{"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/regularization-of-deep-neural-networks-with","title":"Regularization of Deep Neural Networks with Spectral Dropout","arxiv_id":"1711.08591","date":"2017-11-23","proceeding":null,"authors":["Salman Khan","Munawar Hayat","Fatih Porikli"],"abstract":"The big breakthrough on the ImageNet challenge in 2012 was partially due to\nthe `dropout' technique used to avoid overfitting. Here, we introduce a new\napproach called `Spectral Dropout' to improve the generalization ability of\ndeep neural networks. We cast the proposed approach in the form of regular\nConvolutional Neural Network (CNN) weight layers using a decorrelation\ntransform with fixed basis functions. Our spectral dropout method prevents\noverfitting by eliminating weak and `noisy' Fourier domain coefficients of the\nneural network activations, leading to remarkably better results than the\ncurrent regularization methods. Furthermore, the proposed is very efficient due\nto the fixed basis functions used for spectral transformation. In particular,\ncompared to Dropout and Drop-Connect, our method significantly speeds up the\nnetwork convergence rate during the training process (roughly x2), with\nconsiderably higher neuron pruning rates (an increase of ~ 30%). We demonstrate\nthat the spectral dropout can also be used in conjunction with other\nregularization approaches resulting in additional performance gains.","url_abs":"http://arxiv.org/abs/1711.08591v1","url_pdf":"http://arxiv.org/pdf/1711.08591v1.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":[],"methods":[{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"pruning","method_name":"Pruning"},{"method_slug":"spectral-dropout","method_name":"Spectral Dropout"}],"datasets_introduced":[],"methods_introduced":[{"slug":"spectral-dropout","name":"Spectral Dropout","full_name":"Spectral Dropout"}],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1711.08591","atlas_url":"https://app.syntology.ai/?focus=1711.08591","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}