- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 3
Read review
Paper summary
This paper studies score estimation via Monte Carlo when the target distribution is specified by an unnormalized density , so that the score is estimated by sampling rather than by training a neural network. The authors propose an affine post-processing rule that combines score estimates at multiple signal levels to reduce the bias and variance of the base estimator. Experiments on Gaussian, Gaussian-mixture, and several other targets are conducted to verify the effectiveness of the proposed method.
Critical strengths
- Since the score is a smooth function with respect to the signal level , it is reasonable to use score estimation at different levels to construct a more accurate estimator.
- The optimal affine post-processor is analyzed theoretically, and an interesting observation is given: any unbiased estimator with non-zero variance is not optimal among affine post-processors.
Critical weaknesses
-
Equation (24) approximates the score as . The authors should explain more explicitly why such a decomposition exists. I understand that is a smooth function of , but it is not immediately clear why it can be expanded in a basis independent of , considering that is a vector. A brief justification would help.
-
The paper uses a fixed set of design levels , but the main text does not discuss how they are chosen or how sensitive the method is to this choice. The authors should clarify how to choose them in the main text and discuss the impact of this choice.
-
The paper provides a closed-form Wiener oracle and shows that unbiased estimation is not optimal in general. However, the practical Legendre ridge-regression post-processor is not accompanied by a theoretical guarantee that it reduces variance or MSE, nor by a finite-sample error bound. The current theory applies to the oracle rule, not to the practical one. The authors should provide such guarantees or related analyses clearly.
-
The experimental targets include many artificial distributions such as banana, ellipsoidal shell, and Gaussian mixtures. However, all targets are synthetic. It would be better to include at least one real-world application, such as a molecular energy function, to demonstrate practicality in high-dimensional settings.
-
Theorems 1 and 2 and Corollaries 1 and 2 would benefit from additional discussion around them.
Questions for authors
see weakness.