Analytical methods can predict photon transport in turbid media quickly and with low computational cost, making them attractive for real‑time and inverse problems. However, their accuracy is limited by the approximations required for tractable solutions. In contrast, Monte Carlo (MC) simulation is regarded as the gold standard for modeling light transport, but its high computational expense hampers use in applications that need repeated forward evaluations.\ \ We introduce a physics‑guided residual learning framework that combines the efficiency of analytical models with the accuracy of MC simulation. Using the diffusion approximation (DA) as a representative case, DA exhibits systematic errors under relatively strong absorption, weak scattering, or imperfect boundary treatment. Given the absorption coefficient, scattering coefficient, and anisotropy factor, the DA prediction is incorporated as a physical prior, and the network learns only the residual between this analytical prediction and the MC reference.\ \ Across the evaluated optical‑property range, the proposed framework reduces scalar‑reflectance RMSE by roughly a factor of 62 and spatially resolved reflectance RMSLE by about a factor of 14 compared to the raw diffusion approximation. With only 1% of the available training data, it achieves approximately 87% and 93% lower errors respectively than a direct neural‑network surrogate. These results demonstrate that learning corrections to an analytical physical baseline improves both prediction accuracy and data efficiency.\ \ Review: The study provides a compelling example of embedding physical priors into deep learning, offering a practical route to achieve fast yet accurate photon transport predictions for computationally demanding simulations.