Efficient Sign-Based Optimization: Accelerating Convergence via Variance Reduction

Sign stochastic gradient descent (signSGD) is a communication-efficient method that transmits only the sign of stochastic gradients for parameter updating. Existing literature has demonstrated that signSGD can achieve a convergence rate of $\mathcal{O}(d^{1/2}T^{-1/4})$, where $d$ represents the dimension and $T$ is the iteration number. In this paper, we improve this convergence rate to $\mathcal{O}(d^{1/2}T^{-1/3})$ by introducing the Sign-based Stochastic Variance Reduction (SSVR) method, which employs variance reduction estimators to track gradients and leverages their signs to update. For finite-sum problems, our method can be further enhanced to achieve a convergence rate of $\mathcal{O}(m^{1/4}d^{1/2}T^{-1/2})$, where $m$ denotes the number of component functions. Furthermore, we investigate the heterogeneous majority vote in distributed settings and introduce two novel algorithms that attain improved convergence rates of $\mathcal{O}(d^{1/2}T^{-1/2} + dn^{-1/2})$ and $\mathcal{O}(d^{1/4}T^{-1/4})$ respectively, outperforming the previous results of $\mathcal{O}(dT^{-1/4} + dn^{-1/2})$ and $\mathcal{O}(d^{3/8}T^{-1/8})$, where $n$ represents the number of nodes. Numerical experiments across different tasks validate the effectiveness of our proposed methods.

Paper

References (43)

Scroll for more · 31 remaining

Similar papers

Peer review

Reviewer XovB7/10 · confidence 5/52024-06-28

Summary

This paper introduces the Sign-based Stochastic Variance Reduction (SSVR) algorithm, which enhances the convergence rate of the traditional signSGD method. By incorporating variance reduction techniques with sign-based updates, the authors achieve a convergence rate of $O(d^{1/2}T^{-1/3})$ for general stochastic optimization and $O(m^{1/4}d^{1/2}T^{-1/2})$ for finite-sum problems. Additionally, the paper proposes novel algorithms for distributed environments by introducing the unbiased sign operation, resulting in superior convergence rates for heterogeneous data. Numerical experiments further validate the effectiveness of the proposed methods.

Strengths

1. The proposed methods improve the convergence rates over traditional signSGD methods and their variants, achieving faster convergence rates for general non-convex optimization and finite-sum optimization. 2. The SSVR-MV algorithm developed in the paper is communication-efficient and well-suited for distributed settings. The obtained convergence rates significantly enhance previous results in heterogeneous settings. 3. Numerical experiments validate the effectiveness of the proposed methods, demonstrating superior performance compared to existing sign-based optimization methods in terms of convergence speed and accuracy.

Weaknesses

1. Given that signSGD methods typically require large batch sizes to ensure convergence, the authors should include more experimental results to demonstrate the dependency on batch size for the proposed methods. This would clarify whether the proposed method also necessitates large batches for convergence in practice. 2. The authors introduce the stochastic unbiased sign operation in Definition 1, which differs from the traditional sign operation. It would be beneficial to provide a more detailed explanation of their differences and specify when the sign operation is preferred. 3. There are some typos in the paper, as listed below: - Page 2, Lines 44 and 45: $T^{1/2}$ should be $T^{-1/2}$. - Page 3, Line 98: "signed-based" should be "sign-based". - Page 5, Algorithm 2, Step 4: "set $t=\tau$" should be "set $\tau=t$". - Page 9, Figure 2: "SSVR" should be "SSVR-MV". - Page 28, Line 482: $f(x_t;\xi_t^j)$ should be $\nabla f(x_t;\xi_t^j)$.

Questions

1. Can you provide a more detailed discussion about the design of equation (3), especially the error correction term? The authors claim that "This gap can be effectively mitigated by the error correction term we introduced in equation (3)," but it remains unclear how it works based solely on the content in the main body. 2. Could you provide additional experimental results to demonstrate the performance of the proposed SSVR method with different batch sizes?

Rating

7

Confidence

5

Soundness

3

Presentation

3

Contribution

3

Limitations

The paper is theoretical and does not present potential negative societal impacts.

Reviewer Huhp7/10 · confidence 4/52024-07-11

Summary

The paper introduces an enhanced sign-based method called SSVR, designed to improve the convergence rate of signSGD with variance reduction techniques. SSVR achieves an improved convergence rate of $O(d^{1/2}T^{-1/3})$ for stochastic non-convex functions. For non-convex finite-sum optimization problems, the SSVR-FS method demonstrates an improved convergence rate of $ O(m^{1/4}d^{1/2}T^{-1/2}) $. To address heterogeneous data in distributed settings, the authors present novel algorithms that outperform existing methods in terms of convergence rates. The effectiveness of these proposed methods is validated through numerical experiments.

