Towards Optimal Effective Resistance Estimation

We provide new algorithms and conditional hardness for the problem of estimating effective resistances in $n$-node $m$-edge undirected, expander graphs. We provide an $\widetilde{O}(m\epsilon^{-1})$-time algorithm that produces with high probability, an $\widetilde{O}(n\epsilon^{-1})$-bit sketch from which the effective resistance between any pair of nodes can be estimated, to $(1 \pm \epsilon)$-multiplicative accuracy, in $\widetilde{O}(1)$-time. Consequently, we obtain an $\widetilde{O}(m\epsilon^{-1})$-time algorithm for estimating the effective resistance of all edges in such graphs, improving (for sparse graphs) on the previous fastest runtimes of $\widetilde{O}(m\epsilon^{-3/2})$ [Chu et. al. 2018] and $\widetilde{O}(n^2\epsilon^{-1})$ [Jambulapati, Sidford, 2018] for general graphs and $\widetilde{O}(m + n\epsilon^{-2})$ for expanders [Li, Sachdeva 2022]. We complement this result by showing a conditional lower bound that a broad set of algorithms for computing such estimates of the effective resistances between all pairs of nodes require $\widetilde{\Omega}(n^2 \epsilon^{-1/2})$-time, improving upon the previous best such lower bound of $\widetilde{\Omega}(n^2 \epsilon^{-1/13})$ [Musco et. al. 2017]. Further, we leverage the tools underlying these results to obtain improved algorithms and conditional hardness for more general problems of sketching the pseudoinverse of positive semidefinite matrices and estimating functions of their eigenvalues.

Paper

References (55)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer UUSU6/10 · confidence 3/52023-06-29

Summary

The authors consider the problem of estimating effective resistance both for all edges of a graph and between any pair of nodes. This quantity is incredibly useful for sampling from a graph to produce a sparser graph with similar spectral properties. Furthermore, this quantity can provide a useful metric for the importance of an edge to the connectivity of the network. Within this context the authors restrict to a subclass, expander graphs, which have better connectivity properties. The practical application of these graphs is much more limited, but as the authors point out, they are commonly used as a first step towards giving more general algorithms. The authors provide a new algorithm that for sparse graphs gives an asymptotically faster algorithm than the previous work. In addition, the authors improve upon the lower bound runtime by reducing to the triangle detection problem.

Strengths

1. Improves the asymptotic runtime for an important problem on sparse graphs of a commonly studied subclass, expander graphs. 2. Applies a new clever factorization technique for expander graphs with substantial technical depth 3. Gives a new approach for lower bounding the runtime through a reduction to the triangle detection problem that extends to additional problems

Weaknesses

For a broad ML venue the practical application of this theoretical work is quite limited given the restriction to expander graphs. It makes good progress in a well-studied area in the theoretical community but may be more appropriate for a theory-centric venue. To be of more general interest, it would be preferable to see progress upon general graphs by either making asymptotic or practical improvements. I would very much welcome the authors rebuttal (see questions section) but it also feels like these techniques, specifically the asymmetric factorization bounding, which is very clever, could not be reasonably extended to general graphs.

Questions

What's the biggest bottleneck in your opinion for applying similar techniques to the ones you provide in this work to improve the runtime for general graphs?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

