- Rating
- 1
- Confidence
- 4
- Informativeness
- 1
- Presentation
- 1
- Soundness
- 1
Read review
Paper summary
The paper considers the chromatic correlation clustering problem. It suggests a rounding technique for the standard LP relaxation, but argues that it does not provide any improvement on approximation ratios, even over those using the standard LP relaxation. The paper then introduces more constraints to the standard LP relaxation, but it does not give any further improvement, either.
Critical strengths
None
Critical weaknesses
It’s hard for me to say the paper gives some interesting results. It just gives some attempts to resolve some cases, but these attempts do not generalize. Moreover, I believe the paper would benefit a lot from revision. I have to say that it was very complicated for me to understand the paper, and I suspect most researchers with proper background may also find the paper difficult. Please read Questions For Authors, where I list comments and concerns. Considering these points, I don’t think the paper would attract enough interest from the audience of the conference.
Questions for authors
- I don’t think the paper is easy to follow because it uses many specific terms without proper definitions, which I don’t feel are standard enough. For example, even in the first four sentences of the abstract, I had little sense of the meaning of the following terms: “vertex-pair LP”, “cross-color pairwise packing”, “selected-color admission marginal”, “positive affinity”, and “claim-respecting projection of independent claims”. The same tendency happens throughout the whole manuscript, but I don’t see it is worth pointing out every single place of this issue I found here.
- In Line 51, the paper suddenly drops , where , while it never defined what means.
- The description of Algorithm 1 looks misleading. If we follow what is written there, then we set only when is sampled in Line 5, draw in Line 8, and never consider the pair again since is certainly removed from after this point. Then, it makes little sense how we “pack” , because we visit only once throughout the entire execution of the algorithm.
- In Line 181, the paper says the scheme is always feasible (which I understand well-defined) because , but isn’t it simply because we remove at least the pivot every iteration?
- Isn’t Lemma 4.2 satisfied only when we have more condition that is the pivot and still remains in ?
- In Proposition 4.3, where is the definition of an “independent color-wise claim ”?
- In Line 245, “neutral”?
- In Line 259, “endpoint-first charge”?
- In Lemma 6.5, “canonical marginals”?
- In Figure 2, ?
- In Line 351, swap the last two entries of ?