Strengths

1. The proposed methods improves upon existing algorithms in complexities. The sign-based algorithms have broad applications in the ML community and may be of great interest in the field. 2. The proposed sign-based variance-reduced estimator and stochastic unbiased sign operation are novel and efficient. They effectively reduce the gradient estimation error and are communication efficient, particularly for heterogeneous data in distributed settings. 3. The experimental results on various datasets validate the effectiveness of the proposed methods. 4. The paper is well-written, with clear problem settings and contributions. The proofs are also easy to follow.

Weaknesses

1. The authors introduce the bounded gradient assumption (Assumption 4) for distributed settings. Why is this assumption necessary? Is it commonly used in previous literature for this setting? 2. The description of Algorithm 2 is not very rigorous. The variable $\tau$ in step 4 and $\tau$ in step 10 seems different but used the same notation. 3. Some typos: - Line 98: "signed-based" --> "sign-based." - Line 247: "Assumption 5′" --> "Assumption 5" or "Assumption 4′." - Line 487: "$S(v_t^j)$" --> "$S_{R}(v_t^j)$."

Questions

1. Does previous work also require similar bounded gradient assumptions for the majority vote in heterogeneous settings? 2. Is the performance of the proposed method sensitive to variations in batch size?

Rating

7

Confidence

4

Soundness

4

Presentation

4

Contribution

4

Limitations

This paper does not present negative societal impacts.

Reviewer 2Gd96/10 · confidence 4/52024-07-11

Summary

This paper considers application of variance reduction to sign-based optimization, i.e., a setting where the algorithm can only access to $\mathrm{sign}(\nabla f(x^t; \xi^t))$ at step $t$, in the smooth and nonconvex setting. Because $\\{\pm 1\\}$-valued vectors can be transmitted more efficiently than real-valued vectors, the sign-based optimization is considered to be useful especially in distributed optimization. While a naive SGD-type algorithm previously achieved the convergence rate of $O(d^\frac12 T^{-\frac14})$, the proposed variance-reduced algorithm yields the rate of $O(d^\frac12 T^{-\frac13})$ (measured by the $L_1$-norm of the gradient.) Next, they applied the algorithm to finite-sum optimization (with $m$ components) and obtained the convergence rate of $O(m^\frac14 d^\frac12 T^{-\frac14})$. Finally, they applied these results to distributed optimization (the server received $\\{\pm 1\\}$-valued gradient estimators from clients), with an additional technique called \textit{majority vote} to obtain unbiased gradient estimators, and obtained improved communication complexities.

Strengths

### **The problem is well-motivated** Sign-based optimization is an important technique to reduce computational / communication complexities in large-scale optimization. I think it is a natural attempt to apply variance reduction, which is a common technique to accelerate optimization, to achieve more efficient sign-based optimization. ### **Improved convergence rate** Overall, the proposed approach improved the convergence rates of sign-based optimization, which I think solid contributions to the theory of optimization. - For stochastic non-convex optimization, the proposed algorithm achieves $\min{1\leq t \leq T}\mathbb{E}[\\|\nabla f(x^t)\\|]\lesssim d^\frac12 T^{-\frac13}$, which is an improvement from the $d^\frac12 T^{-\frac14}$ rate of SignSGD ([Bernstein et al., 2018](https://arxiv.org/abs/1802.04434)). - For finite-sum optimization, the proposed algorithm achieves $\min{1\leq t \leq T}\mathbb{E}\[\|\nabla f(x^t)\\|]\lesssim m^\frac14 d^\frac12 T^{-\frac12}$, whereas the previous SignSVRG algorithm achieves the rate of $O(m^\frac12 d^\frac12 T^{-\frac12})$ ([Chzhen and Schechtman, 2023](https://arxiv.org/abs/2305.13187). (I am a bit confused because Ii looks like there is no benefit of variance reduction for SignSVRG in terms of the dependency on $m$.) - For distributed optimization, the proposed algorithm achieves $\min{1\leq t \leq T}\mathbb{E}[\\|\nabla f(x^t)\\|]\lesssim d^\frac12 T^{-\frac12} + d n^{-\frac12}$ or $\lesssim d^\frac14 T^{-\frac14}$. The previous bound was $d T^{-\frac14} + d n^{-\frac12}$ ([Sun et al., 2023](https://proceedings.mlr.press/v202/sun23l.html)) or $d^\frac38 T^{-\frac18}$ ([Jin et al., 2023](https://arxiv.org/abs/2002.10940), in $L^2$-norm). This is achieved by a trick called majority vote, which stochastically maps the real-valued gradient vector (or $\nabla f_i(x)$) into $\\{\pm 1\\}$-valued vectors transmitted to the server. ### **Experimental results** The authors showed sufficient amount of experiments to verify the validity of the proposed methods (as a theory paper).

