{"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/noisy-natural-gradient-as-variational","title":"Noisy Natural Gradient as Variational Inference","arxiv_id":"1712.02390","date":"2017-12-06","proceeding":"ICML 2018 7","authors":["Guodong Zhang","Shengyang Sun","David Duvenaud","Roger Grosse"],"abstract":"Variational Bayesian neural nets combine the flexibility of deep learning\nwith Bayesian uncertainty estimation. Unfortunately, there is a tradeoff\nbetween cheap but simple variational families (e.g.~fully factorized) or\nexpensive and complicated inference procedures. We show that natural gradient\nascent with adaptive weight noise implicitly fits a variational posterior to\nmaximize the evidence lower bound (ELBO). This insight allows us to train\nfull-covariance, fully factorized, or matrix-variate Gaussian variational\nposteriors using noisy versions of natural gradient, Adam, and K-FAC,\nrespectively, making it possible to scale up to modern-size ConvNets. On\nstandard regression benchmarks, our noisy K-FAC algorithm makes better\npredictions and matches Hamiltonian Monte Carlo's predictive variances better\nthan existing methods. Its improved uncertainty estimates lead to more\nefficient exploration in active learning, and intrinsic motivation for\nreinforcement learning.","url_abs":"http://arxiv.org/abs/1712.02390v2","url_pdf":"http://arxiv.org/pdf/1712.02390v2.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":"noisy-natural-gradient-as-variational","repo_url":"https://github.com/gd-zhang/noisy-K-FAC","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"noisy-natural-gradient-as-variational","repo_url":"https://github.com/minfanzhang/noisy-K-FAC_added_flipout","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"active-learning","task_name":"Active Learning"},{"task_slug":"efficient-exploration","task_name":"Efficient Exploration"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"variational-inference","task_name":"Variational Inference"}],"methods":[{"method_slug":"adam","method_name":"Adam"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1712.02390","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}