{"id":12599,"date":"2025-02-17T10:21:44","date_gmt":"2025-02-17T01:21:44","guid":{"rendered":"https:\/\/blogs.mathworks.com\/japan-community\/?p=12599"},"modified":"2025-02-17T10:21:44","modified_gmt":"2025-02-17T01:21:44","slug":"transformer-models-from-hype-to-implementation-jp","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/japan-community\/2025\/02\/17\/transformer-models-from-hype-to-implementation-jp\/","title":{"rendered":"\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5\uff1a\u6d41\u884c\u304b\u3089\u5177\u4f53\u7684\u306a\u6d3b\u7528\u3078"},"content":{"rendered":"<div style=\"margin: 2px 10px 9px 4px; padding: 0px; line-height: 21px; min-height: 0px; white-space: pre-wrap; color: #212121; font-family: Helvetica, Arial, sans-serif; font-style: normal; font-size: 14px; font-weight: 400; text-align: left;\">\u203b\u3053\u306e\u6295\u7a3f\u306f 2024 \u5e74 10 \u6708 31 \u65e5\u306b<a href=\"https:\/\/blogs.mathworks.com\/deep-learning\/2024\/10\/31\/transformer-models-from-hype-to-implementation\/?from=jp\">Artificial Intelligence \u3078 \u6295\u7a3f<\/a>\u3055\u308c\u305f\u3082\u306e\u306e\u6284\u8a33\u3067\u3059\u3002<\/div>\n<hr \/>\n<p>&nbsp;<\/p>\n<p>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u4e16\u754c\u3067\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\uff08Transformer\uff09\u30e2\u30c7\u30eb\u304c\u5927\u304d\u306a\u6ce8\u76ee\u3092\u96c6\u3081\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306f\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\uff08NLP\uff09\u304b\u3089\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3\u306b\u81f3\u308b\u591a\u304f\u306e <a href=\"https:\/\/www.mathworks.com\/discovery\/artificial-intelligence.html\">AI \u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3<\/a>\u3067\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u5287\u7684\u306b\u5411\u4e0a\u3055\u305b\u3001\u7ffb\u8a33\u3001\u8981\u7d04\u3001\u3055\u3089\u306b\u306f\u753b\u50cf\u5206\u985e\u306e\u3088\u3046\u306a\u30bf\u30b9\u30af\u3067\u65b0\u3057\u3044\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3092\u6253\u3061\u7acb\u3066\u3066\u3044\u307e\u3059\u3002\u3057\u304b\u3057\u3001\u8a71\u984c\u306e\u80cc\u5f8c\u306b\u306f\u4f55\u304c\u3042\u308b\u306e\u3067\u3057\u3087\u3046\u304b\uff1f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u5358\u306a\u308b AI \u306e\u6700\u65b0\u30c8\u30ec\u30f3\u30c9\u306b\u904e\u304e\u306a\u3044\u306e\u3067\u3057\u3087\u3046\u304b\u3001\u305d\u308c\u3068\u3082 LSTM \u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u3088\u3046\u306a\u4ee5\u524d\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3088\u308a\u3082\u5b9f\u8cea\u7684\u306a\u5229\u70b9\u304c\u3042\u308b\u306e\u3067\u3057\u3087\u3046\u304b\uff1f<\/p>\n<p>\u3053\u306e\u6295\u7a3f\u3067\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u4e3b\u8981\u306a\u5074\u9762\u3092\u63a2\u308a\u3001AI \u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u306b\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u3092\u4f7f\u7528\u3059\u308b\u7406\u7531\u3001\u305d\u3057\u3066 MATLAB \u3067\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u3092\u3069\u306e\u3088\u3046\u306b\u4f7f\u7528\u3059\u308b\u304b\u306b\u3064\u3044\u3066\u8003\u5bdf\u3057\u307e\u3059\u3002<\/p>\n<p style=\"font-size: 22px; color: #c04c0b;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u5165\u9580<\/strong><\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u30012017 \u5e74\u306e\u8ad6\u6587\u300c<a href=\"https:\/\/arxiv.org\/abs\/1706.03762\" class=\"external\" rel=\"nofollow\" target=\"_blank\">Attention Is All You Need<\/a>\u300d\u3067\u7d39\u4ecb\u3055\u308c\u305f\u7279\u5225\u306a\u30af\u30e9\u30b9\u306e<a