{"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/active-mini-batch-sampling-using-repulsive","title":"Active Mini-Batch Sampling using Repulsive Point Processes","arxiv_id":"1804.02772","date":"2018-04-08","proceeding":null,"authors":["Cheng Zhang","Cengiz Öztireli","Stephan Mandt","Giampiero Salvi"],"abstract":"The convergence speed of stochastic gradient descent (SGD) can be improved by\nactively selecting mini-batches. We explore sampling schemes where similar data\npoints are less likely to be selected in the same mini-batch. In particular, we\nprove that such repulsive sampling schemes lowers the variance of the gradient\nestimator. This generalizes recent work on using Determinantal Point Processes\n(DPPs) for mini-batch diversification (Zhang et al., 2017) to the broader class\nof repulsive point processes. We first show that the phenomenon of variance\nreduction by diversified sampling generalizes in particular to non-stationary\npoint processes. We then show that other point processes may be computationally\nmuch more efficient than DPPs. In particular, we propose and investigate\nPoisson Disk sampling---frequently encountered in the computer graphics\ncommunity---for this task. We show empirically that our approach improves over\nstandard SGD both in terms of convergence speed as well as final model\nperformance.","url_abs":"http://arxiv.org/abs/1804.02772v2","url_pdf":"http://arxiv.org/pdf/1804.02772v2.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":"active-mini-batch-sampling-using-repulsive","repo_url":"https://github.com/andrew31416/sampling","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"point-processes","task_name":"Point Processes"}],"methods":[{"method_slug":"sgd","method_name":"SGD"},{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.02772","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}