{"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/improving-back-propagation-by-adding-an","title":"Improving Back-Propagation by Adding an Adversarial Gradient","arxiv_id":"1510.04189","date":"2015-10-14","proceeding":null,"authors":["Arild Nøkland"],"abstract":"The back-propagation algorithm is widely used for learning in artificial\nneural networks. A challenge in machine learning is to create models that\ngeneralize to new data samples not seen in the training data. Recently, a\ncommon flaw in several machine learning algorithms was discovered: small\nperturbations added to the input data lead to consistent misclassification of\ndata samples. Samples that easily mislead the model are called adversarial\nexamples. Training a \"maxout\" network on adversarial examples has shown to\ndecrease this vulnerability, but also increase classification performance. This\npaper shows that adversarial training has a regularizing effect also in\nnetworks with logistic, hyperbolic tangent and rectified linear units. A simple\nextension to the back-propagation method is proposed, that adds an adversarial\ngradient to the training. The extension requires an additional forward and\nbackward pass to calculate a modified input sample, or mini batch, used as\ninput for standard back-propagation learning. The first experimental results on\nMNIST show that the \"adversarial back-propagation\" method increases the\nresistance to adversarial examples and boosts the classification performance.\nThe extension reduces the classification error on the permutation invariant\nMNIST from 1.60% to 0.95% in a logistic network, and from 1.40% to 0.78% in a\nnetwork with rectified linear units. Results on CIFAR-10 indicate that the\nmethod has a regularizing effect similar to dropout in fully connected\nnetworks. Based on these promising results, adversarial back-propagation is\nproposed as a stand-alone regularizing method that should be further\ninvestigated.","url_abs":"http://arxiv.org/abs/1510.04189v2","url_pdf":"http://arxiv.org/pdf/1510.04189v2.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":"improving-back-propagation-by-adding-an","repo_url":"https://github.com/axelbrando/Mixture-Density-Networks-for-distribution-and-uncertainty-estimation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"machine-learning","task_name":"BIG-bench Machine Learning"},{"task_slug":"classification","task_name":"General Classification"}],"methods":[{"method_slug":"dropout","method_name":"Dropout"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1510.04189","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}