{"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/stdp-based-spiking-deep-convolutional-neural","title":"STDP-based spiking deep convolutional neural networks for object recognition","arxiv_id":"1611.01421","date":"2016-11-04","proceeding":null,"authors":["Saeed Reza Kheradpisheh","Mohammad Ganjtabesh","Simon J. Thorpe","Timothée Masquelier"],"abstract":"Previous studies have shown that spike-timing-dependent plasticity (STDP) can\nbe used in spiking neural networks (SNN) to extract visual features of low or\nintermediate complexity in an unsupervised manner. These studies, however, used\nrelatively shallow architectures, and only one layer was trainable. Another\nline of research has demonstrated - using rate-based neural networks trained\nwith back-propagation - that having many layers increases the recognition\nrobustness, an approach known as deep learning. We thus designed a deep SNN,\ncomprising several convolutional (trainable with STDP) and pooling layers. We\nused a temporal coding scheme where the most strongly activated neurons fire\nfirst, and less activated neurons fire later or not at all. The network was\nexposed to natural images. Thanks to STDP, neurons progressively learned\nfeatures corresponding to prototypical patterns that were both salient and\nfrequent. Only a few tens of examples per category were required and no label\nwas needed. After learning, the complexity of the extracted features increased\nalong the hierarchy, from edge detectors in the first layer to object\nprototypes in the last layer. Coding was very sparse, with only a few thousands\nspikes per image, and in some cases the object category could be reasonably\nwell inferred from the activity of a single higher-order neuron. More\ngenerally, the activity of a few hundreds of such neurons contained robust\ncategory information, as demonstrated using a classifier on Caltech 101,\nETH-80, and MNIST databases. We also demonstrate the superiority of STDP over\nother unsupervised techniques such as random crops (HMAX) or auto-encoders.\nTaken together, our results suggest that the combination of STDP with latency\ncoding may be a key to understanding the way that the primate visual system\nlearns, its remarkable processing speed and its low energy consumption.","url_abs":"http://arxiv.org/abs/1611.01421v3","url_pdf":"http://arxiv.org/pdf/1611.01421v3.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":"stdp-based-spiking-deep-convolutional-neural","repo_url":"https://github.com/ggoupy/SpikingConvNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"object-recognition","task_name":"Object Recognition"}],"methods":[{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}