k-local observable complexity from O(4^k) to O(2^k) and lowering measurement-side computation. The approach is mathematically equivalent to full ANOs modulo unitary similarity. The work appears on arXiv as 2605.15410.
Key facts
- Adaptive Non-local Observables (ANOs) enlarge function space of Variational Quantum Algorithms.
- ANOs shift hardware demands from circuit synthesis to measurement design.
- Diagonal ANOs reduce parameter count and classical optimization cost.
- Diagonal matrices are canonical representatives of ANO space modulo unitary similarity.
- Complexity reduction from O(4^k) to O(2^k) for k-local observables.
- Measurement-side classical computation is lowered.
- Proposed method retains full ANO capability.
- Paper published on arXiv with ID 2605.15410.
Entities
Institutions
- arXiv