Consensus-ADMM for General Quadratically Constrained Quadratic Programming
Open accessThe DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index.
Abstract
Nonconvex quadratically constrained quadratic programming (QCQP) problems have numerous applications in signal processing, machine learning, and wireless communications, albeit the general QCQP is NP-hard, and several interesting special cases are NP-hard as well. This paper proposes a new algorithm for general QCQP. The problem is first reformulated in consensus optimization form, to which the alternating direction method of multipliers can be applied. The reformulation is done in such a way that each of the subproblems is a QCQP with only one constraint (QCQP-1), which is efficiently solvable irrespective of (non)convexity. The core components are carefully designed to make the overall algorithm more scalable, including efficient methods for solving QCQP-1, memory efficient implementation, parallel/distributed implementation, and smart initialization. The proposed algorithm is then tested in two applications: multicast beamforming and phase retrieval. The results indicate superior performance over prior state-of-the-art methods.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Boyd, S., & Vandenberghe, L. (2004) · doi:10.1017/cbo9780511804441 · Cited by 56,601
Linear Matrix Inequalities in System and Control Theory
Boyd, S., El Ghaoui, L., Feron, E., & Balakrishnan, V. (1994) · Society for Industrial and Applied Mathematics eBooks · doi:10.1137/1.9781611970777 · Cited by 21,687
Boyd, S., Parikh, N., Chu, E., Peleato, B., & Eckstein, J. (2011) · Foundations and Trends® in Machine Learning · doi:10.1561/2200000016 · Cited by 17,075
Boyd, S. (2010) · now publishers, Inc. eBooks · doi:10.1561/9781601984616 · Cited by 12,677
A practical algorithm for the determination of phase from image and diffraction plane pictures
Gerchberg, R. W. (1972) · Optik · Cited by 4,519
Semidefinite Relaxation of Quadratic Optimization Problems
Luo, Z.-Q., Ma, W.-K., So, A., Ye, Y., & Zhang, S. (2010) · IEEE Signal Processing Magazine · doi:10.1109/msp.2010.936019 · Cited by 3,715 · cited 1×
“Quadratically constrained quadratic programming (QCQP) is an optimization problem that minimizes a quadratic function subject to quadractic inequality and equality constraints [1].”
BackgroundGoemans, M. X., & Williamson, D. P. (1995) · Journal of the ACM · doi:10.1145/227683.227684 · Cited by 3,587 · cited 1×
“Then for each quadratic term, we have that for any point x (0) ∈ C n .”
BackgroundEckstein, J., & Bertsekas, D. P. (1992) · Mathematical Programming · doi:10.1007/bf01581204 · Cited by 2,791 · cited 1×
“ADMM was first designed for convex problems, for which it is known to converge under mild conditions [15].”
BackgroundReconstruction of an object from the modulus of its Fourier transform
Fienup, J. R. (1978) · Optics Letters · doi:10.1364/ol.3.000027 · Cited by 1,836 · cited 1×
“There exist many algorithms designed for phase retrieval under specific scenarios, for example, the classical Gerchberg-Saxton algorithm [29] and Fienup’s algorithm [30] were designed for Fourier-based measurements, and have been successfully applied in the phase retrieval community for decades.”
MethodTransmit beamforming for physical-layer multicasting
Sidiropoulos, N. D., Davidson, T. N., & Luo, Z.-Q. (2006) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2006.872578 · Cited by 1,335 · cited 8×
“At this point we need to remind the reader that, although the consensus ADMM algorithm we derived for non-convex QCQPs has an appealing form and cheap per-iteration complexity, it is after all a heuristic for what is in general an NP-hard problem.”
BackgroundMethodInfluentialSelf-citationYuille, A. L., & Rangarajan, A. (2003) · Neural Computation · doi:10.1162/08997660360581958 · Cited by 1,151 · cited 1×
“Recently [11] proposed feasible point pursuit - successive convex approximation (FPP-SCA) to address this issue, by adding a slack variable to each constraint and an ℓ 1 penalty on the slacks to the cost.”
BackgroundPhase Retrieval with Application to Optical Imaging: A contemporary overview
Shechtman, Y., Eldar, Y. C., Cohen, O., Chapman, H. N., Miao, J., & Segev, M. (2015) · IEEE Signal Processing Magazine · doi:10.1109/msp.2014.2352673 · Cited by 1,131
Phase Retrieval via Wirtinger Flow: Theory and Algorithms
Candès, E. J., Li, X., & Soltanolkotabi, M. (2015) · IEEE Transactions on Information Theory · doi:10.1109/tit.2015.2399924 · Cited by 1,049 · cited 3×
“In a noiseless scenario, we apply the updates (14) to the problem of phase retrieval with random Gaussian measurements, and compare it with two state-of-the-art algorithms designed for this setting: Wirtinger flow [32] and alternating minimization [31].”