2 fair

Limitations

The authors adequately addressed the limitations

Reviewer AnJp6/10 · confidence 2/52023-07-04

Summary

The paper studies the (approximate) estimation of effective resistance on expander graphs, i.e., graphs with edge expansion $\Omega(1/\text{polylog}\ n)$. The main upper-bound result is a data structure (sketch) for effective resistance that takes $\tilde{O}(m/\epsilon)$ time to build, answers each query in $\tilde{O}(1)$ time, and is of size $\tilde{O}(m/\epsilon)$. This improves the Li and Sachdeva [SODA’23] on the construction time on sparse expanders with $O(n/\epsilon)$ edges. The data structure implies an $\tilde{O}(n^2/\epsilon)$-time algorithm to compute all-pairs effective resistance. To complement the upper bound result, the paper further includes a lower bound of $\tilde{\Omega}(n^2/\epsilon^{0.5})$ time for computing all-pairs effective resistance for all algorithms that are not based on fast matrix multiplications (FMM). The main technical ingredient for the upper bound result is to use asymmetric factorization of the effective resistance in expander graphs (in the form of $\ell_{1}$ norm) and apply the count-sketch in $\ell_{1}$ to achieve the $\epsilon^{-1}$ compression. For the lower bound side, the paper designs a novel reduction to the triangle detection problem, which improves the exponent on $\epsilon$ from $1/13$ to $1/2$. My general evaluation of the paper is positive. The paper seems to require a good deal of background to read, and I did not manage to check the correctness of the proofs. The high-level intuitions make sense to me, and conditioning on equation (2) is correct, I think everything goes through for the upper bound. As such, as far as I can see, the paper is technically solid, and it makes progress toward an interesting open problem. On the flip side, it seems the improvements of this paper are rather restrictive – compared to Li and Sachdeva [SODA’23], it is only better on expander with $O(n/\epsilon)$ edges. For the presentations, I think the paper generally makes good efforts to define notions and explain ideas clearly. However, I still believe that the writing assumes a very strong level of knowledge from the reader (see the weakness part for details).

Strengths

As I have mentioned in the general evaluation, I think the paper is technically solid, and the presentation is clear in general. The high-level intuitions provided in the paper are helpful to the readers, and the techniques include novel ideas of the symmetric factorization of the effective resistance.

Weaknesses

Most of them are already mentioned in the general evaluation. Here is a list that would help the authors to address: - On the upper bound side, the improvement compared to Li and Sachdeva [SODA’23] is rather limited – the bound is only better on expanders with $O(n/\epsilon)$ edges. - As I said in the general evaluation, the paper seems to require a good deal of background to read. I realize that the authors have already put efforts into making the paper more accessible by defining most of the notions clearly from the get-go and presenting the high-level ideas concisely. However, in my (very subjective) opinion, reading the paper still requires some non-trivial efforts. - I think some of the presentation problems can be attributed to unclear notations, for instance: 1. $\tilde{O}(\cdot)$ is defined after multiple usages in section 1, and $\tilde{\Omega}(\cdot)$ is never defined. This is especially confusing since I think you hide $\text{polylog}$ vs. $1/\text{polylog}$ factors in the two notations. 2. The edge-vertex incident matrix $B_{G}$ is briefly mentioned but never formally defined (I know that it is): this could be an issue for the general audience at NeurIPS to understand. 3. Is $J$ in line 194 the matrix for the JL projection? It is never mentioned. - Finally, for NeurIPS, it would be good if you could spend more passage talking about downstream applications and what your bound would imply therein.

Questions

Most of them are presentation-related issues, and some of them are comments - Line 63, do you mean $(\tilde{O}(m \epsilon^{-1}), \tilde{O}(1), \tilde{O}(n \epsilon^{-1}))$? (The current notation means sublinear time but linear size.) - Similarly, the order seems to be wrong in line 187 – should be $(\tilde{O}(m \epsilon^{-2}), \tilde{O}(\epsilon^{-2}), \tilde{O}(n \epsilon^{-2}))$ according to your notation. - Line 120: ‘Building on this work’ – what do you mean? Based on this manuscript? - Line 198: there are two ‘vectors’. - Formatting issue: For some reason, most of your citations do not contain the venue of publication; some of them have the venue but not the title – please fix this. - Minor formatting issue: adding *actual citations* in the abstract seems to be non-standard, i.e., I think it is ok to write as (Chu et al. [Arxiv’18]), but not putting the actual pointer there. - Did you state the (conditional) hardness result for triangle detection anywhere in the paper? I would recommend including it in the preliminary. - I do not quite understand the logic of ruling out FFM-based algorithms: from the statement of theorem 3, it seems the reduction between effective resistance and triangle detection is unconditional. So why the reduction does not work for FFM-based algorithms? Is it because of the hardness result of triangle detection?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

I do not see any immediate non-technical/societal limitations on the work. Presentation-wise, it would be good if the paper had discussed more downstream applications.

