Trainable Photonic Measurement for Physics-Informed PDE Learning
Researchers have developed a novel trainable photonic measurement technique for learning physics-informed partial differential equations (PDEs). This approach utilizes a photonic quantum neural field where optical phases represent coordinates, mixed through multi-photon interference, and decoded via photon-number measurements. The photonic circuit itself is optimized as the representation, minimizing the physics-informed residual. This method demonstrated superior accuracy, achieving up to an order of magnitude lower errors with fewer trainable parameters than classical baselines across various PDE benchmarks, particularly in complex regimes where residual derivatives amplify phase mismatch. AI
IMPACT Introduces a novel representation-learning technique for scientific machine learning, potentially improving the accuracy of solving complex differential equations.