{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,7]],"date-time":"2026-06-07T04:08:42Z","timestamp":1780805322460,"version":"3.54.1"},"reference-count":50,"publisher":"Springer Science and Business Media LLC","issue":"24","license":[{"start":{"date-parts":[[2020,2,5]],"date-time":"2020-02-05T00:00:00Z","timestamp":1580860800000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2020,2,5]],"date-time":"2020-02-05T00:00:00Z","timestamp":1580860800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/501100007601","name":"Horizon 2020","doi-asserted-by":"publisher","award":["688007"],"award-info":[{"award-number":["688007"]}],"id":[{"id":"10.13039\/501100007601","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Neural Comput &amp; Applic"],"published-print":{"date-parts":[[2020,12]]},"abstract":"<jats:title>Abstract<\/jats:title><jats:p>Convolutional neural networks (CNNs) lack robustness to test image corruptions that are not seen during training. In this paper, we propose a new layer for CNNs that increases their robustness to several types of corruptions of the input images. We call it a \u2018push\u2013pull\u2019 layer and compute its response as the combination of two half-wave rectified convolutions, with kernels of different size and opposite polarity. Its implementation is based on a biologically motivated model of certain neurons in the visual system that exhibit response suppression, known as push\u2013pull inhibition. We validate our method by replacing the first convolutional layer of the LeNet, ResNet and DenseNet architectures with our push\u2013pull layer. We train the networks on original training images from the MNIST and CIFAR data sets and test them on images with several corruptions, of different types and severities, that are unseen by the training process. We experiment with various configurations of the ResNet and DenseNet models on a benchmark test set with typical image corruptions constructed on the CIFAR test images. We demonstrate that our push\u2013pull layer contributes to a considerable improvement in robustness of classification of corrupted images, while maintaining state-of-the-art performance on the original image classification task. We released the code and trained models at the url <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" ext-link-type=\"uri\" xlink:href=\"http:\/\/github.com\/nicstrisc\/Push-Pull-CNN-layer\">http:\/\/github.com\/nicstrisc\/Push-Pull-CNN-layer<\/jats:ext-link>.<\/jats:p>","DOI":"10.1007\/s00521-020-04751-8","type":"journal-article","created":{"date-parts":[[2020,3,6]],"date-time":"2020-03-06T13:34:43Z","timestamp":1583501683000},"page":"17957-17971","update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":29,"title":["Enhanced robustness of convolutional networks with a push\u2013pull inhibition layer"],"prefix":"10.1007","volume":"32","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-7478-3509","authenticated-orcid":false,"given":"Nicola","family":"Strisciuglio","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Manuel","family":"Lopez-Antequera","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nicolai","family":"Petkov","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2020,2,5]]},"reference":[{"key":"4751_CR1","doi-asserted-by":"publisher","first-page":"14410","DOI":"10.1109\/access.2018.2807385","volume":"6","author":"N Akhtar","year":"2018","unstructured":"Akhtar N, Mian A (2018) Threat of adversarial attacks on deep learning in computer vision: a survey. IEEE Access 6:14410\u201314430. https:\/\/doi.org\/10.1109\/access.2018.2807385","journal-title":"IEEE Access"},{"issue":"3","key":"4751_CR2","doi-asserted-by":"publisher","first-page":"340","DOI":"10.1016\/j.conb.2010.02.012","volume":"20","author":"HJ Alitto","year":"2010","unstructured":"Alitto HJ, Dan Y (2010) Function of inhibition in visual cortical processing. Curr Opin Neurobiol 