BackgroundMethodHong, M., Luo, Z.-Q., & Razaviyayn, M. (2016) · SIAM Journal on Optimization · doi:10.1137/140990309 · Cited by 841 · cited 5×
“The very first step in the proof of [19] shows that the augmented Lagrangian is monotonically non-increasing under certain conditions.”
BackgroundInfluentialPhase recovery, MaxCut and complex semidefinite programming
Waldspurger, I., d’Aspremont, A., & Mallat, S. (2015) · Mathematical Programming · doi:10.1007/s10107-013-0738-9 · Cited by 573 · cited 1×
“SDR-based methods have also been developed, including PhaseLift [31] and PhaseCut [32], however, they again suffer from effectively squaring the problem dimension, and how to recover a good approximate solution to the original problem when they return a higher rank matrix is an open question.”
MethodConvex Optimization-Based Beamforming
Gershman, A., Sidiropoulos, N., Shahbazpanahi, S., Bengtsson, M., & Ottersten, B. (2010) · IEEE Signal Processing Magazine · doi:10.1109/msp.2010.936015 · Cited by 571
Self-citationPhase Retrieval via Matrix Completion
Candès, E. J., Eldar, Y. C., Strohmer, T., & Voroninski, V. (2013) · SIAM Journal on Imaging Sciences · doi:10.1137/110848074 · Cited by 565
Rank-Constrained Separable Semidefinite Programming With Applications to Optimal Beamforming
Huang, Y., & Palomar, D. P. (2010) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2009.2031732 · Cited by 561 · cited 1×
“Even if not, one can resort to rank reduction as in [5], which handles a more general rank reduction problem.”
BackgroundPhase Retrieval Using Alternating Minimization
Netrapalli, P., Jain, P., & Sanghavi, S. (2015) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2015.2448516 · Cited by 535 · cited 3×
“In a noiseless scenario, we apply the updates (14) to the problem of phase retrieval with random Gaussian measurements, and compare it with two state-of-the-art algorithms designed for this setting: Wirtinger flow [32] and alternating minimization [31].”
BackgroundMethodHayes, M. (1982) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/tassp.1982.1163863 · Cited by 431 · cited 1×
“ch measurements can be directly obtained using optics, and have the nice property that the model is identifiable (up to a global phase ambiguity and conjugate inversion) for real signals almost surely [11]. Despite this uniqueness property, the recovery problem per se is known to be very hard from a computational point…”
BackgroundAn alternating direction algorithm for matrix completion with nonnegative factors
Xu, Y., Yin, W., Wen, Z., & Zhang, Y. (2012) · Frontiers of Mathematics in China · doi:10.1007/s11464-012-0194-5 · Cited by 339 · cited 1×
“In [16], ADMM was applied to non-negative matrix factorization (a non-convex problem) with missing values, and it was shown that, if ADMM converges for this non-convex problem, then it converges to a KKT point.”
MethodNew Results on Quadratic Minimization
Ye, Y., & Zhang, S. (2003) · SIAM Journal on Optimization · doi:10.1137/s105262340139001x · Cited by 248 · cited 1×
“For general non-convex QCQPs, the prevailing method to tackle the problem is through semi-definite relaxation (SDR), where the following semi-definite programming (SDP) problem is solved instead minimize where H n denotes the set of n × n complex Hermitian matrices.”
MethodStrong Duality in Nonconvex Quadratic Optimization with Two Quadratic Constraints
Beck, A., & Eldar, Y. C. (2006) · SIAM Journal on Optimization · doi:10.1137/050644471 · Cited by 234
Feasible Point Pursuit and Successive Approximation of Non-Convex QCQPs
Mehanna, O., Huang, K., Gopalakrishnan, B., Konar, A., & Sidiropoulos, N. D. (2015) · IEEE Signal Processing Letters · doi:10.1109/lsp.2014.2370033 · Cited by 229 · cited 15×
“It is possible to develop specialized solvers for FPP-SCA to accelerate it, but it is a non-trivial task which may require a lot of thinking, whereas our algorithm, readily available in Alg.”