Weaknesses

### **Technical novelty** Variance reduction itself is quite common in the optimization field so I am afraid that this paper might be a bit incremental. Section 3.3 (Results under weaker assumptions) looks redundant given Theorem 2, as there are no technical difficulty about weakening the assumption described. ### **Hyperparameter choice** Although this is a common criticism, most variance reduction algorithm require a very careful hyperparameter tuning. It looks like the theoretical guarantees of the proposed algorithms also require a specific set of hyperparameters, and I am not confident in the robustness of the proposed algorithms to the hyperparameters.

Questions

### **On additional terms** - Why is the term of $md/T$ required in Theorem 4? - Why is the term of $d n^{-\frac12}$ required in Theorem 5? ### **On the difficulties to further improve the convergence rates** - In theorem 6, why cannot you improve the convergence rate into $\mathrm{poly}(d) T^{-\frac13}$? ### **Assumptions** - Is it possible to weaken Assumption 3 to averaged smoothness, i.e., $\frac1m \sum_{i=1}^m \\|\nabla f_i(x)-\nabla f_i(y)\\|^2 \leq L^2 \\|x-y\\|^2$

Rating

6

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

N/A

Reviewer hYmq7/10 · confidence 3/52024-07-16

Summary

This paper investigates the stochastic optimization problem and a special case of finite sum optimization problem. They propose the Sign-based Stochastic Variance Reduction (SSVR) method, leveraging variance reduction techiniques, which improves the convergence rate of Sign stochastic gradient descent (signSGD) algorithm. Further, they also study the heterogeneous majority vote in distributed settings and introduce two novel algorithms, whose convergence rates improve the state-of-the-art. Numerical experiments prove the efficiency of their algorithms and support their theoretical guarantees.

Strengths

The paper is clearly written, with detailed discussion and comparision.

Weaknesses

Some places of the paper need further clarification.

Questions

1. How did you get the convergence rate of $O(m^{1/2}d^{1/2}T^{−1/2})$ for the SignSVRG of y? It seems that in their paper they claim $O(T^{-1/2}) convergence rate without $m$ and $n$ in the numerator. Did you calculate this result by yourself? Please provide discussions. 2. There is another paper studying sign-based optimization (Qin et al. (2023)), the setting studied in this paper is the same as your problem (2). Can you also discuss this paper and compare the results? References: 1. E. Chzhen and S. Schechtman. SignSVRG: fixing SignSGD via variance reduction. ArXiv e-prints, 317 arXiv:2305.13187, 2023. 2. Qin, Z., Liu, Z., & Xu, P. (2023). Convergence of Sign-based Random Reshuffling Algorithms for Nonconvex Optimization. arXiv preprint arXiv:2310.15976.

Rating

7

Confidence

3

Soundness

3

Presentation

3

Contribution

3

Limitations

Yes.

Area Chair aAQH2024-08-08

Please read the authors’ rebuttal and reply by August 13, 2024 (11:59 PM, AOE time)

Dear Reviewers, Thank you for your hard work during the review process. The authors have responded to your initial reviews. **If you haven’t already done so, please take the time to carefully read their responses.** It is crucial for the authors to receive acknowledgment of your review by the deadline for the author-reviewer discussion period, which is August 13, 2024 (11:59 PM, Anywhere on Earth). Please address any points of disagreement with the authors as early as possible. Best, Your AC

Reviewer Huhp2024-08-09

Thank the authors for the rely. It addresses all my questions and concerns. I'm willing to keep my score.

Reviewer hYmq2024-08-11

Thanks for the rebuttal. It addressed all my concerns. I will increase my score.

Authorsrebuttal2024-08-11

Thank you very much for your kind reply! We will revise our paper according to the constructive reviews. Best regards, Authors

Reviewer XovB2024-08-13

final rating

Thank you for the detailed responses. They have satisfactorily addressed all my concerns, and I am inclined to recommend acceptance.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC