SD-KDE uses a one-step score-based shift before standard KDE, improving the asymptotic error rate when the score is estimated accurately.
I summarize the theory, the bias/variance expansion, and the experimental results in the companion post: Score-Debiased Kernel Density Estimation.