Fully Connected Quantum Boltzmann Machine via Bilevel Optimization
A recent study advances the quantum approximate optimization algorithm (QAOA) by integrating it into a bilevel optimization framework, introducing a fully connected quantum Boltzmann machine (QBM). The training process consists of an inner loop that focuses on minimizing positive phase energy, while the outer loop is dedicated to learning negative phase contrastive divergence by fine-tuning the structural parameters of the target Hamiltonian. With just one QAOA layer (p=1), the model demonstrates exceptional performance, achieving an average probability of 0.9559 for accurately measuring the target quantum state in ideal conditions. Additionally, it shows impressive resilience to noise, even under the typical noise levels found in contemporary commercial quantum computers.
Key facts
- Extends conventional QAOA circuit to bilevel optimization architecture
- Proposes fully connected QBM
- Inner-loop: positive phase energy minimization
- Outer-loop: negative phase contrastive divergence learning
- Single layer (p=1) achieves average probability 0.9559
- Notable noise robustness demonstrated
- arXiv:2605.07473v1
- Published on arXiv
Entities
Institutions
- arXiv