{"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/learning-to-discover-sparse-graphical-models","title":"Learning to Discover Sparse Graphical Models","arxiv_id":"1605.06359","date":"2016-05-20","proceeding":"ICML 2017 8","authors":["Eugene Belilovsky","Kyle Kastner","Gaël Varoquaux","Matthew Blaschko"],"abstract":"We consider structure discovery of undirected graphical models from\nobservational data. Inferring likely structures from few examples is a complex\ntask often requiring the formulation of priors and sophisticated inference\nprocedures. Popular methods rely on estimating a penalized maximum likelihood\nof the precision matrix. However, in these approaches structure recovery is an\nindirect consequence of the data-fit term, the penalty can be difficult to\nadapt for domain-specific knowledge, and the inference is computationally\ndemanding. By contrast, it may be easier to generate training samples of data\nthat arise from graphs with the desired structure properties. We propose here\nto leverage this latter source of information as training data to learn a\nfunction, parametrized by a neural network that maps empirical covariance\nmatrices to estimated graph structures. Learning this function brings two\nbenefits: it implicitly models the desired structure or sparsity properties to\nform suitable priors, and it can be tailored to the specific problem of edge\nstructure discovery, rather than maximizing data likelihood. Applying this\nframework, we find our learnable graph-discovery method trained on synthetic\ndata generalizes well: identifying relevant edges in both synthetic and real\ndata, completely unknown at training time. We find that on genetics, brain\nimaging, and simulation data we obtain performance generally superior to\nanalytical methods.","url_abs":"http://arxiv.org/abs/1605.06359v3","url_pdf":"http://arxiv.org/pdf/1605.06359v3.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":"learning-to-discover-sparse-graphical-models","repo_url":"https://github.com/eugenium/LearnGraphDiscovery","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1605.06359","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}