{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,1,16]],"date-time":"2026-01-16T09:28:58Z","timestamp":1768555738297,"version":"3.49.0"},"reference-count":15,"publisher":"MIT Press - Journals","issue":"4","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Neural Computation"],"published-print":{"date-parts":[[2017,4]]},"abstract":"<jats:p> Many previous proposals for adversarial training of deep neural nets have included directly modifying the gradient, training on a mix of original and adversarial examples, using contractive penalties, and approximately optimizing constrained adversarial objective functions. In this article, we show that these proposals are actually all instances of optimizing a general, regularized objective we call DataGrad. Our proposed DataGrad framework, which can be viewed as a deep extension of the layerwise contractive autoencoder penalty, cleanly simplifies prior work and easily allows extensions such as adversarial training with multitask cues. In our experiments, we find that the deep gradient regularization of DataGrad (which also has L1 and L2 flavors of regularization) outperforms alternative forms of regularization, including classical L1, L2, and multitask, on both the original data set and adversarial sets. Furthermore, we find that combining multitask optimization with DataGrad adversarial training results in the most robust performance. <\/jats:p>","DOI":"10.1162\/neco_a_00928","type":"journal-article","created":{"date-parts":[[2017,1,17]],"date-time":"2017-01-17T21:43:31Z","timestamp":1484689411000},"page":"867-887","source":"Crossref","is-referenced-by-count":19,"title":["Unifying Adversarial Training Algorithms with Data Gradient Regularization"],"prefix":"10.1162","volume":"29","author":[{"given":"Alexander G.","family":"Ororbia II","sequence":"first","affiliation":[{"name":"Pennsylvania State University, University Park, PA 16802, U.S.A."}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Daniel","family":"Kifer","sequence":"additional","affiliation":[{"name":"Pennsylvania State University, University Park, PA 16802, U.S.A."}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"C. Lee","family":"Giles","sequence":"additional","affiliation":[{"name":"Pennsylvania State University, University Park, PA 16802, U.S.A."}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"281","reference":[{"key":"B1","volume":"19","author":"Bengio Y.","year":"2007","journal-title":"Advances in neural information processing systems"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1162\/neco.1992.4.4.494"},{"key":"B3","first-page":"315","volume":"15","author":"Glorot X.","year":"2011","journal-title":"JMLR W & CP"},{"key":"B4","author":"Goodfellow I. J.","year":"2014","journal-title":"Explaining and harnessing adversarial examples"},{"key":"B5","author":"Gu S.","year":"2014","journal-title":"Towards deep neural network architectures robust to adversarial examples"},{"key":"B6","first-page":"1026","author":"He K.","year":"2015","journal-title":"Proc. IEEE Conference on Computer Vision"},{"key":"B7","author":"Huang R.","year":"2015","journal-title":"Learning with a strong adversary"},{"key":"B8","author":"Lyu C.","year":"2015","journal-title":"A unified gradient regularization family for adversarial examples."},{"key":"B9","author":"Miyato T.","year":"2015","journal-title":"Distributional smoothing with virtual adversarial training"},{"key":"B10","author":"Nguyen A.","year":"2014","journal-title":"Deep neural networks are easily fooled: High confidence predictions for unrecognizable images"},{"key":"B11","author":"Nkland A.","year":"2015","journal-title":"Improving back-propagation by adding an adversarial gradient"},{"key":"B12","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-23528-8_32"},{"key":"B13","doi-asserted-by":"publisher","DOI":"10.1162\/neco.1994.6.1.147"},{"key":"B14","first-page":"833","author":"Rifai S.","year":"2011","journal-title":"Proc. of the 28th International Conference on Machine Learning (ICML-11)"},{"key":"B15","author":"Szegedy C.","year":"2013","journal-title":"Intriguing properties of neural networks"}],"container-title":["Neural Computation"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mitpressjournals.org\/doi\/pdf\/10.1162\/NECO_a_00928","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2021,3,12]],"date-time":"2021-03-12T21:41:42Z","timestamp":1615585302000},"score":1,"resource":{"primary":{"URL":"https:\/\/direct.mit.edu\/neco\/article\/29\/4\/867-887\/8255"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2017,4]]},"references-count":15,"journal-issue":{"issue":"4","published-print":{"date-parts":[[2017,4]]}},"alternative-id":["10.1162\/NECO_a_00928"],"URL":"https:\/\/doi.org\/10.1162\/neco_a_00928","relation":{},"ISSN":["0899-7667","1530-888X"],"issn-type":[{"value":"0899-7667","type":"print"},{"value":"1530-888X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2017,4]]}}}