- Rating
- 1
- Confidence
- 1
- Informativeness
- 1
- Presentation
- 1
- Soundness
- 2
Read review
Paper summary
This paper proposes PALE, a method that employs neighborhood-averaging predictions based on features excluded via masking, utilizing a doubly stochastic kernel, to learn feature-weighted distances for Laplacian Eigenmaps, incorporating ridge regularization into the prediction loss. After selecting features through thresholding, the method constructs spectral coordinates. The paper first demonstrates that the standard eigenmap loss can approach zero on nearly disconnected graphs, indicating that this loss is unsuitable as an objective for learning affinities. Subsequently, under an "independent nuisance" model, the authors prove that the population criterion drives nuisance weights toward their lower bounds. They also provide results concerning metric estimation, support recovery, and spectral convergence for a fixed dimensionality (allowing the same set of samples to be used for both distance estimation and graph construction). Empirical validation includes simulations on surfaces, tori, and spheres, as well as analyses of two molecular datasets: CITE-seq and PBMC data.
Critical strengths
-
The results of Theorem 1 are clean and convincing: neither bi-random normalization, positive off-diagonal elements, nor forward entropy can prevent the eigenmap loss from going to zero. This provides a strict counterexample to "small spectral loss does not represent an informative representation". It is the most valuable conceptual contribution of the entire paper and also provides a real motivation for the predictive criterion.
-
Consistent spectral convergence (Theorem 19 + Corollary 1) that allows simultaneous estimation of distance and composition for the same observation is technically non-trivial. Consistent coverage of a compact metric family avoids sample splitting. This "simultaneity" processing is rare in similar work.
-
The optimization certification mechanism (branch-and-bound global target gap certificate, explicit accounting of numerical errors) rarely combines statistical theory with verifiable calculations. It is more rigorous than the usual practice of only reporting optimizer convergence information, and the reproducibility statement is also more complete.
Critical weaknesses
-
The core theory holds only under a fixed feature dimension : the primary convergence and support recovery results require to be fixed, with . In contrast, the simulations (such as the spherical experiment with and ) and the real-world data applications are inherently high-dimensional.
-
The theoretical constants are vacuous for practical sample sizes: "Condition 1" requires bounded measurements and a value that is bounded away from zero, with a kernel lower bound of . The authors themselves note that for surface/torus designs, evaluates to or . Consequently, all error bounds are scaled by constants on the order of .
-
Empirical gains are marginal and inconsistent: The analysis includes only two real-world datasets, where the improvement in paired MSE over "isotropic LE" is negligible (0.0023). Moreover, as the authors themselves admit, "MLKR spectral" consistently outperforms PALE across both datasets and both dimensions.
Questions for authors
-
In the spherical experiment with and , how are the block size and the number of blocks chosen? Do they satisfy the condition as well as the growth conditions required by Theorem 8? If the experimental setup does not meet these theoretical prerequisites, do the paper’s guarantees regarding spectral convergence still apply?
-
Given that MLKR consistently yields lower errors on the CITE-seq/PBMC dataset, what is the practical advantage of PALE over MLKR when the response view is available? Is there any empirical evidence from real-world datasets where PALE significantly outperforms other methods in a label-free setting?
-
What are the typical number of iterations and the computational runtime required for the branch-and-bound procedure to obtain ? Does this approach scale effectively with ?
-
Since the threshold and the eigengap are interdependent and both require tuning, how can they be selected in a stable manner in practice? Given that spectral error is amplified by a factor of , how do the authors address cases involving real-world data where is small?
2 comments
review
Thank you for your rebuttal.
Given that the paper is extremely long, and I share the same concerns as reviewer HTYb, I have lowered my confidence score to 1; therefore, I am inclined to maintain my current rating.
Response to Reviewer QkGd (2/2)
Thank you for this question, and we apologize that our block scaling was unclear. For b=2N, R=N/2, both growth conditions hold at fixed m=192 as N→∞. Spectral bounds do not certify N=4096.