BackgroundMethodResultInfluentialSelf-citationApproximation Bounds for Quadratic Optimization with Homogeneous Quadratic Constraints
Luo, Z., Sidiropoulos, N. D., Tseng, P., & Zhang, S. (2007) · SIAM Journal on Optimization · doi:10.1137/050642691 · Cited by 225 · cited 1×
“Then for each quadratic term, we have that for any point x (0) ∈ C n .”
BackgroundSelf-citationNew results on Hermitian matrix rank-one decomposition
Ai, W., Huang, Y., & Zhang, S. (2011) · Mathematical Programming · doi:10.1007/s10107-009-0304-7 · Cited by 197 · cited 1×
“d as two inequalities). A QCQP is in general NP-hard, except for some special cases, for example when all the {Ai}m i=1 in “≤” inequality constraints are positive semi-definite [4, §4.4], m is “small” [1,2,14,27], or if the quadratic terms are all homogeneous and {Ai}m i=0 are all Toeplitz [16]. 1 For general non-conve…”
MethodInfluentialHong, M., Luo, Z.-Q. T., & Razaviyayn, M. (2015) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2015.7178689 · Cited by 171
A Conic Quadratic Programming Approach to Physical Layer Multicasting for Large-Scale Antenna Arrays
Tran, L., Hanif, M. F., & Juntti, M. J. (2014) · IEEE Signal Processing Letters · doi:10.1109/lsp.2013.2293840 · Cited by 139 · cited 7×
“ADMM is able to compute a beamformer that is better than one step of SLA and do so in less time; in fact the transmission power obtained via ADMM is only slightly worse than ten steps of SLA.”
MethodInfluentialLiavas, A. P., & Sidiropoulos, N. D. (2015) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2015.2454476 · Cited by 122 · cited 1×
“Some follow-up works on other nonconvex problems have made similar claims [17], [18].”
BackgroundSelf-citationAlternating direction method of multipliers for real and complex polynomial optimization models
Jiang, B., Ma, S., & Zhang, S. (2014) · Optimization · doi:10.1080/02331934.2014.895901 · Cited by 57 · cited 1×
“Some follow-up works on other nonconvex problems have made similar claims [17], [18].”
BackgroundPhase Retrieval Using Feasible Point Pursuit: Algorithms and Cramér–Rao Bound
Qian, C., Sidiropoulos, N. D., Huang, K., Huang, L., & So, H. C. (2016) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2016.2593688 · Cited by 37 · cited 2×
“3) Gaussian Noise: Another interesting scenario is that where measurements are corrupted by additive white Gaussian noise, in which case maximum likelihood estimation can be cast as the following non-convex QCQP (similar to the LS-FPP formulation proposed in [28])”
BackgroundMethodSelf-citationHidden Convexity in QCQP with Toeplitz-Hermitian Quadratics
Konar, A., & Sidiropoulos, N. D. (2015) · IEEE Signal Processing Letters · doi:10.1109/lsp.2015.2419571 · Cited by 29 · cited 2×
“A QCQP is in general NP-hard, except for some special cases, for example when all the { A i } mi =1 in “ ” inequality constraints are positive semi-definite [2, § 4.4], m is “small” [3–6], or if the quadratic terms are all homogeneous and { A i } m i =0 are all Toeplitz [7].”
BackgroundInfluentialSelf-citation
Cited by
Signal Processing for High-Throughput Satellites: Challenges in New Interference-Limited Scenarios
Perez-Neira, A. I., Vazquez, M. A., Shankar, M. R. B., Maleki, S., & Chatzinotas, S. (2019) · IEEE Signal Processing Magazine · doi:10.1109/msp.2019.2894391 · Cited by 170
“ng P 2 in real multibeam satellite systems. This is the case for the work done in [15], which promotes the use of the consensus alternate direction February 13, 2018 DRAFT 15 of multipliers method of [16] to solve the non-convex QCQP P 2. C. Multiple Gateways Multibeam precoding over multiple GWs consists of transmitt…”
Method
Metadata
- Authors
- Huang, K., & Sidiropoulos, N. D.
- Venue
- IEEE Transactions on Signal Processing
- Published
- 2016-10-15
- Volume
- 64
- Issue
- 20
- Pages
- 5297-5310
- ISSN
- 1053-587X (ISSN-L)
- 1941-0476
- OpenAlex
- W2236834722
- Cited by
- 242
- References
- 32
- FWCI
- 25.94
- Type
- journal article
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.