ABSTRACT
Suppose we are given an n-dimensional order-3 symmetric tensor T ∈ (ℝn)⊗ 3 that is the sum of r random rank-1 terms. The problem of recovering the rank-1 components is possible in principle when r ≲ n2 but polynomial-time algorithms are only known in the regime r ≪ n3/2. Similar “statistical-computational gaps” occur in many high-dimensional inference tasks, and in recent years there has been a flurry of work on explaining the apparent computational hardness in these problems by proving lower bounds against restricted (yet powerful) models of computation such as statistical queries (SQ), sum-of-squares (SoS), and low-degree polynomials (LDP). However, no such prior work exists for tensor decomposition, largely because its hardness does not appear to be explained by a “planted versus null” testing problem.
We consider a model for random order-3 tensor decomposition where one component is slightly larger in norm than the rest (to break symmetry), and the components are drawn uniformly from the hypercube. We resolve the computational complexity in the LDP model: O(logn)-degree polynomial functions of the tensor entries can accurately estimate the largest component when r ≪ n3/2 but fail to do so when r ≫ n3/2. This provides rigorous evidence suggesting that the best known algorithms for tensor decomposition cannot be improved, at least by known approaches. A natural extension of the result holds for tensors of any fixed order k ≥ 3, in which case the LDP threshold is r ∼ nk/2.
- Dimitris Achlioptas and Amin Coja-Oghlan. 2008. Algorithmic barriers from phase transitions. In 2008 49th Annual IEEE Symposium on Foundations of Computer Science. 793–802. Google ScholarDigital Library
- Esraa Al-Sharoa, Mahmood Al-Khassaweneh, and Selin Aviyente. 2017. A tensor based framework for community detection in dynamic networks. In International conference on acoustics, speech and signal processing (ICASSP). 2312–2316. Google ScholarDigital Library
- Anima Anandkumar, Dean P Foster, Daniel J Hsu, Sham M Kakade, and Yi-Kai Liu. 2012. A spectral algorithm for latent dirichlet allocation. Advances in neural information processing systems, 25 (2012). Google Scholar
- Animashree Anandkumar, Rong Ge, Daniel Hsu, and Sham Kakade. 2013. A tensor spectral approach to learning mixed membership community models. In Conference on Learning Theory. 867–881. Google Scholar
- Anima Anandkumar, Rong Ge, and Majid Janzamin. 2014. Analyzing tensor power method dynamics: Applications to learning overcomplete latent variable models. arXiv preprint arXiv:1411.1488. Google Scholar
- Animashree Anandkumar, Rong Ge, and Majid Janzamin. 2015. Learning overcomplete latent variable models through tensor methods. In Conference on Learning Theory. 36–112. Google Scholar
- Joseph Anderson, Mikhail Belkin, Navin Goyal, Luis Rademacher, and James Voss. 2014. The more, the merrier: the blessing of dimensionality for learning large Gaussian mixtures. In Conference on Learning Theory. 1135–1164. Google Scholar
- Davide Bacciu and Danilo P Mandic. 2020. Tensor decompositions in deep learning. arXiv preprint arXiv:2002.11835. Google Scholar
- Afonso S Bandeira, Jess Banks, Dmitriy Kunisky, Christopher Moore, and Alexander S Wein. 2021. Spectral planting and the hardness of refuting cuts, colorability, and communities in random graphs. In Conference on Learning Theory. 410–473. Google Scholar
- Afonso S Bandeira, Ben Blum-Smith, Joe Kileel, Amelia Perry, Jonathan Weed, and Alexander S Wein. 2017. Estimation under group actions: recovering orbits from invariants. arXiv preprint arXiv:1712.10163. Google Scholar
- Afonso S Bandeira, Ahmed El Alaoui, Samuel B Hopkins, Tselil Schramm, Alexander S Wein, and Ilias Zadik. 2022. The Franz-Parisi Criterion and Computational Trade-offs in High Dimensional Statistics. arXiv preprint arXiv:2205.09727. Google Scholar
- Afonso S Bandeira, Dmitriy Kunisky, and Alexander S Wein. 2020. Computational Hardness of Certifying Bounds on Constrained PCA Problems. In 11th Innovations in Theoretical Computer Science Conference (ITCS 2020). 151, 78. Google Scholar
