{"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/unsupervised-learning-via-meta-learning","title":"Unsupervised Learning via Meta-Learning","arxiv_id":"1810.02334","date":"2018-10-04","proceeding":"ICLR 2019 5","authors":["Kyle Hsu","Sergey Levine","Chelsea Finn"],"abstract":"A central goal of unsupervised learning is to acquire representations from\nunlabeled data or experience that can be used for more effective learning of\ndownstream tasks from modest amounts of labeled data. Many prior unsupervised\nlearning works aim to do so by developing proxy objectives based on\nreconstruction, disentanglement, prediction, and other metrics. Instead, we\ndevelop an unsupervised meta-learning method that explicitly optimizes for the\nability to learn a variety of tasks from small amounts of data. To do so, we\nconstruct tasks from unlabeled data in an automatic way and run meta-learning\nover the constructed tasks. Surprisingly, we find that, when integrated with\nmeta-learning, relatively simple task construction mechanisms, such as\nclustering embeddings, lead to good performance on a variety of downstream,\nhuman-specified tasks. Our experiments across four image datasets indicate that\nour unsupervised meta-learning approach acquires a learning algorithm without\nany labeled data that is applicable to a wide range of downstream\nclassification tasks, improving upon the embedding learned by four prior\nunsupervised learning methods.","url_abs":"http://arxiv.org/abs/1810.02334v6","url_pdf":"http://arxiv.org/pdf/1810.02334v6.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":[],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"disentanglement","task_name":"Disentanglement"},{"task_slug":"meta-learning","task_name":"Meta-Learning"},{"task_slug":"unsupervised-few-shot-image-classification","task_name":"Unsupervised Few-Shot Image Classification"},{"task_slug":"unsupervised-few-shot-learning","task_name":"Unsupervised Few-Shot Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/unsupervised-few-shot-image-classification-on","task":"Unsupervised Few-Shot Image Classification","dataset":"Mini-Imagenet 5-way (1-shot)","model":"CACTU","rank_in_archive_order":27,"of":28,"metrics":{"Accuracy":"39.90"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-few-shot-image-classification-on-1","task":"Unsupervised Few-Shot Image Classification","dataset":"Mini-Imagenet 5-way (5-shot)","model":"CACTU","rank_in_archive_order":26,"of":28,"metrics":{"Accuracy":"53.97"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1810.02334","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}