{"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/sparse-unsupervised-capsules-generalize","title":"Sparse Unsupervised Capsules Generalize Better","arxiv_id":"1804.06094","date":"2018-04-17","proceeding":null,"authors":["David Rawlinson","Abdelrahman Ahmed","Gideon Kowadlo"],"abstract":"We show that unsupervised training of latent capsule layers using only the\nreconstruction loss, without masking to select the correct output class, causes\na loss of equivariances and other desirable capsule qualities. This implies\nthat supervised capsules networks can't be very deep. Unsupervised sparsening\nof latent capsule layer activity both restores these qualities and appears to\ngeneralize better than supervised masking, while potentially enabling deeper\ncapsules networks. We train a sparse, unsupervised capsules network of similar\ngeometry to Sabour et al (2017) on MNIST, and then test classification accuracy\non affNIST using an SVM layer. Accuracy is improved from benchmark 79% to 90%.","url_abs":"http://arxiv.org/abs/1804.06094v1","url_pdf":"http://arxiv.org/pdf/1804.06094v1.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":"sparse-unsupervised-capsules-generalize","repo_url":"https://github.com/ProjectAGI/sparse-unsupervised-capsules","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"}],"methods":[{"method_slug":"svm","method_name":"SVM"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.06094","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}