- Jess Banks, Sidhanth Mohanty, and Prasad Raghavendra. 2021. Local statistics, semidefinite programming, and community detection. In Proceedings of the 2021 ACM-SIAM Symposium on Discrete Algorithms (SODA). 1298–1316. Google ScholarCross Ref
- Boaz Barak, Samuel Hopkins, Jonathan Kelner, Pravesh K Kothari, Ankur Moitra, and Aaron Potechin. 2019. A nearly tight sum-of-squares lower bound for the planted clique problem. SIAM J. Comput., 48, 2 (2019), 687–735. Google ScholarDigital Library
- Boaz Barak, Jonathan A Kelner, and David Steurer. 2015. Dictionary learning and tensor decomposition via the sum-of-squares method. In Proceedings of the forty-seventh annual ACM symposium on Theory of computing. 143–151. Google ScholarDigital Library
- Gerard Ben Arous, Reza Gheissari, and Aukosh Jagannath. 2020. Algorithmic thresholds for tensor PCA. The Annals of Probability, 48, 4 (2020), 2052–2087. Google Scholar
- Gérard Ben Arous, Alexander S Wein, and Ilias Zadik. 2020. Free Energy Wells and Overlap Gap Property in Sparse PCA. In Conference on Learning Theory. 479–482. Google Scholar
- Quentin Berthet and Philippe Rigollet. 2013. Complexity theoretic lower bounds for sparse principal component detection. In Conference on learning theory. 1046–1066. Google Scholar
- Aditya Bhaskara, Moses Charikar, Ankur Moitra, and Aravindan Vijayaraghavan. 2014. Smoothed analysis of tensor decompositions. In Proceedings of the forty-sixth annual ACM symposium on Theory of computing. 594–603. Google ScholarDigital Library
- Giulio Biroli, Chiara Cammarota, and Federico Ricci-Tersenghi. 2020. How to iron out rough landscapes and get optimal performances: averaged gradient descent and its application to tensor PCA. Journal of Physics A: Mathematical and Theoretical, 53, 17 (2020), 174003. Google ScholarCross Ref
- Cristiano Bocci, Luca Chiantini, and Giorgio Ottaviani. 2014. Refined methods for the identifiability of tensors. Annali di Matematica Pura ed Applicata (1923-), 193, 6 (2014), 1691–1702. Google Scholar
- Nicolas Boumal, Tamir Bendory, Roy R Lederman, and Amit Singer. 2018. Heterogeneous multireference alignment: A single pass approach. In 52nd Annual Conference on Information Sciences and Systems (CISS). 1–6. Google ScholarCross Ref
- Matthew Brennan and Guy Bresler. 2020. Reducibility and statistical-computational gaps from secret leakage. In Conference on Learning Theory. 648–847. Google Scholar
- Matthew Brennan, Guy Bresler, and Wasim Huleihel. 2018. Reducibility and computational lower bounds for problems with planted sparse structure. In Conference On Learning Theory. 48–166. Google Scholar
- Matthew S Brennan, Guy Bresler, Sam Hopkins, Jerry Li, and Tselil Schramm. 2021. Statistical Query Algorithms and Low Degree Tests Are Almost Equivalent. In Conference on Learning Theory. 774–774. Google Scholar
- Guy Bresler and Brice Huang. 2022. The algorithmic phase transition of random k-SAT for low degree polynomials. In 2021 IEEE 62nd Annual Symposium on Foundations of Computer Science (FOCS). 298–309. Google ScholarCross Ref
- Zongchen Chen, Elchanan Mossel, and Ilias Zadik. 2022. Almost-Linear Planted Cliques Elude the Metropolis Process. arXiv preprint arXiv:2204.01911. Google Scholar
- Amin Coja-Oghlan, Oliver Gebhard, Max Hahn-Klimroth, Alexander S Wein, and Ilias Zadik. 2022. Statistical and Computational Phase Transitions in Group Testing. In Conference on Learning Theory. 4764–4781. Google Scholar
- Lieven De Lathauwer, Josphine Castaing, and Jean-Franois Cardoso. 2007. Fourth-order cumulant-based blind identification of underdetermined mixtures. IEEE Transactions on Signal Processing, 55, 6 (2007), 2965–2973. Google ScholarDigital Library
- Aurelien Decelle, Florent Krzakala, Cristopher Moore, and Lenka Zdeborová. 2011. Asymptotic analysis of the stochastic block model for modular networks and its algorithmic applications. Physical Review E, 84, 6 (2011), 066106. Google ScholarCross Ref
- Ilias Diakonikolas and Daniel Kane. 2022. Non-gaussian component analysis via lattice basis reduction. In Conference on Learning Theory. 4535–4547. Google Scholar
- Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. 2017. Statistical query lower bounds for robust estimation of high-dimensional gaussians and gaussian mixtures. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS). 73–84. Google ScholarCross Ref
- Jingqiu Ding, Tommaso d’Orsi, Chih-Hung Liu, David Steurer, and Stefan Tiegel. 2022. Fast algorithm for overcomplete order-3 tensor decomposition. In Conference on Learning Theory. 3741–3799. Google Scholar
- Yunzi Ding, Dmitriy Kunisky, Alexander S Wein, and Afonso S Bandeira. 2019. Subexponential-time algorithms for sparse PCA. arXiv preprint arXiv:1907.11635. Google Scholar
- Zhou Fan, Yi Sun, Tianhao Wang, and Yihong Wu. 2020. Likelihood landscape and maximum likelihood estimation for the discrete orbit recovery model. Communications on Pure and Applied Mathematics. Google Scholar
- Vitaly Feldman, Elena Grigorescu, Lev Reyzin, Santosh S Vempala, and Ying Xiao. 2017. Statistical algorithms and a lower bound for detecting planted cliques. Journal of the ACM (JACM), 64, 2 (2017), 1–37. Google ScholarDigital Library
- David Gamarnik. 2021. The overlap gap property: A topological barrier to optimizing over random structures. Proceedings of the National Academy of Sciences, 118, 41 (2021), e2108492118. Google ScholarCross Ref
- David Gamarnik, Aukosh Jagannath, and Subhabrata Sen. 2021. The overlap gap property in principal submatrix recovery. Probability Theory and Related Fields, 181, 4 (2021), 757–814. Google ScholarCross Ref
- David Gamarnik, Aukosh Jagannath, and Alexander S Wein. 2020. Low-degree hardness of random optimization problems. In 2020 IEEE 61st Annual Symposium on Foundations of Computer Science (FOCS). 131–140. Google ScholarCross Ref
- David Gamarnik, Cristopher Moore, and Lenka Zdeborová. 2022. Disordered Systems Insights on Computational Hardness. arXiv preprint arXiv:2210.08312. Google Scholar
- David Gamarnik and Madhu Sudan. 2017. Limits of Local Algorithms over Sparse Random Graphs. The Annals of Probability, 2353–2376. Google Scholar
- David Gamarnik and Ilias Zadik. 2017. Sparse high-dimensional linear regression. algorithmic barriers and a local search algorithm. arXiv preprint arXiv:1711.04952. Google Scholar
- David Gamarnik and Ilias Zadik. 2019. The landscape of the planted clique problem: Dense subgraphs and the overlap gap property. arXiv preprint arXiv:1904.07174. Google Scholar
- Rong Ge, Qingqing Huang, and Sham M Kakade. 2015. Learning mixtures of gaussians in high dimensions. In Proceedings of the forty-seventh annual ACM symposium on Theory of computing. 761–770. Google ScholarDigital Library
- Rong Ge and Tengyu Ma. 2015. Decomposing Overcomplete 3rd Order Tensors using Sum-of-Squares Algorithms. Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques, 829–849. Google Scholar
- Rong Ge and Tengyu Ma. 2017. On the optimization landscape of tensor decompositions. Advances in Neural Information Processing Systems, 30 (2017). Google Scholar
- Navin Goyal, Santosh Vempala, and Ying Xiao. 2014. Fourier PCA and robust tensor decomposition. In Proceedings of the forty-sixth annual ACM symposium on Theory of computing. 584–593. Google ScholarDigital Library
- Bruce Hajek, Yihong Wu, and Jiaming Xu. 2015. Computational lower bounds for community detection on random graphs. In Conference on Learning Theory. 899–928. Google Scholar
- Johan Håstad. 1989. Tensor rank is NP-complete. In International Colloquium on Automata, Languages, and Programming. 451–460. Google Scholar
- Christopher J Hillar and Lek-Heng Lim. 2013. Most tensor problems are NP-hard. Journal of the ACM (JACM), 60, 6 (2013), 1–39. Google ScholarDigital Library
- Justin Holmgren and Alexander S Wein. 2021. Counterexamples to the Low-Degree Conjecture. In 12th Innovations in Theoretical Computer Science Conference (ITCS 2021). 185. Google Scholar
- Samuel Hopkins. 2018. Statistical Inference and the Sum of Squares Method. Ph. D. Dissertation. Cornell University. Google Scholar
- Samuel B Hopkins, Pravesh K Kothari, Aaron Potechin, Prasad Raghavendra, Tselil Schramm, and David Steurer. 2017. The power of sum-of-squares for detecting hidden structures. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS). 720–731. Google ScholarCross Ref
- Samuel B Hopkins, Tselil Schramm, and Jonathan Shi. 2019. A robust spectral algorithm for overcomplete tensor decomposition. In Conference on Learning Theory. 1683–1722. Google Scholar
- Samuel B Hopkins, Tselil Schramm, Jonathan Shi, and David Steurer. 2016. Fast spectral algorithms from sum-of-squares proofs: tensor decomposition and planted sparse vectors. In Proceedings of the forty-eighth annual ACM symposium on Theory of Computing. 178–191. Google ScholarDigital Library
- Samuel B Hopkins, Jonathan Shi, and David Steurer. 2015. Tensor principal component analysis via sum-of-square proofs. In Conference on Learning Theory. 956–1006. Google Scholar
- Samuel B Hopkins and David Steurer. 2017. Efficient Bayesian estimation from few samples: community detection and related problems. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS). 379–390. Google ScholarCross Ref
- Daniel Hsu and Sham M Kakade. 2013. Learning mixtures of spherical gaussians: moment methods and spectral decompositions. In Proceedings of the 4th conference on Innovations in Theoretical Computer Science. 11–20. Google ScholarDigital Library
- Mark Jerrum. 1992. Large cliques elude the Metropolis process. Random Structures & Algorithms, 3, 4 (1992), 347–359. Google ScholarCross Ref
- Bing-Yi Jing, Ting Li, Zhongyuan Lyu, and Dong Xia. 2021. Community detection on mixture multilayer networks via regularized tensor decomposition. The Annals of Statistics, 49, 6 (2021), 3181–3205. Google ScholarCross Ref
- Bohdan Kivva and Aaron Potechin. 2020. Exact nuclear norm, completion and decomposition for random overcomplete tensors via degree-4 SOS. arXiv preprint arXiv:2011.09416. Google Scholar
- Frederic Koehler and Elchanan Mossel. 2021. Reconstruction on Trees and Low-Degree Polynomials. arXiv preprint arXiv:2109.06915. Google Scholar
- Tamara G. Kolda. 2021. Will the real Jennrich’s Algorithm please stand up? Available online at. www.mathsci.ai/post/jennrich Accessed: 10-22-2022 Google Scholar
- Pravesh K Kothari and Peter Manohar. 2021. A stress-free sum-of-squares lower bound for coloring. arXiv preprint arXiv:2105.07517. Google Scholar
- Pravesh K Kothari, Ryuhei Mori, Ryan O’Donnell, and David Witmer. 2017. Sum of squares lower bounds for refuting any CSP. In Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing. 132–145. Google ScholarDigital Library
- Dmitriy Kunisky. 2021. Hypothesis testing with low-degree polynomials in the Morris class of exponential families. In Conference on Learning Theory. 2822–2848. Google Scholar
- Dmitriy Kunisky. 2022. Lecture Notes on Sum-of-Squares Optimization. Available online at. www.kunisky.com/static/teaching/2022spring-sos/sos-notes.pdf Accessed: 09-29-2022 Google Scholar
- Dmitriy Kunisky, Alexander S Wein, and Afonso S Bandeira. 2022. Notes on computational hardness of hypothesis testing: Predictions using the low-degree likelihood ratio. In ISAAC Congress (International Society for Analysis, its Applications and Computation). 1–50. Google ScholarCross Ref
- Thibault Lesieur, Florent Krzakala, and Lenka Zdeborová. 2015. MMSE of probabilistic low-rank matrix estimation: Universality with respect to the output channel. In 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton). 680–687. Google ScholarDigital Library
- Sue E Leurgans, Robert T Ross, and Rebecca B Abel. 1993. A decomposition for three-way arrays. SIAM J. Matrix Anal. Appl., 14, 4 (1993), 1064–1083. Google ScholarDigital Library
- Allen Liu and Ankur Moitra. 2021. Algorithms from Invariants: Smoothed Analysis of Orbit Recovery over SO(3). arXiv preprint arXiv:2106.02680. Google Scholar
- Yuetian Luo and Anru R Zhang. 2022. Tensor clustering with planted structures: Statistical optimality and computational limits. The Annals of Statistics, 50, 1 (2022), 584–613. Google ScholarCross Ref
- Tengyu Ma, Jonathan Shi, and David Steurer. 2016. Polynomial-time tensor decompositions with sum-of-squares. In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS). 438–446. Google ScholarCross Ref
- Stefano Sarao Mannelli, Florent Krzakala, Pierfrancesco Urbani, and Lenka Zdeborova. 2019. Passed & spurious: Descent algorithms and local minima in spiked matrix-tensor models. In International conference on machine learning. 4333–4342. Google Scholar
- Ankur Moitra. 2014. Algorithmic aspects of machine learning. Lecture notes. Google Scholar