Reviewer 9w9Q7/10 · confidence 4/52023-07-05

Summary

In the paper, the authors study the problem of estimating the effective resistance between pairs of vertices in weighted, undirected expander graphs. The authors present a novel algorithm for effective resistance estimation and effective resistance sketching. Moreover, they give a conditional lower bound based on triangle detection.

Strengths

This is a very nice theory paper that studies the problem of effective resistance estimation which a lot of people in the relevant community care. The contributions of the paper are novel algorithms for both the lower bound and upper bound. The algorithm cleverly uses the countSketch to build a spectral sketch data structure and connects the Laplacian of expanders using Cheeger's inequality. The paper is very well written and it is a pleasure to read. In our opinion, this is an important contribution to the literature.

Weaknesses

For the upper bound, the result looks very similar to the result of [Li, Sachdeva 2022]. It would be great if the authors can provide a more close comparison between the two.

Questions

Minor: 1. Line 63: should be $(\widetilde{O}({m\epsilon^{-1}}),\widetilde{O}(1), \widetilde{O}({n\epsilon^{-1}}))$-effective 2. Line 192: the expression of $r_G(i,j)$ missing the $\delta_{i,j}$ term 3. Both Equation (1) and Equation (2) missing the $\delta_{i,j}$ term 4: Line 240: peridicular should be perpendicular?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

4 excellent

Limitations

N/A

Reviewer btut8/10 · confidence 4/52023-07-10

Summary

The paper theoretically analyzes the time complexity of estimating effective resistances in $n$-node $m$-edge undirected, expander graphs. Specifically, the paper present an algorithm which can achieve $\tilde{O}(m \varepsilon^{-1})$ time complexity for estimating the effective resistance of all edges in such graphs. This complexity bound significantly improves the previous result either for general graphs or for expanders. The paper also present a tight lower bound for computing the estimates of effective resistances.

Strengths

S1. Computing the effective resistance between a pair of graph nodes is an important and fundamental problem. The paper improves the best complexity result of existing work with non-trivial analysis. A tight lower bound is also provided in the paper. S2. The paper's technical part is sound and solid. S3. The paper is well-written and easy to follow.

Weaknesses

W1. The paper lacks experimental results to empirically demonstrate the effectiveness of the proposed algorithm.

Questions

The paper mainly focuses on undirected expander graphs. An interesting question is whether the complexity results given in the paper can be further extended to general graphs.

Rating

8: Strong Accept: Technically strong paper, with novel ideas, excellent impact on at least one area, or high-to-excellent impact on multiple areas, with excellent evaluation, resources, and reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

4 excellent

Presentation

4 excellent

Contribution

4 excellent

Limitations

I do not see any potential negative societal impact in this paper.

Reviewer Craf6/10 · confidence 4/52023-07-18

Summary

Given a graph $G$, the effective resistance approximation problem is one where you are asked to approximate the effective resistances of some subset $S$ of vertex pairs. The authors have two contributions: Firstly, the authors demonstrate an $O(m\epsilon^{-1} + |S|)$ algorithm for computing effective resistances in graphs with constant expansion. This matches the current best known algorithm for computing effective resistances in expanders for dense graphs, and beats it in sparse graphs of total edge count $o(n\epsilon^{-2})$. To achieve this, the authors make use of CountSketch, and demonstrate that there are sets of vectors from which the effective resistances can be extracted in $\tilde{O}(1)$ time. These vectors are obtained through the use of CountSketch on vectors with small $\ell_1$ norm. Furthermore, this data structure is $\tilde{O}(n\epsilon^{-1})$ sparse. Secondly, the authors demonstrate an $\Omega(n^2\epsilon^{-1/2})$ conditional lower bound on the time required to compute the effective resistance on all pairs of edges in a graph. They demonstrate that if an effective resistance approximation algorithm is faster than $O(n^2\epsilon^{-1/2})$ time by a polynomial factor, then the time complexity of triangle detection will be truly subcubic. Currently, the only subcubic algorithms for triangle detection require fast matrix multiplication, so it is unlikely that a combinatorial algorithm for the effective resistance problem on all pairs can be faster. Furthermore, they improve on the lower bounds on a range of similar spectral sum problems using this same reduction, by refining the analysis performed by [4]. They achieve this improvement through the use of several techniques. Firstly, by considering a different power series expansion as in [4], they demonstrate that the triangle counting problem can be reduced to a different problem that looks more like effective resistances. To obtain sufficiently good estimation errors, they also make use of symmetric random signing, where each entry of the adjacency matrix is flipped with probability 1/2.

