{"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/boltzmann-exploration-expectationmaximisation","title":"Boltzmann Exploration Expectation–Maximisation","arxiv_id":null,"date":"2019-12-18","proceeding":"arXiv 2019 12","authors":["Mathias Edman","Neil Dhir"],"abstract":"We present a general method for fitting finite mixture models (FMM). Learning\r\nin a mixture model consists of finding the most likely cluster assignment for each\r\ndata-point, as well as finding the parameters of the clusters themselves. In many\r\nmixture models, this is difficult with current learning methods, where the most common approach is to employ monotone learning algorithms e.g. the conventional expectation-maximisation algorithm. While effective, the success of any\r\nmonotone algorithm is crucially dependant on good parameter initialisation, where\r\na common choice is K-means initialisation, commonly employed for Gaussian\r\nmixture models.\r\nFor other types of mixture models, the path to good initialisation parameters is often\r\nunclear and may require a problem-specific solution. To this end, we propose a\r\ngeneral heuristic learning algorithm that utilises Boltzmann exploration to assign\r\neach observation to a specific base distribution within the mixture model, which we\r\ncall Boltzmann exploration expectation-maximisation (BEEM). With BEEM, hard\r\nassignments allow straight forward parameter learning for each base distribution\r\nby conditioning only on its assigned observations. Consequently, it can be applied\r\nto mixtures of any base distribution where single component parameter learning is\r\ntractable. The stochastic learning procedure is able to escape local optima and is\r\nthus insensitive to parameter initialisation. We show competitive performance on a\r\nnumber of synthetic benchmark cases as well as on real-world datasets.","url_abs":"https://arxiv.org/abs/1912.08869","url_pdf":"https://arxiv.org/pdf/1912.08869.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":"boltzmann-exploration-expectationmaximisation","repo_url":"https://github.com/kaminAI/beem","is_official":0,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"iris-segmentation","task_name":"Iris Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}