{"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/variational-inference-for-monte-carlo","title":"Variational inference for Monte Carlo objectives","arxiv_id":"1602.06725","date":"2016-02-22","proceeding":null,"authors":["Andriy Mnih","Danilo J. Rezende"],"abstract":"Recent progress in deep latent variable models has largely been driven by the\ndevelopment of flexible and scalable variational inference methods. Variational\ntraining of this type involves maximizing a lower bound on the log-likelihood,\nusing samples from the variational posterior to compute the required gradients.\nRecently, Burda et al. (2016) have derived a tighter lower bound using a\nmulti-sample importance sampling estimate of the likelihood and showed that\noptimizing it yields models that use more of their capacity and achieve higher\nlikelihoods. This development showed the importance of such multi-sample\nobjectives and explained the success of several related approaches.\n  We extend the multi-sample approach to discrete latent variables and analyze\nthe difficulty encountered when estimating the gradients involved. We then\ndevelop the first unbiased gradient estimator designed for importance-sampled\nobjectives and evaluate it at training generative and structured output\nprediction models. The resulting estimator, which is based on low-variance\nper-sample learning signals, is both simpler and more effective than the NVIL\nestimator proposed for the single-sample variational objective, and is\ncompetitive with the currently used biased estimators.","url_abs":"http://arxiv.org/abs/1602.06725v2","url_pdf":"http://arxiv.org/pdf/1602.06725v2.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":"variational-inference-for-monte-carlo","repo_url":"https://github.com/artemZholus/variational_memory_addressing","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"variational-inference","task_name":"Variational Inference"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1602.06725","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}