Strengths

Originality: I really like the approaches used by the authors. I think that the technical contributions here in both the upper and lower bounds are very interesting to me. I like the use of the asymmetric CountSketch, and the fact that the vectors picked are small in $\ell_1$ norm. The usage of a different power series expansion to obtain a different reduction for the triangle detection problem is also very interesting to me. The former is novel to me, and the latter reads like a novel improvement to obtain better bounds than a preexisting argument. Quality/Clarity: The full paper is a bit of a rough read, which is the nature of theory papers, but I like that a lot of the intuition is explained in the main section, that makes it a lot easier to understand what is happening. I could follow/guess a lot of what was going to happen in the analysis just from reading the main section which had none of the analysis. Significance: The upper bound is relatively significant. Effective resistance approximation is certainly a very important problem that is used as a subroutine for many other problems. They have made an improvement to the approximation time complexity in expanders in sparse graphs, which is a significant contribution to this area. In terms of lower bounds, they have made a significant refinement of the previous approach, which results in an improvement on the lower bounds for a large number of different problems.

Weaknesses

The primary weakness of the upper bound is that the contribution made here is relatively marginal in scope. The paper has made improvements in solving a problem that is important, but have only done so in a limited setting (sparse expanders). The lower bound is also a very interesting refinement of the previous approach, but it is somewhat unclear to me why this lower bound specifically is one that is important to look at. I feel like there are several other natural lower bounds for effective resistances that are more natural to want to consider. This latter point might just be me since I am not very well versed in this area, but even though this reduction to triangle detection without the use of FMM seems pretty standard is used in many papers, and I understand the arguments as to why it is done, I can't really find myself getting behind it.

Questions

The lower bound obtained by the authors is one on the time required to compute the effective resistance on all pairs of edges. I wonder if there are any lower bounds on the time required to compute the effective resistance on just the edges of the graph. Looking through the use cases of effective resistances in the introduction. It seems to me that a lot of them only care about the effective resistance of just the edges in the graph, while maybe only the most recent works on over-squashing in GNNs care about the total effective resistance. I am also very confused about the implications of the lower bound with regards to the upper bound. The lower bound is $\Omega(n^2\epsilon^{-1/2})$, while the upper bound is $O(m + n\epsilon^{-2} + n^2)$. Even if the graph were not sparse, the algorithm takes $O(n\epsilon^{-2} + n^2)$ time, while the lower bound takes $\Omega(n^2\epsilon^{-1/2})$. If we pick $\epsilon$ to be small, say $\sqrt{n}^{-1}$, isn't the upper bound faster than the lower bound? I don't think I understand lines 82-84, where the authors claim that this lower bound holds on expanders.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

4 excellent

Presentation

4 excellent

Contribution

3 good

Limitations

Typos: Line 240: peridicular Line 521: resistance Line 536: reisistance

Reviewer T2X26/10 · confidence 3/52023-07-25

Summary

The paper studies the problem of estimating the effective resistances in $n$-node $m$-edge undirected, expander graphs, and provides a $O(m/\epsilon)$ time algorithm which produces an effective resistances sketch in $O(n/\epsilon)$ size of space. This improves the runtime of the previous algorithm that estimates the effective resistance of all edges in such graphs and can be extended to the general problems of sketching the pseudoinverse of positive semidefinite matrices and estimating functions of their eigenvalues. The authors also give a new corresponding conditional lower bound where the $\epsilon$ dependence is improved.

