Insurance Reserve Intelligence Platform: PINN/KINN for term-life reserves (arXiv:2609.30765v1)
This arXiv preprint describes an "Insurance Reserve Intelligence Platform" that combines a classical Thiele-equation solver with a Physics-Informed Neural Network (PINN) augmented by Knowledge-Informed Neural Network (KINN) losses to model term-life insurance reserves. The model predicts a standardized reserve ratio using seven features, reports test-set performance (R2=0.9887, MAE=785.48, RMSE=1212.76), and claims PINN/KINN inference is ~119.53× faster than the classical solver on 200 policies while noting limitations in monotonicity and out-of-distribution generalization.