{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T16:57:14Z","timestamp":1791565034705,"version":"4.3.4"},"reference-count":54,"publisher":"National Academy of Sciences","issue":"41","license":[{"start":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T00:00:00Z","timestamp":1791504000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000001","name":"National Science Foundation","doi-asserted-by":"publisher","award":["CCF-1231216"],"award-info":[{"award-number":["CCF-1231216"]}],"id":[{"id":"10.13039\/100000001","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["www.pnas.org"],"crossmark-restriction":true},"short-container-title":["Proc. Natl. Acad. Sci. U.S.A."],"published-print":{"date-parts":[[2026,10,13]]},"abstract":"<jats:p>Existing models of learning in the brain explain how given circuits learn, but not how biology could assemble those circuits in the first place. To address this gap, we formulate Self-Assembling Learning\u2014the study of how learning systems can emerge from lower-level interactions\u2014and introduce one example mechanism, the Self-Assembling Motif (SAM). SAM is self-assembling at two scales: The motif emerges from initially random connectivity under heterosynaptic plasticity rules, and networks of SAMs, composed hierarchically, self-organize into dynamics that provably approximate a generalized form of stochastic gradient descent\u2014matching backpropagation-level performance. This suggests that biological learning need not be prescribed but can emerge from local rules\u2014and to a far greater extent than previously thought.<\/jats:p>","DOI":"10.1073\/pnas.2606245123","type":"journal-article","created":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T16:31:21Z","timestamp":1791563481000},"update-policy":"https:\/\/doi.org\/10.1073\/pnas.cm10313","source":"Crossref","is-referenced-by-count":0,"title":["How biological synapses self-assemble gradient learning"],"prefix":"10.1073","volume":"123","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-0076-621X","authenticated-orcid":false,"given":"Qianli","family":"Liao","sequence":"first","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"K. Lisa Yang Integrative Computational Neuroscience Center, Yang Tan Collective, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Center for Brains, Minds, and Machines, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"McGovern Institute for Brain Research, Massachusetts Institute of Technology","place":["Cambridge"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Liu","family":"Ziyin","sequence":"additional","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Research Laboratory of Electronics, Massachusetts Institute of Technology","place":["Cambridge"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yulu","family":"Gan","sequence":"additional","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Center for Brains, Minds, and Machines, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology","place":["Cambridge"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Brian","family":"Cheung","sequence":"additional","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Center for Brains, Minds, and Machines, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology","place":["Cambridge"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mark T.","family":"Harnett","sequence":"additional","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"K. Lisa Yang Integrative Computational Neuroscience Center, Yang Tan Collective, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"McGovern Institute for Brain Research, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology","place":["Cambridge"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3944-0455","authenticated-orcid":false,"given":"Tomaso","family":"Poggio","sequence":"additional","affiliation":[{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"K. Lisa Yang Integrative Computational Neuroscience Center, Yang Tan Collective, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Center for Brains, Minds, and Machines, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"McGovern Institute for Brain Research, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042nb2s44","id-type":"ROR","asserted-by":"publisher"}],"name":"Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology","place":["Cambridge"]},{"id":[{"id":"https:\/\/ror.org\/042t93s57","id-type":"ROR","asserted-by":"publisher"}],"name":"Italian Institute of Technology","place":["Genova"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"341","published-online":{"date-parts":[[2026,10,9]]},"reference":[{"key":"e_1_3_4_1_2","doi-asserted-by":"publisher","DOI":"10.1038\/srep27755"},{"key":"e_1_3_4_2_2","doi-asserted-by":"publisher","DOI":"10.1038\/nn.4244"},{"key":"e_1_3_4_3_2","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.5023-14.2015"},{"key":"e_1_3_4_4_2","doi-asserted-by":"publisher","DOI":"10.1146\/annurev-vision-082114-035447"},{"key":"e_1_3_4_5_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41583-020-0277-3"},{"key":"e_1_3_4_6_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.tics.2018.12.005"},{"key":"e_1_3_4_7_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-024-08548-w"},{"key":"e_1_3_4_8_2","unstructured":"J. Raugel et al. Disentangling the factors of convergence between brains and computer vision models. arXiv [Preprint] (2025). https:\/\/arxiv.org\/abs\/2508.18226 (Accessed 31 August 2026)."