{"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/reparameterization-gradient-for-non","title":"Reparameterization Gradient for Non-differentiable Models","arxiv_id":"1806.00176","date":"2018-06-01","proceeding":"NeurIPS 2018 12","authors":["Wonyeol Lee","Hangyeol Yu","Hongseok Yang"],"abstract":"We present a new algorithm for stochastic variational inference that targets\nat models with non-differentiable densities. One of the key challenges in\nstochastic variational inference is to come up with a low-variance estimator of\nthe gradient of a variational objective. We tackle the challenge by\ngeneralizing the reparameterization trick, one of the most effective techniques\nfor addressing the variance issue for differentiable models, so that the trick\nworks for non-differentiable models as well. Our algorithm splits the space of\nlatent variables into regions where the density of the variables is\ndifferentiable, and their boundaries where the density may fail to be\ndifferentiable. For each differentiable region, the algorithm applies the\nstandard reparameterization trick and estimates the gradient restricted to the\nregion. For each potentially non-differentiable boundary, it uses a form of\nmanifold sampling and computes the direction for variational parameters that,\nif followed, would increase the boundary's contribution to the variational\nobjective. The sum of all the estimates becomes the gradient estimate of our\nalgorithm. Our estimator enjoys the reduced variance of the reparameterization\ngradient while remaining unbiased even for non-differentiable models. The\nexperiments with our preliminary implementation confirm the benefit of reduced\nvariance and unbiasedness.","url_abs":"http://arxiv.org/abs/1806.00176v2","url_pdf":"http://arxiv.org/pdf/1806.00176v2.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":"reparameterization-gradient-for-non","repo_url":"https://github.com/wonyeol/reparam-nondiff","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"variational-inference","task_name":"Variational Inference"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.00176","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}