href=\"https:\/\/www.mathworks.com\/discovery\/deep-learning.html\">\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0<\/a>\u30e2\u30c7\u30eb\u3067\u3059\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u6838\u5fc3\u306f\u3001\u8a00\u8a9e\u3084\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u306e\u3088\u3046\u306a\u9010\u6b21\u30c7\u30fc\u30bf\u3092\u3001\u5f93\u6765\u306e<a href=\"https:\/\/www.mathworks.com\/discovery\/rnn.html\">\u30ea\u30ab\u30ec\u30f3\u30c8\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af<\/a>\uff08RNN\uff09\u3084<a href=\"https:\/\/www.mathworks.com\/discovery\/lstm.html\">\u9577\u77ed\u671f\u8a18\u61b6<\/a>\uff08LSTM\uff09\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3088\u308a\u3082\u52b9\u7387\u7684\u306b\u51e6\u7406\u3059\u308b\u3088\u3046\u306b\u8a2d\u8a08\u3055\u308c\u3066\u3044\u308b\u70b9\u3067\u3059\u3002<\/p>\n<h6><\/h6>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16475 \" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/ml_models.png\" alt=\"Transformer models (e.g. BERT model) are a sub-category of deep learning models (e.g. LSTM), which are a sub-category of machine learning models (e.g. linear regression)\" width=\"669\" height=\"431\" \/><\/p>\n<p><strong>Figure:<\/strong> <a href=\"https:\/\/www.mathworks.com\/discovery\/machine-learning-models.html\">\u6a5f\u68b0\u5b66\u7fd2\u30e2\u30c7\u30eb<\/a>\u306e\u7a2e\u5225\u3068\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb<\/p>\n<p>&nbsp;<\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u80cc\u5f8c\u306b\u3042\u308b\u4e3b\u8981\u306a\u6280\u8853\u306f\u3001<strong>\u81ea\u5df1\u6ce8\u610f\uff08self-attention\uff09\u6a5f\u69cb<\/strong>\u3067\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30e2\u30c7\u30eb\u306f\u5165\u529b\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u7570\u306a\u308b\u90e8\u5206\u306b\u540c\u6642\u306b\u7126\u70b9\u3092\u5f53\u3066\u308b\u3053\u3068\u304c\u3067\u304d\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u306e\u4f4d\u7f6e\u306b\u95a2\u4fc2\u306a\u304f\u51e6\u7406\u3067\u304d\u307e\u3059\u3002RNN \u3068\u306f\u7570\u306a\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u30c7\u30fc\u30bf\u3092\u30b9\u30c6\u30c3\u30d7\u30d0\u30a4\u30b9\u30c6\u30c3\u30d7\u3067\u51e6\u7406\u3059\u308b\u306e\u3067\u306f\u306a\u304f\u3001\u4e26\u5217\u306b\u5165\u529b\u3092\u51e6\u7406\u3057\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u5165\u529b\u30b7\u30fc\u30b1\u30f3\u30b9\u5168\u4f53\u306e\u95a2\u4fc2\u3092\u540c\u6642\u306b\u6349\u3048\u308b\u3053\u3068\u304c\u3067\u304d\u3001\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306b\u5bfe\u3057\u3066\u975e\u5e38\u306b\u9ad8\u901f\u3067\u3001\u304b\u3064\u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u304c\u6ce8\u76ee\u3092\u96c6\u3081\u3066\u3044\u308b\u306e\u306f\u3001\u305d\u306e\u6027\u80fd\u3060\u3051\u3067\u306a\u304f\u67d4\u8edf\u6027\u306b\u3082\u3042\u308a\u307e\u3059\u3002LSTM \u3068\u306f\u7570\u306a\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u5165\u529b\u306e\u51e6\u7406\u6642\u306b\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u9806\u5e8f\u306b\u4f9d\u5b58\u3057\u307e\u305b\u3093\u3002\u305d\u306e\u4ee3\u308f\u308a\u306b\u3001\u5404\u30c8\u30fc\u30af\u30f3\u306e\u4f4d\u7f6e\u306b\u95a2\u3059\u308b\u60c5\u5831\u3092\u8ffd\u52a0\u3059\u308b\u305f\u3081\u306b<strong>\u4f4d\u7f6e\u30a8\u30f3\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0<\/strong>\u3092\u4f7f\u7528\u3057\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u306e\u5c40\u6240\u7684\u304a\u3088\u3073\u5927\u5c40\u7684\u306a\u95a2\u4fc2\u3092\u6349\u3048\u308b\u5fc5\u8981\u304c\u3042\u308b\u30bf\u30b9\u30af\u306b\u5bfe\u3057\u3066\u3088\u308a\u9069\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p style=\"font-size: 22px; color: #c04c0b;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306e\u4e3b\u8981\u69cb\u9020<\/strong><\/p>\n<p>\u00a0\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u3092\u3088\u308a\u3088\u304f\u7406\u89e3\u3059\u308b\u305f\u3081\u306b\u3001\u305d\u306e\u4e3b\u8981\u306a\u69cb\u6210\u8981\u7d20\u3092\u898b\u3066\u307f\u307e\u3057\u3087\u3046\u3002<\/p>\n<ol>\n<li><strong>\u4f4d\u7f6e\u30a8\u30f3\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0<\/strong>:<br \/>\n\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u30c7\u30fc\u30bf\u3092\u4e26\u5217\u306b\u51e6\u7406\u3059\u308b\u305f\u3081\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u306e\u30c8\u30fc\u30af\u30f3\u306e\u9806\u5e8f\u3092\u7406\u89e3\u3059\u308b\u65b9\u6cd5\u304c\u5fc5\u8981\u3067\u3059\u3002\u4f4d\u7f6e\u30a8\u30f3\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u306f\u3001\u30c8\u30fc\u30af\u30f3\u306e\u4f4d\u7f6e\u306b\u95a2\u3059\u308b\u60c5\u5831\u3092\u5165\u529b\u306b\u6ce8\u5165\u3057\u3001\u4e26\u5217\u51e6\u7406\u3055\u308c\u308b\u306b\u3082\u304b\u304b\u308f\u3089\u305a\u3001\u30e2\u30c7\u30eb\u304c\u30b7\u30fc\u30b1\u30f3\u30b9\u69cb\u9020\u3092\u7406\u89e3\u3067\u304d\u308b\u3088\u3046\u306b\u3057\u307e\u3059\u3002<\/li>\n<li><strong>Encoder-Decoder \u6a5f\u69cb<\/strong>:<br \/>\n\u5143\u306e\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u30fb\u30c7\u30b3\u30fc\u30c0\u30fc\u306e\u69cb\u9020\u306b\u57fa\u3065\u3044\u3066\u3044\u307e\u3059\u3002\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u306f\u5165\u529b\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u53d7\u3051\u53d6\u308a\u3001\u8907\u6570\u306e\u5c64\u3092\u901a\u3058\u3066\u51e6\u7406\u3057\u3001\u5185\u90e8\u8868\u73fe\u3092\u4f5c\u6210\u3057\u307e\u3059\u3002\u30c7\u30b3\u30fc\u30c0\u30fc\u306f\u3053\u306e\u8868\u73fe\u3092\u4f7f\u7528\u3057\u3066\u3001\u7ffb\u8a33\u3001\u5206\u985e\u3001\u307e\u305f\u306f\u4ed6\u306e\u7a2e\u985e\u306e\u4e88\u6e2c\u3068\u306a\u308b\u51fa\u529b\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u751f\u6210\u3057\u307e\u3059\u3002<\/li>\n<li><strong>Multi-Head Attention<\/strong>:<br \/>\n\u81ea\u5df1\u6ce8\u610f\uff08self-attention\uff09\u306f\u30e2\u30c7\u30eb\u304c\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u95a2\u9023\u90e8\u5206\u306b\u7126\u70b9\u3092\u5f53\u3066\u308b\u3053\u3068\u3092\u53ef\u80fd\u306b\u3057\u307e\u3059\u3002Multi-head attention \u306f\u8907\u6570\u306e attention \u64cd\u4f5c\u3092\u4e26\u5217\u3067\u5b9f\u884c\u3057\u3001\u30e2\u30c7\u30eb\u304c\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u7570\u306a\u308b\u5074\u9762\u3092\u4e00\u5ea6\u306b\u5b66\u7fd2\u3067\u304d\u308b\u3088\u3046\u306b\u3057\u307e\u3059\u3002\u5404 head \u306f\u5165\u529b\u306e\u7570\u306a\u308b\u90e8\u5206\u306b\u7126\u70b9\u3092\u5f53\u3066\u308b\u3053\u3068\u304c\u3067\u304d\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306b\u3088\u308a\u591a\u304f\u306e\u67d4\u8edf\u6027\u3068\u7cbe\u5ea6\u3092\u4e0e\u3048\u307e\u3059\u3002<\/li>\n<li><strong>\u30d5\u30a3\u30fc\u30c9\u30d5\u30a9\u30ef\u30fc\u30c9\u5c64:\u00a0<\/strong><br \/>\nAttention \u5c64\u306e\u5f8c\u3001\u5404\u30c8\u30fc\u30af\u30f3\u306f\u5b8c\u5168\u306b\u63a5\u7d9a\u3055\u308c\u305f\u30d5\u30a3\u30fc\u30c9\u30d5\u30a9\u30ef\u30fc\u30c9\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u901a\u904e\u3057\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u5c64\u306f\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u3067\u306e\u5404\u30c8\u30fc\u30af\u30f3\u306e\u95a2\u4fc2\u3092\u30e2\u30c7\u30eb\u304c\u3088\u308a\u6df1\u304f\u7406\u89e3\u3059\u308b\u306e\u3092\u52a9\u3051\u307e\u3059\u3002<\/li>\n<\/ol>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16478 size-full\" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/transformer_model_original.png\" alt=\"Originally proposed architecture of a transformer models showing inputs, outputs, and processing blocks\" width=\"439\" height=\"646\" \/><\/p>\n<h6><\/h6>\n<p><strong>Figure:<\/strong> <a