},{"key":"e_1_3_4_9_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41593-024-01668-6"},{"key":"e_1_3_4_10_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41593-020-00753-w"},{"key":"e_1_3_4_11_2","doi-asserted-by":"crossref","unstructured":"Q. Liao J. Leibo T. Poggio \u201cHow important is weight symmetry in backpropagation?\u201d in Proceedings of the AAAI Conference on Artificial Intelligence 2016 D. Schuurmans M. P. Wellman Eds. (AAAI Press Palo Alto CA 2016).","DOI":"10.1609\/aaai.v30i1.10279"},{"key":"e_1_3_4_12_2","doi-asserted-by":"publisher","DOI":"10.1038\/ncomms13276"},{"key":"e_1_3_4_13_2","unstructured":"M. Akrout C. Wilson P. Humphreys T. Lillicrap D. B. Tweed \u201cDeep learning without weight transport\u201d in Advances in Neural Information Processing Systems H. Wallach et al. Eds. (Curran Associates Inc. Red Hook NY 2019) vol. 32."},{"key":"e_1_3_4_14_2","doi-asserted-by":"crossref","unstructured":"J. Kolen J. Pollack \u201cBack-propagation without weight transport\u201d in Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN\u201994) (IEEE Piscataway NJ 1994) vol. 3 pp. 1375\u20131380.","DOI":"10.1109\/ICNN.1994.374486"},{"key":"e_1_3_4_15_2","unstructured":"A. N\u00f8kland \u201cDirect feedback alignment provides learning in deep neural networks\u201d in Advances in Neural Information Processing Systems D. D. Lee M. Sugiyama U. von Luxburg I. Guyon R. Garnett Eds. (Curran Associates Inc. Red Hook NY 2016) vol. 29."},{"key":"e_1_3_4_16_2","unstructured":"A. N\u00f8kland L. H. Eidnes \u201cTraining neural networks with local error signals\u201d in International Conference on Machine Learning K. Chaudhuri R. Salakhutdinov Eds. (PMLR 2019) pp. 4839\u20134850."},{"key":"e_1_3_4_17_2","unstructured":"W. Xiao H. Chen Q. Liao T. Poggio \u201cBiologically-plausible learning algorithms can scale to large datasets\u201d in International Conference on Learning Representations (ICLR) 2019 (International Conference on Learning Representations 2019)."},{"key":"e_1_3_4_18_2","first-page":"8721","article-title":"Dendritic cortical microcircuits approximate the backpropagation algorithm","volume":"31","author":"Sacramento J.","year":"2018","unstructured":"J. Sacramento, R. Ponte Costa, Y. Bengio, W. Senn, Dendritic cortical microcircuits approximate the backpropagation algorithm. Adv. Neural Inf. Process. Syst. 31, 8721\u20138732 (2018).","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"e_1_3_4_19_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41593-021-00857-x"},{"key":"e_1_3_4_20_2","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1758"},{"key":"e_1_3_4_21_2","doi-asserted-by":"publisher","DOI":"10.1038\/s42256-024-00845-3"},{"key":"e_1_3_4_22_2","unstructured":"M. Cameron Y. Chen T. J. Sejnowski A biologically-plausible alternative to backpropagation using pseudoinverse feedback. OpenReview [Preprint] (2025). https:\/\/openreview.net\/forum?id=B5Dj4EhZPP (Accessed 31 August 2026)."},{"key":"e_1_3_4_23_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-37562-1"},{"key":"e_1_3_4_24_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-40141-z"},{"key":"e_1_3_4_25_2","doi-asserted-by":"crossref","unstructured":"R. Abel S. Ullman Biologically inspired learning model for instructed vision. Adv. Neural Inf. Process. Syst. 37 45315\u201345358 (2024).","DOI":"10.52202\/079017-1440"},{"key":"e_1_3_4_26_2","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.5757-08.2009"},{"key":"e_1_3_4_27_2","doi-asserted-by":"publisher","DOI":"10.1016\/S0166-2236(96)80018-X"},{"key":"e_1_3_4_28_2","doi-asserted-by":"publisher","DOI":"10.1038\/nrn2356"},{"key":"e_1_3_4_29_2","doi-asserted-by":"publisher","DOI":"10.1146\/annurev-vision-121219-081716"},{"key":"e_1_3_4_30_2","unstructured":"A. Krizhevsky \u201cLearning multiple layers of features from tiny images\u201d (Tech. Rep. 2009)."},{"key":"e_1_3_4_31_2","unstructured":"Y. LeCun \u201cThe MNIST database of handwritten digits\u201d (Tech. Rep. 1998)."},{"key":"e_1_3_4_32_2","unstructured":"H. Xiao K. Rasul R. Vollgraf Fashion-MNIST: A novel image dataset for benchmarking machine learning algorithms. arXiv [Preprint] (2017). https:\/\/arxiv.org\/abs\/1708.07747 (Accessed 31 August 2026)."