Q-Capsule: A Localized Capsule-Based Quantum Neural Architecture for Barren Plateau Mitigation
Q-Capsule is a quantum neural network architecture designed to overcome barren plateaus, where gradients vanish during training. It uses localized capsules, register partitioning, and adaptive depth growth guided by Quantum Fisher Information Matrix (QFIM) to maintain stable gradients and high efficiency. This approach results in superior performance compared to globally entangling models while requiring fewer gates.