Strengths

1. The technical contribution is solid. The paper provides the algorithms with better theoretical bounds, along with the corresponding lower bound. To achieve this, one key point of the paper is it considers an asymmetric factorization and the asymmetric CountSketch in $\ell_1$. 2. The organization of this paper is good. The introduction section is detailed and the authors then first give a technical overview which can help the readers first get a brief understanding of the key point in the algorithm design.

Weaknesses

1. This paper is a pure theory paper. However, given that it is a NeurIPS submission, I think the experimental evaluations will make the paper stronger. 2. I feel like some parts of the paper are a little technically dense and not very easy to follow. Also, the paper uses many standard facts in spectral graph theory. I think it will make the presentation more clear if the authors can give more detail in the preliminaries section.

Questions

1.As summarized in table 1, [3] has a runtime $O(m + n/\epsilon^2)$, will this bound be better than this work if $m/n > 1/\epsilon$?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

See the above questions.

Reviewer UUSU2023-08-10

Thanks for all the detailed responses! I appreciate the authors pointing out that their asymmetric factorization bound does apply (just not as well) to general graphs which I had overlooked in my review. It's always difficult to conjecture how well these techniques could lead to further improvement for general graphs. But I'm curious if the authors have any opinion on the most likely avenue for achieving nearly linear time on general graphs (improving asymmetric factorization bound or applying expander decomposition techniques that they mention in the rebuttal). My intuition is still is that improving their asymmetric factorization bound will be more difficult than figuring out a different approach for general graphs. But regardless, the authors have several nice results that warrant acceptance in my opinion.

Authorsrebuttal2023-08-11

Thanks for the question! As you already mentioned (and we discussed in our general response), it is challenging to conjecture how to obtain the runtime for general graphs. (Though this is certainly an interesting open problem for future work.) It’s unclear how to improve the asymmetric factorization bound _directly_ and we are unsure whether a better direct bound will be possible. One approach for leveraging our results in future work is to use algorithms on expanders (such as ours, or others) as a subroutine in a general graph algorithm in order to get improved runtimes for effective resistance estimation, or at least better query times for sketching. This might include (1) the technique of repeatedly computing effective resistances in expander decompositions/partitions (like what is done [2] and [3]) and (2) a recursive Gaussian elimination approach like what is also used in [3]. However this is speculative and for future research.

Reviewer UUSU2023-08-14

Thanks for the response! I appreciate the authors willingness to speculate on future directions of their work, and their approaches mentioned seem reasonable. I especially appreciate the authors speculating upon improving the asymmetric factorization bound given that they have the best intuition upon this technique and the potential for extending it.

Reviewer Craf2023-08-11

This clarifies a lot of the questions I had, thank you very much!

Reviewer AnJp2023-08-13

Reponse

Thanks for responding to my questions. I think many clarifications in the responses are worth adding to the paper to help readers understand. Please take advantage of the extra page in the final version to include these. Also, it'll be helpful to add some aspects of these discussions to a full version of your paper.

Authorsrebuttal2023-08-16

Thank you for the feedback! We indeed plan to incorporate many of the clarifications discussed in the author-reviewer discussions in the next iteration of the submission. Thanks!

Reviewer btut2023-08-14

Response to the Authors' Rebuttal

I am very glad to hear that the authors are considering to introduce some empirical experiments in the future. The extension to general graphs are also highly expected. Looking forward to seeing these future work. Thanks!

Authorsrebuttal2023-08-16

Thank you for the response and we are glad to hear that you share our excitement for this line of research! To be clear, we think that future work on empirical experiments or general graphs may benefit from our submission, but are outside its scope. Thanks!

Reviewer T2X22023-08-18

Thanks for the detailed response, it addresses most of my concerns!

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC