{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T04:53:01Z","timestamp":1782795181149,"version":"3.54.5"},"reference-count":81,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"4","license":[{"start":{"date-parts":[[2025,4,1]],"date-time":"2025-04-01T00:00:00Z","timestamp":1743465600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2025,4,1]],"date-time":"2025-04-01T00:00:00Z","timestamp":1743465600000},"content-version":"am","delay-in-days":0,"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2025,4,1]],"date-time":"2025-04-01T00:00:00Z","timestamp":1743465600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,4,1]],"date-time":"2025-04-01T00:00:00Z","timestamp":1743465600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"Packard Fellowship in Science and Engineering"},{"name":"Sloan Fellowship in Mathematics"},{"name":"NSF-CAREER","award":["1846369"],"award-info":[{"award-number":["1846369"]}]},{"name":"Defense Advanced Research Projects Agency (DARPA) Learning with Less Labels (LwLL) and Fast Network Interface Cards (FastNICS) Programs"},{"name":"NSF-Civil Infrastructure Systems","award":["1813877"],"award-info":[{"award-number":["1813877"]}]},{"name":"NSF-Civil Infrastructure Systems","award":["2008443"],"award-info":[{"award-number":["2008443"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Inform. Theory"],"published-print":{"date-parts":[[2025,4]]},"DOI":"10.1109\/tit.2025.3530335","type":"journal-article","created":{"date-parts":[[2025,1,15]],"date-time":"2025-01-15T19:40:15Z","timestamp":1736970015000},"page":"2991-3037","source":"Crossref","is-referenced-by-count":3,"title":["Implicit Balancing and Regularization: Generalization and Convergence Guarantees for Overparameterized Asymmetric Matrix Sensing"],"prefix":"10.1109","volume":"71","author":[{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0000-0003-2101-6418","authenticated-orcid":false,"given":"Mahdi","family":"Soltanolkotabi","sequence":"first","affiliation":[{"name":"Ming Hsieh Institute of Electrical and Computer Engineering, University of Southern California, Los Angeles, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0000-0002-0543-9456","authenticated-orcid":false,"given":"Dominik","family":"St\u00f6ger","sequence":"additional","affiliation":[{"name":"Department of Mathematics, KU Eichstätt-Ingolstadt, Ingolstadt, Germany"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0009-0002-9983-1738","authenticated-orcid":false,"given":"Changzhi","family":"Xie","sequence":"additional","affiliation":[{"name":"Thomas Lord Department of Computer Science, University of Southern California, Los Angeles, CA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","doi-asserted-by":"publisher","DOI":"10.1145\/3446776"},{"key":"ref2","first-page":"1","article-title":"Implicit bias of gradient descent on linear convolutional networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"31","author":"Gunasekar"},{"key":"ref3","first-page":"2937","article-title":"On lazy training in differentiable programming","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Chizat"},{"key":"ref4","first-page":"2","article-title":"Algorithmic regularization in over-parameterized matrix sensing and neural networks with quadratic activations","volume-title":"Proc. Mach. Learn. Res.","volume":"75","author":"Li"},{"key":"ref5","first-page":"23831","article-title":"Small random initialization is akin to spectral learning: Optimization and generalization guarantees for overparameterized low-rank matrix reconstruction","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"34","author":"St\u00f6ger"},{"key":"ref6","first-page":"5413","article-title":"Neural networks can learn representations with gradient descent","volume-title":"Proc. Conf. Learn. Theory","author":"Damian"},{"key":"ref7","article-title":"Neural networks efficiently learn low-dimensional representations with SGD","author":"Mousavi-Hosseini","year":"2022","journal-title":"arXiv:2209.14863"},{"key":"ref8","first-page":"9768","article-title":"Learning single-index models with shallow neural networks","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Bietti"},{"key":"ref9","first-page":"384","article-title":"Algorithmic regularization in learning deep homogeneous models: Layers are automatically balanced","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"31","author":"Du"},{"key":"ref10","first-page":"1429","article-title":"Global convergence of gradient descent for asymmetric low-rank matrix factorization","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"34","author":"Ye"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1137\/22M1519833"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1137\/070697835"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2011.2111771"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1017\/9781108231596"},{"key":"ref15","first-page":"65","article-title":"Non-square matrix sensing without spurious local minima via the Burer-Monteiro approach","volume-title":"Proc. Artif. Intell. Statist.","author":"Park"},{"key":"ref16","first-page":"1067","article-title":"Gradient descent can take exponential time to escape saddle points","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"30","author":"Du"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.5555\/3045390.3045493"},{"key":"ref18","article-title":"How over-parameterization slows down gradient descent in matrix sensing: The curses of symmetry and initialization","volume-title":"Proc. 12th Int. Conf. Learn. Represent.","author":"Xiong"},{"issue":"169","key":"ref19","first-page":"1","article-title":"On the computational and statistical complexity of over-parameterized matrix sensing","volume":"25","author":"Zhuo","year":"2024","journal-title":"J. Mach. Learn. Res."},{"key":"ref20","article-title":"Non-convex matrix sensing: Breaking the quadratic rank barrier in the sample complexity","author":"St\u00f6ger","year":"2024","journal-title":"arXiv:2408.13276"},{"key":"ref21","first-page":"6152","article-title":"Implicit regularization in matrix factorization","volume-title":"Proc. 31st Int. Conf. Neural Inf. Process. Syst.","author":"Gunasekar"},{"key":"ref22","first-page":"21174","article-title":"Implicit regularization in deep learning may not be explainable by norms","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Razin"},{"key":"ref23","article-title":"Towards resolving the implicit bias of gradient descent for matrix factorization: Greedy low-rank learning","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Li"},{"key":"ref24","first-page":"15200","article-title":"Understanding incremental learning of gradient descent: A fine-grained analysis of matrix sensing","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Jin"},{"key":"ref25","article-title":"A validation approach to over-parameterized matrix and image recovery","author":"Ding","year":"2022","journal-title":"arXiv:2209.10675"},{"key":"ref26","article-title":"The power of preconditioning in overparameterized low-rank matrix sensing","author":"Xu","year":"2023","journal-title":"arXiv:2302.01186"},{"issue":"10","key":"ref27","first-page":"1","article-title":"A line-search descent algorithm for strict saddle functions with complexity guarantees","volume":"24","author":"O\u2019Neill","year":"2023","journal-title":"J. Mach. Learn. Res."},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1093\/imaiai\/iaae009"},{"key":"ref29","article-title":"Sign-RIP: A robust restricted isometry property for low-rank matrix recovery","author":"Ma","year":"2021","journal-title":"arXiv:2102.02969"},{"key":"ref30","first-page":"26767","article-title":"Rank overspecified robust matrix recovery: Subgradient method and exact recovery","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Ding"},{"key":"ref31","first-page":"7411","article-title":"Implicit regularization in deep matrix factorization","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Arora"},{"key":"ref32","first-page":"2968","article-title":"Implicit regularization for optimal sparse recovery","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Va\u0161kevi\u010d ius"},{"key":"ref33","first-page":"3635","article-title":"Kernel and rich regimes in overparametrized models","volume-title":"Proc. Conf. Learn. Theory","author":"Woodworth"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1093\/imaiai\/iaad012"},{"key":"ref35","article-title":"Non-negative least squares via overparametrization","author":"Chou","year":"2022","journal-title":"arXiv:2207.08437"},{"key":"ref36","article-title":"How to induce regularization in linear models: A guide to reparametrizing gradient flow","author":"Chou","year":"2023","journal-title":"arXiv:2308.04921"},{"key":"ref37","first-page":"34626","article-title":"Implicit bias of gradient descent on reparametrized models: On equivalence to mirror descent","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"35","author":"Li"},{"key":"ref38","first-page":"20558","article-title":"Implicit regularization in matrix sensing via mirror descent","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Wu"},{"key":"ref39","first-page":"521","article-title":"Gradient descent with identity initialization efficiently learns positive definite linear transformations by deep residual networks","volume-title":"Proc. Int. Conf. Mach. Learn.","volume":"80","author":"Bartlett"},{"key":"ref40","first-page":"244","article-title":"On the optimization of deep networks: Implicit acceleration by overparameterization","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Arora"},{"key":"ref41","article-title":"A convergence analysis of gradient descent for deep linear neural networks","author":"Arora","year":"2018","journal-title":"arXiv:1810.02281"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1093\/imaiai\/iaaa039"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1016\/j.acha.2023.101595"},{"key":"ref44","article-title":"Convergence of gradient descent for learning linear neural networks","author":"Maxime Nguegnang","year":"2021","journal-title":"arXiv:2108.02040"},{"key":"ref45","first-page":"10153","article-title":"Understanding the dynamics of gradient flow in overparameterized linear models","volume-title":"Proc. 38th Int. Conf. Mach. Learn.","author":"Tarmoun"},{"key":"ref46","first-page":"7760","article-title":"On the explicit role of initialization on the convergence and implicit bias of overparametrized linear networks","volume-title":"Proc. 38th Int. Conf. Mach. Learn.","author":"Min"},{"key":"ref47","article-title":"Large learning rate tames homogeneity: Convergence and balancing effect","author":"Wang","year":"2022","journal-title":"arXiv:2110.03677"},{"key":"ref48","article-title":"Gradient descent on neural networks typically occurs at the edge of stability","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Cohen"},{"issue":"6","key":"ref49","doi-asserted-by":"crossref","first-page":"717","DOI":"10.1007\/s10208-009-9045-5","article-title":"Exact matrix completion via convex optimization","volume":"9","author":"Cand\u00e8s","year":"2009","journal-title":"Found. Comput. Math."},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2010.2044061"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1002\/cpa.21432"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1137\/110848074"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2013.2294644"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2017.2701342"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1109\/JSTSP.2016.2539100"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-09745-4_2"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2019.2937282"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1109\/tit.2020.2992234"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.2015.2399924"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.1002\/cpa.21638"},{"key":"ref61","doi-asserted-by":"publisher","DOI":"10.1007\/s10208-019-09429-9"},{"key":"ref62","doi-asserted-by":"publisher","DOI":"10.1016\/j.acha.2018.01.001"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1093\/imaiai\/iax022"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2021.3051425"},{"key":"ref65","doi-asserted-by":"publisher","DOI":"10.1007\/s10208-017-9365-9"},{"key":"ref66","first-page":"2973","article-title":"Matrix completion has no spurious local minimum","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"29","author":"Ge"},{"issue":"114","key":"ref67","first-page":"1","article-title":"Sharp restricted isometry bounds for the inexistence of spurious local minima in nonconvex matrix recovery","volume":"20","author":"Zhang","year":"2019","journal-title":"J. Mach. Learn. Res."},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.1007\/s10107-006-0706-8"},{"key":"ref69","doi-asserted-by":"publisher","DOI":"10.1007\/978-0-387-40065-5_4"},{"key":"ref70","first-page":"1724","article-title":"How to escape saddle points efficiently","volume-title":"Proc. 34th Int. Conf. Mach. Learn.","author":"Jin"},{"key":"ref71","first-page":"797","article-title":"Escaping from saddle points\u2014Online stochastic gradient for tensor decomposition","volume-title":"Proc. 28th Conf. Learn. Theory","author":"Ge"},{"key":"ref72","first-page":"1674","article-title":"Non-convex learning via stochastic gradient Langevin dynamics: A nonasymptotic analysis","volume-title":"Proc. Conf. Learn. Theory","volume":"65","author":"Raginsky"},{"key":"ref73","first-page":"1980","article-title":"A hitting time analysis of stochastic gradient Langevin dynamics","volume-title":"Proc. Conf. Learn. Theory","volume":"65","author":"Zhang"},{"key":"ref74","doi-asserted-by":"publisher","DOI":"10.1007\/s10107-019-01363-6"},{"key":"ref75","doi-asserted-by":"publisher","DOI":"10.1137\/22m1506456"},{"key":"ref76","article-title":"On the global convergence of gradient descent for multi-layer ResNets in the mean-field regime","author":"Ding","year":"2021","journal-title":"arXiv:2110.02926"},{"key":"ref77","doi-asserted-by":"publisher","DOI":"10.1109\/JSAIT.2020.2991332"},{"key":"ref78","first-page":"1675","article-title":"Gradient descent finds global minima of deep neural networks","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Du"},{"key":"ref79","first-page":"4951","article-title":"Overparameterized nonlinear learning: Gradient descent takes the shortest path?","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Oymak"},{"key":"ref80","doi-asserted-by":"publisher","DOI":"10.1137\/0707001"},{"key":"ref81","doi-asserted-by":"publisher","DOI":"10.1002\/cpa.20294"}],"container-title":["IEEE Transactions on Information Theory"],"original-title":[],"link":[{"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/ieeexplore.ieee.org\/ielam\/18\/10935829\/10843299-aam.pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/2.zoppoz.workers.dev:443\/http\/xplorestaging.ieee.org\/ielx8\/18\/10935829\/10843299.pdf?arnumber=10843299","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,3,27]],"date-time":"2025-03-27T02:28:51Z","timestamp":1743042531000},"score":1,"resource":{"primary":{"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/ieeexplore.ieee.org\/document\/10843299\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,4]]},"references-count":81,"journal-issue":{"issue":"4"},"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/doi.org\/10.1109\/tit.2025.3530335","relation":{},"ISSN":["0018-9448","1557-9654"],"issn-type":[{"value":"0018-9448","type":"print"},{"value":"1557-9654","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,4]]}}}