{"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/learned-optimizers-that-scale-and-generalize","title":"Learned Optimizers that Scale and Generalize","arxiv_id":"1703.04813","date":"2017-03-14","proceeding":"ICML 2017 8","authors":["Olga Wichrowska","Niru Maheswaranathan","Matthew W. Hoffman","Sergio Gomez Colmenarejo","Misha Denil","Nando de Freitas","Jascha Sohl-Dickstein"],"abstract":"Learning to learn has emerged as an important direction for achieving\nartificial intelligence. Two of the primary barriers to its adoption are an\ninability to scale to larger problems and a limited ability to generalize to\nnew tasks. We introduce a learned gradient descent optimizer that generalizes\nwell to new tasks, and which has significantly reduced memory and computation\noverhead. We achieve this by introducing a novel hierarchical RNN architecture,\nwith minimal per-parameter overhead, augmented with additional architectural\nfeatures that mirror the known structure of optimization tasks. We also develop\na meta-training ensemble of small, diverse optimization tasks capturing common\nproperties of loss landscapes. The optimizer learns to outperform RMSProp/ADAM\non problems in this corpus. More importantly, it performs comparably or better\nwhen applied to small convolutional neural networks, despite seeing no neural\nnetworks in its meta-training set. Finally, it generalizes to train Inception\nV3 and ResNet V2 architectures on the ImageNet dataset for thousands of steps,\noptimization problems that are of a vastly different scale than those it was\ntrained on. We release an open source implementation of the meta-training\nalgorithm.","url_abs":"http://arxiv.org/abs/1703.04813v4","url_pdf":"http://arxiv.org/pdf/1703.04813v4.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":"learned-optimizers-that-scale-and-generalize","repo_url":"https://github.com/Kolin96/learning-to-learn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"bottleneck-residual-block","method_name":"Bottleneck Residual Block"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1703.04813","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}