Read this blog by Yang Song, who first proposed deep generative models based on score matching. You can skip the section on Stochastic Differential Equations, but note that towards the bottom, there is a subsection “Controllable generation for inverse problem solving” which can be interesting.
Unfortunately, the blog skips details on how to actually implement tractable score matching objectives, which are needed for practical implementations. See the Bishop book, 20.3 for a high-level overview of those details. Again, you can skip the part on differential equations.
In case you are interested, more details can be found in Murphy’s book: Section 24.3 introduces denoising and sliced score matching (skip the relation to contrastive divergence section), while section 25.3 gives the (surprisingly simple) objectives used in practice.
Finally, if you want to get the full picture, these are the classic papers by Yang Song: