{"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/eigenvalue-corrected-noisy-natural-gradient","title":"Eigenvalue Corrected Noisy Natural Gradient","arxiv_id":"1811.12565","date":"2018-11-30","proceeding":null,"authors":["Juhan Bae","Guodong Zhang","Roger Grosse"],"abstract":"Variational Bayesian neural networks combine the flexibility of deep learning\nwith Bayesian uncertainty estimation. However, inference procedures for\nflexible variational posteriors are computationally expensive. A recently\nproposed method, noisy natural gradient, is a surprisingly simple method to fit\nexpressive posteriors by adding weight noise to regular natural gradient\nupdates. Noisy K-FAC is an instance of noisy natural gradient that fits a\nmatrix-variate Gaussian posterior with minor changes to ordinary K-FAC.\nNevertheless, a matrix-variate Gaussian posterior does not capture an accurate\ndiagonal variance. In this work, we extend on noisy K-FAC to obtain a more\nflexible posterior distribution called eigenvalue corrected matrix-variate\nGaussian. The proposed method computes the full diagonal re-scaling factor in\nKronecker-factored eigenbasis. Empirically, our approach consistently\noutperforms existing algorithms (e.g., noisy K-FAC) on regression and\nclassification tasks.","url_abs":"http://arxiv.org/abs/1811.12565v1","url_pdf":"http://arxiv.org/pdf/1811.12565v1.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":"eigenvalue-corrected-noisy-natural-gradient","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":"eigenvalue-corrected-noisy-natural-gradient","repo_url":"https://github.com/pomonam/NoisyNaturalGradient","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"eigenvalue-corrected-noisy-natural-gradient","repo_url":"https://github.com/shwang/NNG","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.12565","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}