{"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/fathom-reference-workloads-for-modern-deep","title":"Fathom: Reference Workloads for Modern Deep Learning Methods","arxiv_id":"1608.06581","date":"2016-08-23","proceeding":null,"authors":["Robert Adolf","Saketh Rama","Brandon Reagen","Gu-Yeon Wei","David Brooks"],"abstract":"Deep learning has been popularized by its recent successes on challenging\nartificial intelligence problems. One of the reasons for its dominance is also\nan ongoing challenge: the need for immense amounts of computational power.\nHardware architects have responded by proposing a wide array of promising\nideas, but to date, the majority of the work has focused on specific algorithms\nin somewhat narrow application domains. While their specificity does not\ndiminish these approaches, there is a clear need for more flexible solutions.\nWe believe the first step is to examine the characteristics of cutting edge\nmodels from across the deep learning community.\n  Consequently, we have assembled Fathom: a collection of eight archetypal deep\nlearning workloads for study. Each of these models comes from a seminal work in\nthe deep learning community, ranging from the familiar deep convolutional\nneural network of Krizhevsky et al., to the more exotic memory networks from\nFacebook's AI research group. Fathom has been released online, and this paper\nfocuses on understanding the fundamental performance characteristics of each\nmodel. We use a set of application-level modeling tools built around the\nTensorFlow deep learning framework in order to analyze the behavior of the\nFathom workloads. We present a breakdown of where time is spent, the\nsimilarities between the performance profiles of our models, an analysis of\nbehavior in inference and training, and the effects of parallelism on scaling.","url_abs":"http://arxiv.org/abs/1608.06581v1","url_pdf":"http://arxiv.org/pdf/1608.06581v1.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":"fathom-reference-workloads-for-modern-deep","repo_url":"https://github.com/rdadolf/fathom","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"deep-learning","task_name":"Deep Learning"},{"task_slug":"specificity","task_name":"Specificity"}],"methods":[],"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}