Lower Bounds on Adaptive Sensing for Matrix Recovery

We study lower bounds on adaptive sensing algorithms for recovering low rank matrices using linear measurements. Given an $n \times n$ matrix $A$, a general linear measurement $S(A)$, for an $n \times n$ matrix $S$, is just the inner product of $S$ and $A$, each treated as $n^2$-dimensional vectors. By performing as few linear measurements as possible on a rank-$r$ matrix $A$, we hope to construct a matrix $\hat{A}$ that satisfies $\|A - \hat{A}\|_F^2 \le c\|A\|_F^2$, for a small constant $c$. It is commonly assumed that when measuring $A$ with $S$, the response is corrupted with an independent Gaussian random variable of mean $0$ and variance $\sigma^2$. Cand\'es and Plan study non-adaptive algorithms for low rank matrix recovery using random linear measurements. At a certain noise level, it is known that their non-adaptive algorithms need to perform $\Omega(n^2)$ measurements, which amounts to reading the entire matrix. An important question is whether adaptivity helps in decreasing the overall number of measurements. We show that any adaptive algorithm that uses $k$ linear measurements in each round and outputs an approximation to the underlying matrix with probability $\ge 9/10$ must run for $t = \Omega(\log(n^2/k)/\log\log n)$ rounds showing that any adaptive algorithm which uses $n^{2-\beta}$ linear measurements in each round must run for $\Omega(\log n/\log\log n)$ rounds to compute a reconstruction with probability $\ge 9/10$. Hence any adaptive algorithm that has $o(\log n/\log\log n)$ rounds must use an overall $\Omega(n^2)$ linear measurements. Our techniques also readily extend to obtain lower bounds on adaptive algorithms for tensor recovery and obtain measurement-vs-rounds trade-off for many sensing problems in numerical linear algebra, such as spectral norm low rank approximation, Frobenius norm low rank approximation, singular vector approximation, and more.

Paper

Similar papers

Peer review

Reviewer Txfq7/10 · confidence 3/52023-06-27

Summary

Sparse recovery for vectors has been studied for a long time. It has been shown that allowing adaptive queries will give extra power to reduce the number of queries. We also have upper and lower bounds in this setting. More recently, sparse recovery for low-rank matrices has also been extensively studied. With non-adaptive queries, $\Omega(n^2)$ queries are necessary. With adaptive queries, based on the power method, we get an algorithm with $O(nr)$ linear measurements in each round over $O(\log n)$ adaptive rounds. It is an interesting question to see if the power method is optimal in this setting. This paper gives an affirmative answer to this question by showing a measurement-vs-rounds trade-off for recovering low-rank matrices using linear measurements. More specifically, let $A$ be an $n$-by-$n$ matrix with rank $r$. In each round, the algorithm can make $k$ queries. Each query is an $n$-by-$n$ matrix $S$, and the answer is $\langle S, A\rangle$ with some Gaussian noise. The goal is to reconstruct a matrix $\hat{A}$ such that $\|\hat{A}-A\|_F\leq c\|A\|_F$. The main result of this paper is that any adaptive algorithm which uses $n^{2-\beta}$ linear measurements in each round must run for $o(\log n/ \log \log n)$ rounds to compute a good reconstruction with high probability. Their techniques also apply to obtain measurements-vs-rounds trade-offs for other numerical linear algebra problems, including low-rank approximation in several different norms, singular vector approximation, etc. Technically, the construction of the hard instance is $A=\frac{\alpha}{\sqrt{n}}\sum_{i=1}^r u_iv_i^\top$, where $u_i,v_i$ are independent, Gaussian random vectors. In the proof of their main result, they first reduce to show the lower bound for a deterministic algorithm with perfect linear measurements of the random matrix $\frac{\alpha}{\sqrt{n}}\sum_{i=1}^r u_iv_i^\top+G$, where $G$ is a Gaussian matrix, outputting a reconstruction of $\frac{\alpha}{\sqrt{n}}\sum_{i=1}^r u_iv_i^\top$. The key observation is that the distribution of the responses in the first round is close to $N(0, I_k)$, and therefore the algorithm cannot have a lot of “information” about the target matrix. The proof relies on a random tensor concentration result and Bayes risk lower bounds.

Strengths

This paper fills a gap in the field of sparse recovery and makes significant progress in understanding the limitations of adaptive queries in solving numerical linear algebra problems. Their main result is the lower bound for the sparse recovery of low-rank matrices, nearly matching the upper bound via the power method. Even if they use some techniques in prior works, the proofs are still non-trivial. This paper is well-motivated, and the idea of their techniques and proofs are clearly presented. Most of the claims are sound to me.

Weaknesses

For the applications, it is difficult to judge the significance since there is not enough comparison between the results in this paper and prior works. It seems that some applications are for the regimes incomparable to the previous literature.