href=\"https:\/\/arxiv.org\/abs\/1706.03762\" class=\"external\" rel=\"nofollow\" target=\"_blank\">Vaswani et al, 2017<\/a>\u3067\u767a\u8868\u3055\u308c\u305f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3<\/p>\n<h6><\/h6>\n<p style=\"font-size: 18px;\"><strong>\u305d\u306e\u4ed6\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3<\/strong><\/p>\n<ul>\n<li><strong>Encoder-Decoder Framework:<\/strong> \u5143\u3005\u63d0\u6848\u3055\u308c\u305f\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u30fb\u30c7\u30b3\u30fc\u30c0\u30fc\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u52a0\u3048\u3001\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u3001\u30c7\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3082\u5b9f\u88c5\u3055\u308c\u3066\u3044\u307e\u3059\u3002\n<ol>\n<li>\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u30fb\u30c7\u30b3\u30fc\u30c0\u30fc\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u3001\u4e3b\u306b\u6a5f\u68b0\u7ffb\u8a33\u30bf\u30b9\u30af\u3067\u4f7f\u7528\u3055\u308c\u3001\u4e00\u90e8\u3067\u306f\u7269\u4f53\u691c\u51fa\uff08\u4f8b: <a href=\"https:\/\/arxiv.org\/abs\/2005.12872\" class=\"external\" rel=\"nofollow\" target=\"_blank\">Detection Transformer<\/a>\uff09\u3084\u753b\u50cf\u30bb\u30b0\u30e1\u30f3\u30c6\u30fc\u30b7\u30e7\u30f3\uff08\u4f8b: <a href=\"https:\/\/arxiv.org\/abs\/2304.02643\" class=\"external\" rel=\"nofollow\" target=\"_blank\">Segment Anything Model<\/a>\uff09\u306b\u3082\u4f7f\u7528\u3055\u308c\u307e\u3059\u3002<\/li>\n<li>\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u3001BERT \u3084\u305d\u306e\u30d0\u30ea\u30a8\u30fc\u30b7\u30e7\u30f3\u306e\u3088\u3046\u306a\u30e2\u30c7\u30eb\u3067\u4f7f\u7528\u3055\u308c\u3001\u4e3b\u306b\u5206\u985e\u3084\u8cea\u554f\u5fdc\u7b54\u30bf\u30b9\u30af\u3001\u57cb\u3081\u8fbc\u307f\u30e2\u30c7\u30eb\u306b\u7528\u3044\u3089\u308c\u307e\u3059\u3002<\/li>\n<li>\u30c7\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u3001GPT \u3084 LLaMA \u306e\u3088\u3046\u306a\u30e2\u30c7\u30eb\u3067\u4f7f\u7528\u3055\u308c\u3001\u4e3b\u306b\u30c6\u30ad\u30b9\u30c8\u751f\u6210\u3001\u8981\u7d04\u3001\u30c1\u30e3\u30c3\u30c8\u306b\u7528\u3044\u3089\u308c\u307e\u3059\u3002<\/li>\n<\/ol>\n<\/li>\n<li><strong>Multi-Head Attention Mechanism:<\/strong> \u30de\u30eb\u30c1\u30d8\u30c3\u30c9\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u306b\u306f\u3001\u81ea\u5df1\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\uff08self attention\uff09\u3068\u30af\u30ed\u30b9\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\uff08cross attention\uff09\u3068\u3044\u3046 2 \u3064\u306e\u30d0\u30ea\u30a8\u30fc\u30b7\u30e7\u30f3\u304c\u3042\u308a\u307e\u3059\u3002\u81ea\u5df1\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u304c\u540c\u3058\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u95a2\u9023\u90e8\u5206\u306b\u7126\u70b9\u3092\u5f53\u3066\u308b\u3053\u3068\u3092\u53ef\u80fd\u306b\u3057\u307e\u3059\u3002\u81ea\u5df1\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u30e1\u30ab\u30cb\u30ba\u30e0\u306f\u3001\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u30fb\u30c7\u30b3\u30fc\u30c0\u30fc\u3001\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u3001\u30c7\u30b3\u30fc\u30c0\u30fc\u306e\u307f\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u5b58\u5728\u3057\u307e\u3059\u3002\u4e00\u65b9\u3001\u30af\u30ed\u30b9\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u304c\u7570\u306a\u308b\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u95a2\u9023\u90e8\u5206\u306b\u7126\u70b9\u3092\u5f53\u3066\u308b\u3053\u3068\u3092\u53ef\u80fd\u306b\u3057\u307e\u3059\u3002\u4f8b\u3048\u3070\u3001\u7ffb\u8a33\u30bf\u30b9\u30af\u3067\u82f1\u8a9e\u306e\u6587\uff08\u30af\u30a8\u30ea\uff09\u304c\u30d5\u30e9\u30f3\u30b9\u8a9e\u306e\u6587\uff08\u5024\uff09\u306b\u6ce8\u610f\u3092\u5411\u3051\u308b\u3088\u3046\u306b\u3001\u4e00\u65b9\u306e\u30b7\u30fc\u30b1\u30f3\u30b9\u304c\u3082\u3046\u4e00\u65b9\u306e\u30b7\u30fc\u30b1\u30f3\u30b9\u306b\u6ce8\u610f\u3092\u5411\u3051\u307e\u3059\u3002\u3053\u306e\u30e1\u30ab\u30cb\u30ba\u30e0\u306f\u3001\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u30fb\u30c7\u30b3\u30fc\u30c0\u30fc\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u306e\u307f\u898b\u3089\u308c\u307e\u3059\u3002<\/li>\n<\/ul>\n<h6><\/h6>\n<p style=\"font-size: 22px; color: #c04c0b;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u5229\u70b9<\/strong><\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u3001\u4ee5\u524d\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3068\u6bd4\u3079\u3066\u30b7\u30fc\u30b1\u30f3\u30b9\u30c7\u30fc\u30bf\u306e\u6271\u3044\u65b9\u306b\u5927\u304d\u306a\u5909\u9769\u3092\u3082\u305f\u3089\u3057\u307e\u3057\u305f\u3002\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306f\u3001\u9577\u8ddd\u96e2\u306e\u4f9d\u5b58\u95a2\u4fc2\u3084\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u51e6\u7406\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u4e3b\u306a\u5229\u70b9\u306f\u4ee5\u4e0b\u306e\u901a\u308a\u3067\u3059\u3002<\/p>\n<ol>\n<li><strong>\u4e26\u5217\u51e6\u7406<\/strong>: \u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u304c\u63a1\u7528\u3055\u308c\u308b\u4e3b\u306a\u7406\u7531\u306e\u4e00\u3064\u306f\u3001\u30c7\u30fc\u30bf\u3092\u4e26\u5217\u306b\u51e6\u7406\u3067\u304d\u308b\u80fd\u529b\u3067\u3059\u3002LSTM \u3068\u306f\u7570\u306a\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u81ea\u5df1\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u3092\u7528\u3044\u3066\u30b7\u30fc\u30b1\u30f3\u30b9\u5168\u4f53\u3092\u4e00\u5ea6\u306b\u5206\u6790\u3057\u307e\u3059\u3002\u3053\u306e\u4e26\u5217\u6027\u306b\u3088\u308a\u3001\u7279\u306b\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3067\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u304c\u9ad8\u901f\u5316\u3055\u308c\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u9060\u304f\u96e2\u308c\u305f\u90e8\u5206\u9593\u306e\u4f9d\u5b58\u95a2\u4fc2\u3092\u6349\u3048\u308b\u80fd\u529b\u304c\u5927\u5e45\u306b\u5411\u4e0a\u3057\u307e\u3059\u3002<\/li>\n<li><strong>\u9577\u8ddd\u96e2\u4f9d\u5b58\u95a2\u4fc2\u306e\u51e6\u7406:<\/strong> \u5f93\u6765\u306e\u30b7\u30fc\u30b1\u30f3\u30b9\u30e2\u30c7\u30eb\u3067\u3042\u308b LSTM \u306f\u3001\u9577\u3044\u30b7\u30fc\u30b1\u30f3\u30b9\u306e\u521d\u671f\u90e8\u5206\u306e\u60c5\u5831\u3092\u4fdd\u6301\u3059\u308b\u306e\u306b\u82e6\u52b4\u3057\u307e\u3059\u3002\u30b2\u30fc\u30c8\u306a\u3069\u306e\u30e1\u30ab\u30cb\u30ba\u30e0\u3092\u4f7f\u7528\u3057\u3066\u3053\u306e\u30e1\u30e2\u30ea\u3092\u7ba1\u7406\u3057\u3088\u3046\u3068\u3057\u307e\u3059\u304c\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u304c\u9577\u304f\u306a\u308b\u306b\u3064\u308c\u3066\u52b9\u679c\u304c\u8584\u308c\u307e\u3059\u3002\u4e00\u65b9\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u81ea\u5df1\u30a2\u30c6\u30f3\u30b7\u30e7\u30f3\u3092\u6d3b\u7528\u3057\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u306e\u5404\u30c8\u30fc\u30af\u30f3\u306e\u91cd\u8981\u6027\u3092\u305d\u306e\u4f4d\u7f6e\u306b\u95a2\u4fc2\u306a\u304f\u8a55\u4fa1\u3067\u304d\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u306e\u8981\u7d04\u3084\u30c6\u30ad\u30b9\u30c8\u751f\u6210\u306e\u3088\u3046\u306a\u9577\u671f\u4f9d\u5b58\u95a2\u4fc2\u306e\u7406\u89e3\u304c\u5fc5\u8981\u306a\u30bf\u30b9\u30af\u306b\u7279\u306b\u52b9\u679c\u7684\u3067\u3059\u3002<\/li>\n<li><strong>\u30b9\u30b1\u30fc\u30e9\u30d3\u30ea\u30c6\u30a3:<\/strong> \u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u4e26\u5217\u6027\u306f\u3001\u73fe\u4ee3\u306e\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3001\u7279\u306b\u5927\u898f\u6a21\u306a\u884c\u5217\u6f14\u7b97\u3092\u51e6\u7406\u3059\u308b\u3088\u3046\u8a2d\u8a08\u3055\u308c\u305f