{"id":8342,"date":"2026-03-23T05:12:00","date_gmt":"2026-03-22T20:12:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8342"},"modified":"2026-03-21T14:15:19","modified_gmt":"2026-03-21T05:15:19","slug":"mamba-3-improved-sequence-modeling-using-state-space-principles","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8342","title":{"rendered":"Mamba-3: Improved Sequence Modeling using State Space Principles"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Mamba-3: Improved Sequence Modeling using State Space Principles\u00a0<\/strong>[74.4]<br>\u7dda\u5f62\u30e2\u30c7\u30eb\u306e\u72b6\u614b\u7a7a\u9593\u30e2\u30c7\u30eb(SSM)\u306e\u8996\u70b9\u306b\u89e6\u767a\u3055\u308c\u305f3\u3064\u306e\u4e2d\u6838\u7684\u65b9\u6cd5\u8ad6\u7684\u6539\u5584\u3092\u7d39\u4ecb\u3059\u308b\u3002 \u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306e\u6539\u826f\u3068\u3068\u3082\u306b\u3001Mamba-3\u30e2\u30c7\u30eb\u306f\u3001\u691c\u7d22\u3001\u72b6\u614b\u8ffd\u8de1\u3001\u4e0b\u6d41\u8a00\u8a9e\u30e2\u30c7\u30ea\u30f3\u30b0\u30bf\u30b9\u30af\u9593\u3067\u5927\u304d\u306a\u9032\u6b69\u3092\u9042\u3052\u307e\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2603.15569v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2603.15569v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 16 Mar 2026 17:30:08 GMT)<\/li>\n\n\n\n<li>\u300cWe combine: (1) a more expressive recurrence derived from SSM discretization, (2) a complex-valued state update rule that enables richer state tracking, and (3) a multi-input, multi-output (MIMO) formulation for better model performance without increasing decode latency.\u300d\u3001\u300cAt 1.5B scale, Mamba-3 (MIMO) improves downstream language modeling accuracy by +2.2 over Transformers, +1.9 points over Mamba-2, and +1.8 over GDN, while Mamba-3 (SISO) improves over the next best model, GDN, by +0.6 points.\u300d\u3068Mamba\u306e\u6700\u65b0\u7248\u3002\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\u3067\u306fTransformer\u3068\u72b6\u614b\u7a7a\u9593\u30e2\u30c7\u30eb\u306e\u30cf\u30a4\u30d6\u30ea\u30c3\u30c9\u69cb\u6210\u304c\u591a\u304f\u3001\u671f\u5f85\u5927\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[235],"class_list":["post-8342","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-mamba"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8342","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8342"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8342\/revisions"}],"predecessor-version":[{"id":8343,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8342\/revisions\/8343"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8342"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8342"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8342"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}