20(3):340\u2013346. https:\/\/doi.org\/10.1016\/j.conb.2010.02.012","journal-title":"Curr Opin Neurobiol"},{"issue":"3","key":"4751_CR3","doi-asserted-by":"publisher","first-page":"177","DOI":"10.1007\/s00422-012-0486-6","volume":"106","author":"G Azzopardi","year":"2012","unstructured":"Azzopardi G, Petkov N (2012) A CORF computational model of a simple cell that relies on LGN input outperforms the Gabor function model. Biol Cybern 106(3):177\u2013189. https:\/\/doi.org\/10.1007\/s00422-012-0486-6","journal-title":"Biol Cybern"},{"issue":"7","key":"4751_CR4","doi-asserted-by":"publisher","first-page":"e98424","DOI":"10.1371\/journal.pone.0098424","volume":"9","author":"G Azzopardi","year":"2014","unstructured":"Azzopardi G, Rodr\u00edguez-S\u00e1nchez A, Piater J, Petkov N (2014) A push\u2013pull corf model of a simple cell with antiphase inhibition improves snr and contour detection. PLoS ONE 9(7):e98424. https:\/\/doi.org\/10.1371\/journal.pone.0098424","journal-title":"PLoS ONE"},{"issue":"1","key":"4751_CR5","doi-asserted-by":"publisher","first-page":"46","DOI":"10.1016\/j.media.2014.08.002","volume":"19","author":"G Azzopardi","year":"2015","unstructured":"Azzopardi G, Strisciuglio N, Vento M, Petkov N (2015) Trainable cosfire filters for vessel delineation with application to retinal images. Med Image Anal 19(1):46\u201357. https:\/\/doi.org\/10.1016\/j.media.2014.08.002","journal-title":"Med Image Anal"},{"key":"4751_CR6","unstructured":"Badrinarayanan V, Kendall A, Cipolla R (2015) Segnet: a deep convolutional encoder-decoder architecture for image segmentation. CoRR arXiv:1511.00561"},{"issue":"6060","key":"4751_CR7","doi-asserted-by":"publisher","first-page":"362","DOI":"10.1038\/320362a0","volume":"320","author":"J Bolz","year":"1986","unstructured":"Bolz J, Gilbert CD (1986) Generation of end-inhibition in the visual cortex via interlaminar connections. Nature 320(6060):362\u2013365","journal-title":"Nature"},{"key":"4751_CR8","unstructured":"Carlini N, Wagner DA (2016) Towards evaluating the robustness of neural networks. CoRR arXiv:abs\/1608.04644"},{"key":"4751_CR9","unstructured":"Cohen TS, Welling M (2016) Steerable cnns. CoRR arXiv:abs\/1612.08498"},{"key":"4751_CR10","doi-asserted-by":"publisher","unstructured":"Dodge S, Karam L (2017) A study and comparison of human and deep learning recognition performance under visual distortions. In: 2017 26th international conference on computer communication and networks (ICCCN), pp 1\u20137. https:\/\/doi.org\/10.1109\/ICCCN.2017.8038465","DOI":"10.1109\/ICCCN.2017.8038465"},{"issue":"4","key":"4751_CR11","doi-asserted-by":"publisher","first-page":"759","DOI":"10.1016\/S0896-6273(02)00819-X","volume":"35","author":"TC Freeman","year":"2002","unstructured":"Freeman TC, Durand S, Kiper DC, Carandini M (2002) Suppression without inhibition in visual cortex. Neuron 35(4):759\u2013771. https:\/\/doi.org\/10.1016\/S0896-6273(02)00819-X","journal-title":"Neuron"},{"issue":"4","key":"4751_CR12","doi-asserted-by":"publisher","first-page":"193","DOI":"10.1007\/BF00344251","volume":"36","author":"K Fukushima","year":"1980","unstructured":"Fukushima K (1980) Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol Cybern 36(4):193\u2013202. https:\/\/doi.org\/10.1007\/BF00344251","journal-title":"Biol Cybern"},{"key":"4751_CR13","unstructured":"Geirhos R, Temme CRM, Rauber J, Sch\u00fctt HH, Bethge M, Wichmann FA (2018) Generalisation in humans and deep neural networks. In: Advances in neural information processing systems 31 (NeurIPS2018), pp 7538\u20137550"},{"key":"4751_CR14","unstructured":"Goodfellow I, Shlens J, Szegedy C (2015) Explaining and harnessing adversarial examples. In: International conference on learning representations. arXiv:1412.6572"},{"issue":"8","key":"4751_CR15","doi-asserted-by":"publisher","first-page":"609","DOI":"10.1016\/j.imavis.2003.12.004","volume":"22","author":"C Grigorescu","year":"2004","unstructured":"Grigorescu C, Petkov N, Westenberg M (2004) Contour and boundary detection improved by surround suppression of texture edges. Image Vis Comput 22(8):609\u2013622. https:\/\/doi.org\/10.1016\/j.imavis.2003.12.004","journal-title":"Image Vis Comput"},{"key":"4751_CR16","unstructured":"He K, Zhang X, Ren S, Sun J (2015) Deep residual learning for image recognition. CoRR arXiv:abs\/1512.03385"},{"key":"4751_CR17","unstructured":"Hendrycks D, Dietterich T (2019) Benchmarking neural network robustness to common corruptions and perturbations. In: Proceedings of the international conference on learning representations"},{"key":"4751_CR18","doi-asserted-by":"crossref","unstructured":"Huang G, Liu Z, van der Maaten L, Weinberger KQ (2017) Densely connected convolutional networks. In: 2017 IEEE conference on computer vision and pattern recognition (CVPR), pp 2261\u20132269","DOI":"10.1109\/CVPR.2017.243"},{"issue":"1","key":"4751_CR19","doi-asserted-by":"publisher","first-page":"106","DOI":"10.1113\/jphysiol.1962.sp006837","volume":"160","author":"D Hubel","year":"1962","unstructured":"Hubel D, Wiesel T (1962) Receptive fields, binocular interaction and functional architecture in the cat\u2019s visual cortex. J Physiol-Lond 160(1):106\u2013154","journal-title":"J Physiol-Lond"},{"key":"4751_CR20","doi-asserted-by":"crossref","unstructured":"Hui TW, Tang X, Loy CC (2018) Liteflownet: a lightweight convolutional neural network for optical flow estimation. In: Proceedings of IEEE conference on computer vision and pattern recognition (CVPR), pp 8981\u20138989","DOI":"10.1109\/CVPR.2018.00936"},{"key":"4751_CR21","doi-asserted-by":"publisher","first-page":"37","DOI":"10.3389\/fncir.2016.00037","volume":"10","author":"J Kremkow","year":"2016","unstructured":"Kremkow J, Perrinet LU, Monier C, Alonso JM, Aertsen A, Fr\u00e9gnac Y, Masson GS (2016) Push\u2013pull receptive field organization and synaptic depression: mechanisms for reliably encoding naturalistic stimuli in v1. Front Neural Circuits 10:37. https:\/\/doi.org\/10.3389\/fncir.2016.00037","journal-title":"Front Neural Circuits"},{"key":"4751_CR22","first-page":"1097","volume-title":"Advances in neural information processing systems 25","author":"A Krizhevsky","year":"2012","unstructured":"Krizhevsky A, Sutskever I (2012) Imagenet classification with deep convolutional neural networks. In: Pereira F, Burges CJC, Bottou L, Weinberger KQ (eds) Advances in neural information processing systems 25. Curran Associates, Inc., New York, pp 1097\u20131105"},{"key":"4751_CR23","unstructured":"Kurakin A, Goodfellow IJ, Bengio S (2016) Adversarial examples in the physical world. CoRR arXiv:abs\/1607.02533. URL http:\/\/dblp.uni-trier.de\/db\/journals\/corr\/corr1607.html#KurakinGB16"},{"key":"4751_CR24","volume-title":"Feature grouping","author":"Y LeCun","year":"1999","unstructured":"LeCun Y, Haffner P, Bottou L, Bengio Y (1999) Object recognition with gradient-based learning. In: Forsyth D (ed) Feature grouping. Springer, Berlin"},{"key":"4751_CR25","unstructured":"Lee CY, Xie S, Gallagher P, Zhang Z, Tu Z (2015) Deeply-supervised nets. In: Proceedings of the eighteenth international conference on artificial intelligence and statistics, proceedings of machine learning research, vol 38, pp 562\u2013570. PMLR"},{"issue":"46","key":"4751_CR26","doi-asserted-by":"publisher","first-page":"16466","DOI":"10.1523\/JNEUROSCI.3221-12.2012","volume":"32","author":"Y Li","year":"2012","unstructured":"Li Y, Ma WP, Li LY, Ibrahim LA, Wang SZ, Tao HW (2012) Broadening of inhibitory tuning underlies contrast-dependent sharpening of orientation selectivity in mouse visual cortex. J Neurosci 32(46):16466\u201316477. https:\/\/doi.org\/10.1523\/JNEUROSCI.3221-12.2012","journal-title":"J Neurosci"},{"issue":"3","key":"4751_CR27","doi-asserted-by":"publisher","first-page":"542","DOI":"10.1016\/j.neuron.2011.06.017","volume":"71","author":"BH