{"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/doubly-reparameterized-gradient-estimators","title":"Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives","arxiv_id":"1810.04152","date":"2018-10-09","proceeding":"ICLR 2019 5","authors":["George Tucker","Dieterich Lawson","Shixiang Gu","Chris J. Maddison"],"abstract":"Deep latent variable models have become a popular model choice due to the\nscalable learning algorithms introduced by (Kingma & Welling, 2013; Rezende et\nal., 2014). These approaches maximize a variational lower bound on the\nintractable log likelihood of the observed data. Burda et al. (2015) introduced\na multi-sample variational bound, IWAE, that is at least as tight as the\nstandard variational lower bound and becomes increasingly tight as the number\nof samples increases. Counterintuitively, the typical inference network\ngradient estimator for the IWAE bound performs poorly as the number of samples\nincreases (Rainforth et al., 2018; Le et al., 2018). Roeder et al. (2017)\npropose an improved gradient estimator, however, are unable to show it is\nunbiased. We show that it is in fact biased and that the bias can be estimated\nefficiently with a second application of the reparameterization trick. The\ndoubly reparameterized gradient (DReG) estimator does not suffer as the number\nof samples increases, resolving the previously raised issues. The same idea can\nbe used to improve many recently introduced training techniques for latent\nvariable models. In particular, we show that this estimator reduces the\nvariance of the IWAE gradient, the reweighted wake-sleep update (RWS)\n(Bornschein & Bengio, 2014), and the jackknife variational inference (JVI)\ngradient (Nowozin, 2018). Finally, we show that this computationally efficient,\nunbiased drop-in gradient estimator translates to improved performance for all\nthree objectives on several modeling tasks.","url_abs":"http://arxiv.org/abs/1810.04152v2","url_pdf":"http://arxiv.org/pdf/1810.04152v2.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":"doubly-reparameterized-gradient-estimators","repo_url":"https://github.com/ElleryL/GradientEstimator","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"doubly-reparameterized-gradient-estimators","repo_url":"https://github.com/adrianjav/heterogeneous_vaes","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"doubly-reparameterized-gradient-estimators","repo_url":"https://github.com/nbip/IWAE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"variational-inference","task_name":"Variational Inference"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1810.04152","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1810.04152"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/adrianjav/heterogeneous_vaes","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/nbip/IWAE","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/ElleryL/GradientEstimator","reach":{"status":"ok"}}],"summary":{"unverified":2},"by_repo_kind":{"listed":{"samples":2,"ran":0,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"37bf889d9d1db89e","entry":"bernoullisample","repo":"nbip/IWAE","repo_kind":"listed","path":"src/utils.py","file_url":"https://github.com/nbip/IWAE/blob/HEAD/src/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"37bf889d9d1db89e"}},{"code_sha256_prefix":"d372626c71f42bf0","entry":"logmeanexp","repo":"nbip/IWAE","repo_kind":"listed","path":"src/utils.py","file_url":"https://github.com/nbip/IWAE/blob/HEAD/src/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"d372626c71f42bf0"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}