{"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/meta-learning-update-rules-for-unsupervised","title":"Meta-Learning Update Rules for Unsupervised Representation Learning","arxiv_id":"1804.00222","date":"2018-03-31","proceeding":"ICLR 2019 5","authors":["Luke Metz","Niru Maheswaranathan","Brian Cheung","Jascha Sohl-Dickstein"],"abstract":"A major goal of unsupervised learning is to discover data representations\nthat are useful for subsequent tasks, without access to supervised labels\nduring training. Typically, this involves minimizing a surrogate objective,\nsuch as the negative log likelihood of a generative model, with the hope that\nrepresentations useful for subsequent tasks will arise as a side effect. In\nthis work, we propose instead to directly target later desired tasks by\nmeta-learning an unsupervised learning rule which leads to representations\nuseful for those tasks. Specifically, we target semi-supervised classification\nperformance, and we meta-learn an algorithm -- an unsupervised weight update\nrule -- that produces representations useful for this task. Additionally, we\nconstrain our unsupervised update rule to a be a biologically-motivated,\nneuron-local function, which enables it to generalize to different neural\nnetwork architectures, datasets, and data modalities. We show that the\nmeta-learned update rule produces useful features and sometimes outperforms\nexisting unsupervised learning techniques. We further show that the\nmeta-learned unsupervised update rule generalizes to train networks with\ndifferent widths, depths, and nonlinearities. It also generalizes to train on\ndata with randomly permuted input dimensions and even generalizes from image\ndatasets to a text task.","url_abs":"http://arxiv.org/abs/1804.00222v3","url_pdf":"http://arxiv.org/pdf/1804.00222v3.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":"meta-learning-update-rules-for-unsupervised","repo_url":"https://github.com/tensorflow/models","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null},{"paper_slug":"meta-learning-update-rules-for-unsupervised","repo_url":"https://github.com/tensorflow/models/tree/master/research/learning_unsupervised_learning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"meta-learning","task_name":"Meta-Learning"},{"task_slug":"representation-learning","task_name":"Representation Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.00222","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}