Machine Learning-Optimized Construction of Stable Anti-Coherent States for Quantum Sensing.

Highly entangled states like the GHZ are valuable to metrology because they allow measurements at the fundamental Heisenberg limit, but are only sensitive to external fields oriented along a fixed, known sensing axis nhat = zhat. This limitation can be circumvented by using anticoherent states, which allow for near-Heisenberg limited sensing along arbitrary -- and apriori unknown -- axes nhat. Unfortunately, both the GHZ and the anticoherent states are extremely difficult to prepare due to noise and decoherence present in near-term experimental platforms. In this work we tackle this problem using cavity QED quantum simulations implemented in QuTiP and PIQS to model multi-qubit systems under realistic decoherence conditions. We develop machine learning architectures that systematically explore the space of rotation and one-axis twisting operations to discover optimal control sequences for anti-coherent state preparation. This work establishes a framework for ML-guided discovery of noise-resilient quantum states and control protocols, with applications to quantum sensing, metrology, and fault-tolerant quantum computing.