Questions

1. Does the lower bound require that all the iterations take $k$ queries uniformly? What if the algorithm can adaptively decide the number of queries in each iteration? 2. It would be better to show some known upper bounds for the problems in Table 1. 3. Line 52: $M\in \mathbb{R}^{n\times n} \rightarrow \mathbb{R}^t$. $t$ should be $k$.

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

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

3 good

Presentation

3 good

Contribution

3 good

Limitations

N/A.

Reviewer cbbb5/10 · confidence 2/52023-07-06

Summary

This paper focuses on investigating the lower bound of the adaptive low-rank matrix sensing problem. Specifically, the authors demonstrate that when the noise level significantly exceeds the signal, any adaptive algorithm requiring fewer than $o(\log(n)/\log\log n)$ rounds must utilize at least $\Omega(n^2)$ linear measurements in total. This finding highlights an intriguing trade-off between the number of measurements and the number of rounds in various matrix sensing problems within numerical linear algebra. The paper presents a clear message, and the theoretical results are robust and captivating. However, one aspect worth considering is the general interest in the noise level examined in this study.

Strengths

- The theoretical results are very sound - The presentation is very clean and easy to follow - The adaptive settings of matrix recovery is of general interest

Weaknesses

This paper examines the case where the noise level is assumed to be $O(1)$, while the signal of each entry is considered to be $O(1/\sqrt{N})$. It is worth noting that in a more typical scenario where both signals and noises are of magnitude $O(1)$, a single round with $\Omega(nr)$ measurements, as presented in [5], would suffice. This raises questions about the significance of studying this extreme case and the role that $\sigma_l$ plays in the lower-bound tradeoff. To provide more justifications for studying this extreme case, it would be beneficial for the authors to elaborate on the motivations and implications behind their choice of noise and signal levels. By offering insights into why this specific scenario is relevant and shedding light on the insights gained from this extreme case analysis, the authors can strengthen the value and message of their paper.

Questions

The authors make an assumption that $M^{(i)}$ represents an orthonormal basis and argue that this assumption does not affect the generality of their approach due to the possibility of a change of basis. However, it is important to consider whether this change of basis has any impact on the homogeneous noise level, i.e., $Ag$ for a Gaussian vector $g$ is not a homogeneous Gaussian vector anymore if $A$ is not unitary. It would be valuable for the authors to provide further elaboration on this point.

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

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

4 excellent

Presentation

4 excellent

Contribution

2 fair

Limitations

I do not forsee any potential negative societal impact of this work.

Reviewer m7zx6/10 · confidence 2/52023-07-06

Summary

In this paper, the authors discuss the power of adaptive algorithms in low-rank approximation. This is a setting where one observes general linear measurements of a matrix and wants to produce the best r-rank approximation of it. It is known that non-adaptive algorithms need order n^2 measurements and that with log n rounds a spectral algo works with nearly linear measurements. The authors discuss whether something can be done with o(log n) rounds. ## Contribution The authors prove that for an algo that works with o(log n/log log n) rounds, we need order n^{2-o(1)} measurements.

Strengths

The authors point to the interesting importance of having access to approx log n rounds, as they prove that having access to o(log n/log log n) rounds is like having one round in terms of measurement complexity. I find the contribution clean, correct, and interesting.

Weaknesses

It is a little bit disappointing that the authors cannot only assume o(log n) rounds, and need to assume o(log n/log log n) rounds. Could they comment more on that weakness?

Questions

It seems that for the lower bound the authors only assume that the target matrix is r-rank Gaussian spike plus Gaussian noise. Is that true? If so, please highlight it as this is a rather simple "bad" example to build a lower bound, which makes the result even more appealing.

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

2 fair

Presentation

2 fair

Contribution

3 good

Limitations

See above.

Reviewer NmzP7/10 · confidence 2/52023-07-07

Summary

This paper studies the problem of low-rank matrix reconstruction from linear measurements, which is a matrix generalization of the well known sparse reconstruction setting. The provide new lower bounds for this problem under different error metrics, such as the Frobenius norm, essentially showing that non-trivial reconstruction error is impossible unless $\Omega(\log n)$ adaptive rounds of measurements are made. The hard instance is a rank-1 matrix planted into i.i.d Gaussian noise.

Strengths

- The problem solved is fundamental and to the best of my knowledge the contributions are new and generally applicable. - The technical steps are clearly explained and well written, and look sound.

Weaknesses

- The presentation in Section 1 and 1.3 could be improved. While I like that the explanation goes into technical detail, it could be significantly improved by having an outline of the 2-3 most significant contributions and a more modular presentation.

Questions

Is there any potential connection between this work and the hardness results used in [1] for matrix completion based on planted clique (or the references within)? Are there any implications for matrix completion? [1] Yudong Chen, Incoherence-Optimal Matrix Completion, IEEE Transactions on Information Theory, 2015

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

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

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

NA

Reviewer Txfq2023-08-14

I thank the authors for their response. I keep my score.

Reviewer NmzP2023-08-18

I thank the authors for the response.

Reviewer m7zx2023-08-20

I thank the authors for their response and I maintain my score.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC