{"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/building-high-level-features-using-large","title":"Building high-level features using large scale unsupervised learning","arxiv_id":"1112.6209","date":"2011-12-29","proceeding":null,"authors":["Quoc V. Le","Marc'Aurelio Ranzato","Rajat Monga","Matthieu Devin","Kai Chen","Greg S. Corrado","Jeff Dean","Andrew Y. Ng"],"abstract":"We consider the problem of building high-level, class-specific feature\ndetectors from only unlabeled data. For example, is it possible to learn a face\ndetector using only unlabeled images? To answer this, we train a 9-layered\nlocally connected sparse autoencoder with pooling and local contrast\nnormalization on a large dataset of images (the model has 1 billion\nconnections, the dataset has 10 million 200x200 pixel images downloaded from\nthe Internet). We train this network using model parallelism and asynchronous\nSGD on a cluster with 1,000 machines (16,000 cores) for three days. Contrary to\nwhat appears to be a widely-held intuition, our experimental results reveal\nthat it is possible to train a face detector without having to label images as\ncontaining a face or not. Control experiments show that this feature detector\nis robust not only to translation but also to scaling and out-of-plane\nrotation. We also find that the same network is sensitive to other high-level\nconcepts such as cat faces and human bodies. Starting with these learned\nfeatures, we trained our network to obtain 15.8% accuracy in recognizing 20,000\nobject categories from ImageNet, a leap of 70% relative improvement over the\nprevious state-of-the-art.","url_abs":"http://arxiv.org/abs/1112.6209v5","url_pdf":"http://arxiv.org/pdf/1112.6209v5.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":"building-high-level-features-using-large","repo_url":"https://github.com/greenelab/tybalt","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"tasks":[{"task_slug":"high","task_name":"Vocal Bursts Intensity Prediction"}],"methods":[{"method_slug":"sparse-autoencoder","method_name":"Sparse Autoencoder"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1112.6209","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}