Liu","year":"2011","unstructured":"Liu BH, tang Li YT, Ma WP, Pan CJ, Zhang LI, Tao HW (2011) Broad inhibition sharpens orientation selectivity by expanding input dynamic range in mouse simple cells. Neuron 71(3):542\u2013554. https:\/\/doi.org\/10.1016\/j.neuron.2011.06.017","journal-title":"Neuron"},{"key":"4751_CR28","unstructured":"Lu J, Sibai H, Fabry E, Forsyth DA (2017) Standard detectors aren\u2019t (currently) fooled by physical adversarial stop signs. CoRR arXiv:abs\/1710.03337"},{"key":"4751_CR29","unstructured":"Madry A, Makelov A, Schmidt L, Tsipras D, Vladu A (2018) Towards deep learning models resistant to adversarial attacks. CoRR arXiv:abs\/1706.06083"},{"issue":"11","key":"4751_CR30","doi-asserted-by":"publisher","first-page":"1297","DOI":"10.1364\/JOSA.70.001297","volume":"70","author":"S Mar\u0109elja","year":"1980","unstructured":"Mar\u0109elja S (1980) Mathematical description of the responses of simple cortical cells$$\\ast $$. J Opt Soc Am 70(11):1297\u20131300. https:\/\/doi.org\/10.1364\/JOSA.70.001297","journal-title":"J Opt Soc Am"},{"key":"4751_CR31","unstructured":"Metzen JH, Genewein T, Fischer V, Bischoff B (2017) On detecting adversarial perturbations. In: Proceedings of 5th international conference on learning representations (ICLR). arXiv:1702.04267"},{"key":"4751_CR32","doi-asserted-by":"crossref","unstructured":"Moosavi-Dezfooli S, Fawzi A, Frossard P (2016) Deepfool: a simple and accurate method to fool deep neural networks. In: CVPR. IEEE Computer Society, pp 2574\u20132582","DOI":"10.1109\/CVPR.2016.282"},{"key":"4751_CR33","doi-asserted-by":"crossref","unstructured":"Moosavi-Dezfooli S, Fawzi A, Fawzi O, Frossard P (2017) Universal adversarial perturbations. In: CVPR. IEEE Computer Society, pp 86\u201394","DOI":"10.1109\/CVPR.2017.17"},{"key":"4751_CR34","unstructured":"Papernot N, McDaniel PD, Wu X, Jha S, Swami A (2015) Distillation as a defense to adversarial perturbations against deep neural networks. CoRR arXiv:abs\/1511.04508"},{"key":"4751_CR35","unstructured":"Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recognition. CoRR arXiv:abs\/1409.1556"},{"key":"4751_CR36","unstructured":"Song X, Zhao X, Hu H, Fang L (2018) Edgestereo: a context integrated residual pyramid network for stereo matching. CoRR arXiv:abs\/1803.05196"},{"key":"4751_CR37","doi-asserted-by":"publisher","unstructured":"Strisciuglio N, Petkov N (2017) Delineation of line patterns in images using b-cosfire filters. In: IWOBI, pp 1\u20136. https:\/\/doi.org\/10.1109\/IWOBI.2017.7985538","DOI":"10.1109\/IWOBI.2017.7985538"},{"key":"4751_CR38","doi-asserted-by":"crossref","unstructured":"Strisciuglio N, Azzopardi G, Petkov N (2019) Brain-inspired robust delineation operator. In: Computer Vision\u2014ECCV 2018 Workshops, pp 555\u2013565","DOI":"10.1007\/978-3-030-11015-4_41"},{"issue":"12","key":"4751_CR39","doi-asserted-by":"publisher","first-page":"5852","DOI":"10.1109\/TIP.2019.2922096","volume":"28","author":"N Strisciuglio","year":"2019","unstructured":"Strisciuglio N, Azzopardi G, Petkov N (2019) Robust inhibition-augmented operator for delineation of curvilinear structures. IEEE Trans Image Process 28(12):5852\u20135866. https:\/\/doi.org\/10.1109\/TIP.2019.2922096","journal-title":"IEEE Trans Image Process"},{"key":"4751_CR40","unstructured":"Szegedy C, Inc G, Zaremba W, Sutskever I, Inc G, Bruna J, Erhan D, Inc G, Goodfellow I, Fergus R (2014) Intriguing properties of neural networks. In: In ICLR"},{"key":"4751_CR41","doi-asserted-by":"crossref","unstructured":"Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A (2015) Going deeper with convolutions. In: Computer vision and pattern recognition (CVPR). arXiv:1409.4842","DOI":"10.1109\/CVPR.2015.7298594"},{"issue":"3","key":"4751_CR42","doi-asserted-by":"publisher","first-page":"595","DOI":"10.1523\/JNEUROSCI.2099-17.2017","volume":"38","author":"MM Taylor","year":"2018","unstructured":"Taylor MM, Sedigh-Sarvestani M, Vigeland L, Palmer LA, Contreras D (2018) Inhibition in simple cell receptive fields is broad and off-subregion biased. J Neurosci 38(3):595\u2013612. https:\/\/doi.org\/10.1523\/JNEUROSCI.2099-17.2017","journal-title":"J Neurosci"},{"key":"4751_CR43","doi-asserted-by":"crossref","unstructured":"Temel D, Kwon G, Prabhuhankar M, AlRegib G (2017) CURE-TSR: challenging unreal and real environments for traffic sign recognition. In: Advances in neural information processing systems (NIPS) machine learning for intelligent transportations systems workshop","DOI":"10.1109\/ICMLA.2018.00028"},{"key":"4751_CR44","doi-asserted-by":"crossref","unstructured":"Temel D, Lee J, AlRegib G (2018) CURE-OR: challenging unreal and real environments for object recognition. CoRR arXiv:abs\/1810.08293","DOI":"10.1109\/ICMLA.2018.00028"},{"key":"4751_CR45","unstructured":"Vasiljevic I, Chakrabarti A, Shakhnarovich G (2016) Examining the impact of blur on recognition by convolutional networks. CoRR arXiv:abs\/1611.05760"},{"key":"4751_CR46","unstructured":"Weiler M, Hamprecht FA, Storath M (2017) Learning steerable filters for rotation equivariant cnns. CoRR arXiv:abs\/1711.07289"},{"key":"4751_CR47","unstructured":"Worrall DE, Garbin SJ, Turmukhambetov D, Brostow GJ (2016) Harmonic networks: deep translation and rotation equivariance. CoRR arXiv:abs\/1612.04642"},{"key":"4751_CR48","doi-asserted-by":"crossref","unstructured":"Zagoruyko S, Komodakis N (2016) Wide residual networks. In: BMVC","DOI":"10.5244\/C.30.87"},{"key":"4751_CR49","doi-asserted-by":"publisher","unstructured":"Zheng S, Song Y, Leung T, Goodfellow I (2016) Improving the robustness of deep neural networks via stability training. In: 2016 IEEE conference on computer vision and pattern recognition (CVPR), pp 4480\u20134488. https:\/\/doi.org\/10.1109\/CVPR.2016.485","DOI":"10.1109\/CVPR.2016.485"},{"key":"4751_CR50","doi-asserted-by":"publisher","unstructured":"Zoumpourlis G, Doumanoglou A, Vretos N, Daras P (2017) Non-linear convolution filters for cnn-based learning. In: IEEE international conference on computer vision, ICCV 2017, Venice, Italy, October 22\u201329, 2017, pp 4771\u20134779. https:\/\/doi.org\/10.1109\/ICCV.2017.510","DOI":"10.1109\/ICCV.2017.510"}],"container-title":["Neural Computing and Applications"],"original-title":[],"language":"en","link":[{"URL":"http:\/\/link.springer.com\/content\/pdf\/10.1007\/s00521-020-04751-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"http:\/\/link.springer.com\/article\/10.1007\/s00521-020-04751-8\/fulltext.html","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"http:\/\/link.springer.com\/content\/pdf\/10.1007\/s00521-020-04751-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2021,2,4]],"date-time":"2021-02-04T00:51:09Z","timestamp":1612399869000},"score":1,"resource":{"primary":{"URL":"http:\/\/link.springer.com\/10.1007\/s00521-020-04751-8"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2020,2,5]]},"references-count":50,"journal-issue":{"issue":"24","published-print":{"date-parts":[[2020,12]]}},"alternative-id":["4751"],"URL":"https:\/\/doi.org\/10.1007\/s00521-020-04751-8","relation":{},"ISSN":["0941-0643","1433-3058"],"issn-type":[{"value":"0941-0643","type":"print"},{"value":"1433-3058","type":"electronic"}],"subject":[],"published":{"date-parts":[[2020,2,5]]},"assertion":[{"value":"25 April 2019","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"23 January 2020","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"5 February 2020","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Compliance with ethical standards"}},{"value":"The authors declare that they have no further conflict of interest.","order":2,"name":"Ethics","group":{"name":"EthicsHeading","label":"Conflict of interest"}}]}}