},{"key":"e_1_3_4_33_2","doi-asserted-by":"crossref","unstructured":"J. Yang R. Shi B. Ni \u201cMedMNIST classification decathlon: A lightweight AutoML benchmark for medical image analysis\u201d in IEEE 18th International Symposium on Biomedical Imaging (ISBI) (IEEE Piscataway NJ 2021) pp. 191\u2013195.","DOI":"10.1109\/ISBI48211.2021.9434062"},{"key":"e_1_3_4_34_2","doi-asserted-by":"crossref","unstructured":"J. Yang et al. MedMNIST v2\u2014A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Sci. Data 10 41 (2023).","DOI":"10.1038\/s41597-022-01721-8"},{"key":"e_1_3_4_35_2","unstructured":"Y. Netzer et al. \u201cReading digits in natural images with unsupervised feature learning\u201d in NIPS Workshop on Deep Learning and Unsupervised Feature Learning (Granada 2011) vol. 2011 p. 4."},{"key":"e_1_3_4_36_2","unstructured":"A. Coates A. Ng H. Lee \u201cAn analysis of single-layer networks in unsupervised feature learning\u201d in Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics G. Gordon D. Dunson M. Dud\u00edk Eds. (JMLR Workshop and Conference Proceedings 2011) pp. 215\u2013223."},{"key":"e_1_3_4_37_2","unstructured":"O. Roy M. Vetterli \u201cThe effective rank: A measure of effective dimensionality\u201d in 2007 15th European Signal Processing Conference (IEEE 2007) pp. 606\u2013610."},{"key":"e_1_3_4_38_2","unstructured":"L. Ziyin I. Chuang T. Galanti T. Poggio \u201cFormation of representations in neural networks\u201d in Proceedings of the International Conference on Learning Representations (International Conference on Learning Representations Singapore 2025)."},{"key":"e_1_3_4_39_2","unstructured":"M. Huh et al. The low-rank simplicity bias in deep networks. Trans. Mach. Learn. Res. (2023)."},{"key":"e_1_3_4_40_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.tins.2011.02.001"},{"key":"e_1_3_4_41_2","doi-asserted-by":"crossref","unstructured":"M. Tjia X. Yu L. S. Jammu J. Lu Y. Zuo Pyramidal neurons in different cortical layers exhibit distinct dynamics and plasticity of apical dendritic spines. Front. Neural Circuits 11 43 (2017).","DOI":"10.3389\/fncir.2017.00043"},{"key":"e_1_3_4_42_2","doi-asserted-by":"publisher","DOI":"10.1038\/nature08577"},{"key":"e_1_3_4_43_2","doi-asserted-by":"publisher","DOI":"10.1038\/nature08389"},{"key":"e_1_3_4_44_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuron.2005.01.003"},{"key":"e_1_3_4_45_2","doi-asserted-by":"publisher","DOI":"10.1038\/nn.4061"},{"key":"e_1_3_4_46_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.tins.2012.11.006"},{"key":"e_1_3_4_47_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuron.2018.09.001"},{"key":"e_1_3_4_48_2","doi-asserted-by":"publisher","DOI":"10.1126\/science.aaz3136"},{"key":"e_1_3_4_49_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-026-10190-7"},{"key":"e_1_3_4_50_2","unstructured":"L. Ziyin I. Chuang T. Poggio Heterosynaptic circuits are universal gradient machines. arXiv [Preprint] (2025). https:\/\/arxiv.org\/abs\/2505.02248 (Accessed 31 August 2026)."},{"key":"e_1_3_4_51_2","unstructured":"M. G. Bacvanski L. Ziyin T. Poggio Does feedback alignment work at biological timescales? arXiv [Preprint] (2026). https:\/\/arxiv.org\/abs\/2510.18808v2 (Accessed 31 August 2026)."},{"key":"e_1_3_4_52_2","unstructured":"M. G. Bacvanski L. Ziyin T. Poggio \u201cContinuous-time heteroassociative memory at biological timescales\u201d in New Frontiers in Associative Memory Workshop at ICLR (2026)."},{"key":"e_1_3_4_53_2","unstructured":"M. G. Bacvanski L. Ziyin T. Poggio \u201cStabilizing continuous-time Kolen\u2013Pollack learning with a scale-balance condition\u201d in High-dimensional Learning Dynamics Workshop at ICML 2026 (2026)."},{"key":"e_1_3_4_54_2","unstructured":"Q. Liao PNAS 2026 code release. Github. https:\/\/github.com\/liaoq\/SAM2026. Deposited 3 July 2026."}],"container-title":["Proceedings of the National Academy of Sciences"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/pnas.org\/doi\/pdf\/10.1073\/pnas.2606245123","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T16:31:57Z","timestamp":1791563517000},"score":1,"resource":{"primary":{"URL":"https:\/\/pnas.org\/doi\/10.1073\/pnas.2606245123"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,9]]},"references-count":54,"journal-issue":{"issue":"41","published-print":{"date-parts":[[2026,10,13]]}},"alternative-id":["10.1073\/pnas.2606245123"],"URL":"https:\/\/doi.org\/10.1073\/pnas.2606245123","relation":{},"ISSN":["0027-8424","1091-6490"],"issn-type":[{"value":"0027-8424","type":"print"},{"value":"1091-6490","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,9]]},"assertion":[{"value":"2026-02-20","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-08-18","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-10-09","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e2606245123"}}