{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/how-important-is-weight-symmetry-in","title":"How Important is Weight Symmetry in Backpropagation?","arxiv_id":"1510.05067","date":"2015-10-17","proceeding":null,"authors":["Qianli Liao","Joel Z. Leibo","Tomaso Poggio"],"abstract":"Gradient backpropagation (BP) requires symmetric feedforward and feedback\nconnections -- the same weights must be used for forward and backward passes.\nThis \"weight transport problem\" (Grossberg 1987) is thought to be one of the\nmain reasons to doubt BP's biologically plausibility. Using 15 different\nclassification datasets, we systematically investigate to what extent BP really\ndepends on weight symmetry. In a study that turned out to be surprisingly\nsimilar in spirit to Lillicrap et al.'s demonstration (Lillicrap et al. 2014)\nbut orthogonal in its results, our experiments indicate that: (1) the\nmagnitudes of feedback weights do not matter to performance (2) the signs of\nfeedback weights do matter -- the more concordant signs between feedforward and\ntheir corresponding feedback connections, the better (3) with feedback weights\nhaving random magnitudes and 100% concordant signs, we were able to achieve the\nsame or even better performance than SGD. (4) some\nnormalizations/stabilizations are indispensable for such asymmetric BP to work,\nnamely Batch Normalization (BN) (Ioffe and Szegedy 2015) and/or a \"Batch\nManhattan\" (BM) update rule.","url_abs":"http://arxiv.org/abs/1510.05067v4","url_pdf":"http://arxiv.org/pdf/1510.05067v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"how-important-is-weight-symmetry-in","repo_url":"https://github.com/jsalbert/biotorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"how-important-is-weight-symmetry-in","repo_url":"https://github.com/willwx/sign-symmetry","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"handwritten-digit-recognition","task_name":"Handwritten Digit Recognition"},{"task_slug":"image-classification","task_name":"Image Classification"}],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"sgd","method_name":"SGD"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/handwritten-digit-recognition-on-mnist","task":"Handwritten Digit Recognition","dataset":"MNIST","model":"Sign-symmetry","rank_in_archive_order":2,"of":2,"metrics":{"PERCENTAGE ERROR":"0.91"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-10","task":"Image Classification","dataset":"CIFAR-10","model":"Sign-symmetry","rank_in_archive_order":246,"of":265,"metrics":{"Percentage correct":"80.98"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"Sign-symmetry","rank_in_archive_order":203,"of":211,"metrics":{"Percentage correct":"48.75"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-stl-10","task":"Image Classification","dataset":"STL-10","model":"Sign-symmetry","rank_in_archive_order":115,"of":117,"metrics":{"Percentage correct":"57.32"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-svhn","task":"Image Classification","dataset":"SVHN","model":"Sign-symmetry","rank_in_archive_order":46,"of":62,"metrics":{"Percentage error":"10.16"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1510.05067","atlas_url":"https://app.syntology.ai/?focus=1510.05067","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}