{"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/dice-the-infinitely-differentiable-monte","title":"DiCE: The Infinitely Differentiable Monte-Carlo Estimator","arxiv_id":"1802.05098","date":"2018-02-14","proceeding":null,"authors":["Jakob Foerster","Gregory Farquhar","Maruan Al-Shedivat","Tim Rocktäschel","Eric P. Xing","Shimon Whiteson"],"abstract":"The score function estimator is widely used for estimating gradients of\nstochastic objectives in stochastic computation graphs (SCG), eg, in\nreinforcement learning and meta-learning. While deriving the first-order\ngradient estimators by differentiating a surrogate loss (SL) objective is\ncomputationally and conceptually simple, using the same approach for\nhigher-order derivatives is more challenging. Firstly, analytically deriving\nand implementing such estimators is laborious and not compliant with automatic\ndifferentiation. Secondly, repeatedly applying SL to construct new objectives\nfor each order derivative involves increasingly cumbersome graph manipulations.\nLastly, to match the first-order gradient under differentiation, SL treats part\nof the cost as a fixed sample, which we show leads to missing and wrong terms\nfor estimators of higher-order derivatives. To address all these shortcomings\nin a unified way, we introduce DiCE, which provides a single objective that can\nbe differentiated repeatedly, generating correct estimators of derivatives of\nany order in SCGs. Unlike SL, DiCE relies on automatic differentiation for\nperforming the requisite graph manipulations. We verify the correctness of DiCE\nboth through a proof and numerical evaluation of the DiCE derivative estimates.\nWe also use DiCE to propose and evaluate a novel approach for multi-agent\nlearning. Our code is available at https://www.github.com/alshedivat/lola.","url_abs":"http://arxiv.org/abs/1802.05098v3","url_pdf":"http://arxiv.org/pdf/1802.05098v3.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":"dice-the-infinitely-differentiable-monte","repo_url":"https://github.com/alshedivat/lola","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"dice-the-infinitely-differentiable-monte","repo_url":"https://github.com/alexis-jacq/LOLA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"dice-the-infinitely-differentiable-monte","repo_url":"https://github.com/alexis-jacq/LOLA_DICE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"dice-the-infinitely-differentiable-monte","repo_url":"https://github.com/longtermrisk/marltoolbox","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"dice-the-infinitely-differentiable-monte","repo_url":"https://github.com/yannbouteiller/gym-airsimdroneracinglab","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"meta-learning","task_name":"Meta-Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1802.05098","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}