- Ankur Moitra and Alexander S Wein. 2019. Spectral methods from tensor networks. In Proceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing. 926–937. Google ScholarDigital Library
- Andrea Montanari and Alexander S Wein. 2022. Equivalence of Approximate Message Passing and Low-Degree Polynomials in Rank-One Matrix Estimation. arXiv preprint arXiv:2212.06996. Google Scholar
- Elchanan Mossel and Sébastien Roch. 2005. Learning nonsingular phylogenies and hidden Markov models. In Proceedings of the thirty-seventh annual ACM symposium on Theory of computing. 366–375. Google ScholarDigital Library
- Mohamed Ouerfelli, Mohamed Tamaazousti, and Vincent Rivasseau. 2022. Random tensor theory for tensor decomposition. In Proceedings of the AAAI Conference on Artificial Intelligence. Google ScholarCross Ref
- Amelia Perry, Jonathan Weed, Afonso S Bandeira, Philippe Rigollet, and Amit Singer. 2019. The sample complexity of multireference alignment. SIAM Journal on Mathematics of Data Science, 1, 3 (2019), 497–517. Google ScholarCross Ref
- Aaron Potechin and Goutham Rajendran. 2020. Machinery for proving sum-of-squares lower bounds on certification problems. arXiv preprint arXiv:2011.04253. Google Scholar
- Stephan Rabanser, Oleksandr Shchur, and Stephan Günnemann. 2017. Introduction to tensor decompositions and their applications in machine learning. arXiv preprint arXiv:1711.10781. Google Scholar
- Prasad Raghavendra, Tselil Schramm, and David Steurer. 2018. High dimensional estimation via sum-of-squares proofs. In Proceedings of the International Congress of Mathematicians: Rio de Janeiro 2018. 3389–3423. Google Scholar
- Emile Richard and Andrea Montanari. 2014. A statistical model for tensor PCA. Advances in neural information processing systems, 27 (2014). Google Scholar
- Tselil Schramm and David Steurer. 2017. Fast and robust tensor decomposition with applications to dictionary learning. In Conference on Learning Theory. 1760–1793. Google Scholar
- Tselil Schramm and Alexander S Wein. 2022. Computational barriers to estimation from low-degree polynomials. The Annals of Statistics, 50, 3 (2022), 1833–1858. Google ScholarCross Ref
- Nicholas D Sidiropoulos, Lieven De Lathauwer, Xiao Fu, Kejun Huang, Evangelos E Papalexakis, and Christos Faloutsos. 2017. Tensor decomposition for signal processing and machine learning. IEEE Transactions on Signal Processing, 65, 13 (2017), 3551–3582. Google ScholarDigital Library
- Alexander Spence Wein. 2018. Statistical estimation in the presence of group actions. Ph. D. Dissertation. Massachusetts Institute of Technology. Google Scholar
- Alexander S Wein. 2022. Average-Case Complexity of Tensor Decomposition for Low-Degree Polynomials. arXiv preprint arXiv:2211.05274. Google Scholar
- Alexander S Wein. 2022. Optimal low-degree hardness of maximum independent set. Mathematical Statistics and Learning, 4, 3 (2022), 221–251. Google ScholarCross Ref
- Alexander S Wein, Ahmed El Alaoui, and Cristopher Moore. 2019. The Kikuchi hierarchy and tensor PCA. In 2019 IEEE 60th Annual Symposium on Foundations of Computer Science (FOCS). 1446–1468. Google ScholarCross Ref
- Yihong Wu and Jiaming Xu. 2021. Statistical problems with planted structures: Information-theoretical and computational limits. Information-Theoretic Methods in Data Science, 383 (2021). Google Scholar
- Ilias Zadik, Min Jae Song, Alexander S Wein, and Joan Bruna. 2022. Lattice-based methods surpass sum-of-squares in clustering. In Conference on Learning Theory. 1247–1248. Google Scholar
Index Terms
- Average-Case Complexity of Tensor Decomposition for Low-Degree Polynomials
Recommendations
Hard Functions for Low-Degree Polynomials over Prime Fields
In this article, we present a new hardness amplification for low-degree polynomials over prime fields, namely, we prove that if some function is mildly hard to approximate by any low-degree polynomials then the sum of independent copies of the function ...
Relations between Average-Case and Worst-Case Complexity
The consequences of the worst-case assumption NP=P are very well understood. On the other hand, we only know a few consequences of the analogous average-case assumption “NP is easy on average.” In this paper we establish several new results on the worst-...
Comments