{"id":6567,"date":"2026-03-19T15:00:15","date_gmt":"2026-03-19T06:00:15","guid":{"rendered":"http:\/\/www-dsc.naist.jp\/dsc_naist\/?p=6567"},"modified":"2026-03-27T19:01:23","modified_gmt":"2026-03-27T10:01:23","slug":"dsc-internal-talk-in-march2026","status":"publish","type":"post","link":"http:\/\/www-dsc.naist.jp\/dsc_naist\/en\/dsc-internal-talk-in-march2026\/","title":{"rendered":"DSC internal talk in March"},"content":{"rendered":"<p>Assoc. Prof. Hidetaka Kamigaito gave a lecture in March.<\/p>\n<p>The details are as follows.<\/p>\n<p><span style=\"font-family: georgia, palatino, serif; font-size: 12pt;\">==================== <\/span><\/p>\n<p>Assoc. Prof. Hidetaka Kamigaito (Natural Language Processing Laboratory)<\/p>\n<p>TITLE:<br \/>\nRecent Advances in Minimum Bayes Risk Decoding<\/p>\n<p>ABSTRACT\uff1a<br \/>\nAs represented by large language models (LLMs), generative AI models are fundamental tools in various fields. However, their generation results are sometimes unstable. To mitigate the issue, we can use Minimum Bayes Risk (MBR) decoding, which considers the consensus of generated candidates. In this presentation, I\u2019ll introduce recent advances in MBR decoding, focusing primarily on its use in LLMs.<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<div class=\"block-list-appender wp-block\" tabindex=\"-1\" contenteditable=\"false\" data-block=\"true\">\n<div class=\"block-editor-default-block-appender\" data-root-client-id=\"\">\n<p class=\"block-editor-default-block-appender__content\" tabindex=\"0\" role=\"button\" aria-label=\"\u30c7\u30d5\u30a9\u30eb\u30c8\u30d6\u30ed\u30c3\u30af\u3092\u8ffd\u52a0\">\u00a0<\/p>\n<\/div>\n<\/div>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Assoc. Prof. Hidetaka Kamigaito gave a lecture in March. The details are as follows. ==================== Assoc. Prof. Hidetaka Kamigaito (Natural Language Processing Laboratory) TITLE: Recent Advances in Minimum Bayes Risk Decoding ABSTRACT\uff1a As represented by large language models (LLMs), generative AI models are fundamental tools in various fields. However, their generation results are sometimes unstable. [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_locale":"en_US","_original_post":"http:\/\/www-dsc.naist.jp\/dsc_naist\/?p=6564","_links_to":"","_links_to_target":""},"categories":[3],"tags":[],"event_taxonomy":[15],"acf":[],"_links":{"self":[{"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/posts\/6567"}],"collection":[{"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/comments?post=6567"}],"version-history":[{"count":2,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/posts\/6567\/revisions"}],"predecessor-version":[{"id":6569,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/posts\/6567\/revisions\/6569"}],"wp:attachment":[{"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/media?parent=6567"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/categories?post=6567"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/tags?post=6567"},{"taxonomy":"event_taxonomy","embeddable":true,"href":"http:\/\/www-dsc.naist.jp\/dsc_naist\/wp-json\/wp\/v2\/event_taxonomy?post=6567"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}