GPU \u4e0a\u3067\u975e\u5e38\u306b\u52b9\u7387\u7684\u3067\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30c7\u30fc\u30bf\u30b5\u30a4\u30ba\u304c\u5897\u52a0\u3057\u3066\u3082\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u3046\u307e\u304f\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u3067\u304d\u3001\u30c6\u30ad\u30b9\u30c8\u3001\u97f3\u58f0\u3001\u8996\u899a\u30c7\u30fc\u30bf\u306a\u3069\u5927\u91cf\u306e\u30c7\u30fc\u30bf\u3092\u6271\u3046\u73fe\u5b9f\u306e\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3067\u91cd\u8981\u306a\u7279\u6027\u3068\u306a\u308a\u307e\u3059\u3002<\/li>\n<li><strong>\u30c9\u30e1\u30a4\u30f3\u3092\u8d85\u3048\u305f\u6709\u7528\u6027<\/strong>: \u5f53\u521d\u306f NLP \u30bf\u30b9\u30af\u306e\u305f\u3081\u306b\u8a2d\u8a08\u3055\u308c\u305f\u306b\u3082\u304b\u304b\u308f\u3089\u305a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u3055\u307e\u3056\u307e\u306a\u30c9\u30e1\u30a4\u30f3\u3067\u306e\u9069\u5fdc\u6027\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002\u753b\u50cf\u30c7\u30fc\u30bf\u306b\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3092\u9069\u7528\u3059\u308b\u30d3\u30b8\u30e7\u30f3\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u304b\u3089\u3001\u6642\u7cfb\u5217\u4e88\u6e2c\u3084\u751f\u7269\u533b\u5b66\u30c7\u30fc\u30bf\u5206\u6790\u306b\u81f3\u308b\u307e\u3067\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u67d4\u8edf\u306a\u8a2d\u8a08\u306f\u3055\u307e\u3056\u307e\u306a\u5206\u91ce\u3067\u6210\u529f\u3092\u53ce\u3081\u3066\u3044\u307e\u3059\u3002<\/li>\n<li><strong>\u4e8b\u524d\u8a13\u7df4\u6e08\u307f\u30e2\u30c7\u30eb:<\/strong> BERT\uff08Bidirectional Encoder Representations from Transformers\uff09\u3001GPT\uff08Generative Pre-trained Transformer\uff09\u3001ViT\uff08Vision Transformer\uff09\u306a\u3069\u306e\u4e8b\u524d\u8a13\u7df4\u6e08\u307f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306b\u3088\u308a\u3001\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u3092\u4e00\u304b\u3089\u69cb\u7bc9\u30fb\u8a13\u7df4\u3059\u308b\u3053\u3068\u306a\u304f\u3059\u3050\u306b\u4f7f\u7528\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u4e8b\u524d\u8a13\u7df4\u6e08\u307f\u30e2\u30c7\u30eb\u3092\u4f7f\u7528\u3059\u308b\u3068\u3001\u7279\u5b9a\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306b\u5bfe\u3057\u3066\u306e\u5fae\u8abf\u6574\u3067\u6e08\u307f\u3001\u6642\u9593\u3068\u8a08\u7b97\u8cc7\u6e90\u306e\u4e21\u65b9\u3092\u7bc0\u7d04\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ol>\n<h6><\/h6>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16481 \" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/bert_model.png\" alt=\"Architecture of BERT transformer model\" width=\"311\" height=\"357\" \/><\/p>\n<p><strong>Figure:<\/strong> BERT\u30e2\u30c7\u30eb\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3002\u3053\u308c\u306f\u3001\u5143\u3005\u63d0\u6848\u3055\u308c\u305f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u30a8\u30f3\u30b3\u30fc\u30c0\u30fc\u90e8\u5206\u306e\u307f\u3092\u4f7f\u7528\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<h6><\/h6>\n<p style=\"font-size: 18px;\"><strong>LSTM\u3092\u9078\u3076\u3079\u304d\u30b1\u30fc\u30b9<\/strong><\/p>\n<p>LSTM \u306f\u3001\u7279\u5b9a\u306e\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3084\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u4f9d\u7136\u3068\u3057\u3066\u6709\u7528\u3067\u3059\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u5f37\u529b\u3067\u3059\u304c\u3001\u8a08\u7b97\u30b3\u30b9\u30c8\u304c\u9ad8\u3044\u3068\u3044\u3046\u6b20\u70b9\u304c\u3042\u308a\u307e\u3059\u3002LSTM \u306f\u3001\u77ed\u3044\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u6271\u3046\u30bf\u30b9\u30af\u3001\u4f8b\u3048\u3070\u3001\u30c7\u30fc\u30bf\u304c\u9650\u3089\u308c\u305f\u6642\u7cfb\u5217\u4e88\u6e2c\u306a\u3069\u3067\u3001\u3088\u308a\u30b7\u30f3\u30d7\u30eb\u306a\u69cb\u9020\u3068\u4f4e\u3044\u8a08\u7b97\u8981\u6c42\u304c\u6709\u5229\u3068\u306a\u308b\u5834\u5408\u306b\u3088\u304f\u9078\u3070\u308c\u307e\u3059\u3002\u4f8b\u3048\u3070\u3001<a href=\"https:\/\/blogs.mathworks.com\/deep-learning\/2024\/09\/04\/embedded-ai-integration-with-matlab-and-simulink\/\">\u7d44\u307f\u8fbc\u307fAI\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3<\/a>\u306e\u305f\u3081\u306b\u30e2\u30c7\u30eb\u3092\u8a13\u7df4\u3059\u308b\u5834\u5408\u3001LSTM \u306f\u826f\u3044\u9078\u629e\u80a2\u3067\u3059\u3002\u307e\u305f\u3001\u4e8b\u524d\u8a13\u7df4\u6e08\u307f\u30e2\u30c7\u30eb\u304c\u5229\u7528\u3067\u304d\u306a\u3044\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3067\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u4ee3\u308f\u308a\u306b LSTM \u3092\u8a2d\u8a08\u3059\u308b\u3053\u3068\u3082\u8003\u3048\u3089\u308c\u307e\u3059\u3002<\/p>\n<h6><\/h6>\n<p style=\"font-size: 22px; color: #c04c0b;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u5fdc\u7528\u4f8b<\/strong><\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u3001NLP \u3092\u306f\u3058\u3081\u3068\u3059\u308b\u591a\u304f\u306e\u5206\u91ce\u3067\u5f79\u7acb\u3064\u3053\u3068\u304c\u5b9f\u8a3c\u3055\u308c\u3066\u304d\u3066\u3044\u307e\u3059\u3002<\/p>\n<ul>\n<li><strong><a href=\"https:\/\/www.mathworks.com\/discovery\/natural-language-processing.html\">\u81ea\u7136\u8a00\u8a9e\u51e6\u7406<\/a> (Natural Language Processing, NLP):<\/strong> \u6a5f\u68b0\u7ffb\u8a33\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u8981\u7d04\u3001\u3055\u3089\u306b\u306f\u30c1\u30e3\u30c3\u30c8\u30dc\u30c3\u30c8\u306b\u81f3\u308b\u307e\u3067\u3001BERT \u3084 GPT \u306e\u3088\u3046\u306a\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30d9\u30fc\u30b9\u306e\u30e2\u30c7\u30eb\u306f\u3001\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u65b0\u3057\u3044\u57fa\u6e96\u3092\u6253\u3061\u7acb\u3066\u3066\u3044\u307e\u3059\u3002\u9577\u3044\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u51e6\u7406\u3057\u3001\u6587\u8108\u3092\u6349\u3048\u308b\u80fd\u529b\u306b\u3088\u308a\u3001\u307b\u3068\u3093\u3069\u306e NLP \u30bf\u30b9\u30af\u3067\u9078\u3070\u308c\u308b\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3068\u306a\u3063\u3066\u3044\u307e\u3059\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u6700\u3082\u91cd\u8981\u306a\u6210\u679c\u306e\u4e00\u3064\u306f\u3001GPT \u3084 LLaMA \u306e\u3088\u3046\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\uff08LLM\uff09\u306e\u958b\u767a\u3067\u3059\u3002\u3053\u308c\u3089\u306f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306b\u57fa\u3065\u3044\u3066\u3044\u307e\u3059\u3002<\/li>\n<li><strong><a href=\"https:\/\/www.mathworks.com\/discovery\/computer-vision.html\">\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3<\/a>:<\/strong> ViT \u306e\u5c0e\u5165\u306b\u3088\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306f\u3001\u7279\u306b\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306b\u304a\u3051\u308b\u753b\u50cf\u5206\u985e\u30bf\u30b9\u30af\u3067\u3001<a href=\"https:\/\/www.mathworks.com\/discovery\/convolutional-neural-network.html\">\u7573\u307f\u8fbc\u307f\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af<\/a>\uff08Convolutional Neural Network, CNN\uff09\u3092\u51cc\u99d5\u3057\u59cb\u3081\u3066\u3044\u307e\u3059\u3002<\/li>\n<li><strong><a href=\"https:\/\/www.mathworks.com\/help\/deeplearning\/ug\/time-series-forecasting-using-deep-learning.html\">\u6642\u7cfb\u5217\u4e88\u6e2c<\/a><\/strong>: \u4f1d\u7d71\u7684\u306b LSTM \u304c\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u306b\u4f7f\u7528\u3055\u308c\u3066\u304d\u307e\u3057\u305f\u304c\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306f\u3088\u308a\u9577\u3044\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u51e6\u7406\u3057\u3001\u8907\u96d1\u306a\u30d1\u30bf\u30fc\u30f3\u3092\u6349\u3048\u308b\u80fd\u529b\u304b\u3089\u3001\u3053\u308c\u3089\u306e\u30bf\u30b9\u30af\u306b\u307e\u3059\u307e\u3059\u9069\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/li>\n<\/ul>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16484 \" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/transformers_applications.png\" alt=\"Applications of transformer models include NLP, computer vision, and time-series forecasting\" width=\"557\" height=\"356\" \/><\/p>\n<h6><\/h6>\n<p style=\"font-size: 18px;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u3068\u751f\u6210AI\uff08GenAI\uff09<\/strong><\/p>\n<p>BERT \u3084 GPT \u306e\u3088\u3046\u306a\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30d9\u30fc\u30b9\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306f\u3001\u6700\u5148\u7aef\u306e NLP \u30b7\u30b9\u30c6\u30e0\u306e\u57fa\u76e4\u3068\u306a\u308a\u3001\u4eba\u9593\u306e\u8a00\u8a9e\u3092\u7406\u89e3\u3057\u751f\u6210\u3059\u308b\u80fd\u529b\u306b\u304a\u3044\u3066\u524d\u4f8b\u306e\u306a\u3044\u7cbe\u5ea6\u3067\u306e\u30d6\u30ec\u30fc\u30af\u30b9\u30eb\u30fc\u3092\u53ef\u80fd\u306b\u3057\u3066\u3044\u307e\u3059\u3002BERT \u306f\u53cc\u65b9\u5411\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3092\u901a\u3058\u3066\u8a00\u8a9e\u3092\u7406\u89e3\u3059\u308b\u3053\u3068\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u304a\u308a\u3001\u8cea\u554f\u5fdc\u7b54\u3084<a href=\"https:\/\/www.mathworks.com\/discovery\/sentiment-analysis.html\">\u611f\u60c5\u5206\u6790<\/a>\u3068\u3044\u3063\u305f\u30bf\u30b9\u30af\u306b\u975e\u5e38\u306b\u52b9\u679c\u7684\u3067\u3059\u3002\u4e00\u65b9\u3001GPT \u3084\u305d\u306e\u4ed6\u306e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\uff08LLM\uff09\u306f\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u5185\u306e\u6b21\u306e\u5358\u8a9e\u3092\u4e88\u6e2c\u3059\u308b\u3053\u3068\u3067\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3059\u308b\u3053\u3068\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u304a\u308a\u3001\u4e00\u8cab\u6027\u306e\u3042\u308b\u4eba\u9593\u3089\u3057\u3044\u30b3\u30f3\u30c6\u30f3\u30c4\u3092\u751f\u6210\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n<p>\u751f\u6210 AI\uff08GenAI\uff09\u306f\u3053\u306e\u52e2\u3044\u3092\u6d3b\u304b\u3057\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u3092\u5229\u7528\u3057\u3066\u30c6\u30ad\u30b9\u30c8\u3001\u753b\u50cf\u3001\u3055\u3089\u306b\u306f\u97f3\u697d\u3092\u4f5c\u6210\u3057\u307e\u3059\u3002\u5927\u898f\u6a21\u30e2\u30c7\u30eb\u3092\u7279\u5b9a\u306e\u30c9\u30e1\u30a4\u30f3\u306b\u7279\u5316\u3057\u305f\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3067\u5fae\u8abf\u6574\u3059\u308b\u80fd\u529b\u306b\u3088\u308a\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30d9\u30fc\u30b9\u306e GenAI \u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u306f\u3001\u30b3\u30f3\u30c6\u30f3\u30c4\u751f\u6210\u3001\u30ab\u30b9\u30bf\u30de\u30fc\u30b5\u30fc\u30d3\u30b9\u306e\u81ea\u52d5\u5316\u3001\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u958b\u767a\u306a\u3069\u3067\u307e\u3059\u307e\u3059\u6d17\u7df4\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n<p style=\"font-size: 22px; color: #c04c0b;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc x MATLAB<\/strong><\/p>\n<p style=\"font-size: 18px;\"><strong>NLP\u306b\u304a\u3051\u308b\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc<\/strong><\/p>\n<p>MATLAB \u3068 Text Analytics Toolbox \u3092\u4f7f\u7528\u3059\u308b\u3068\u3001\u4e8b\u524d\u8a13\u7df4\u6e08\u307f <a href=\"https:\/\/www.mathworks.com\/help\/textanalytics\/ref\/bert.html\">BERT\u30e2\u30c7\u30eb<\/a>\u3092\u30ed\u30fc\u30c9\u3067\u304d\u307e\u3059\u3002\u3053\u306e BERT \u30e2\u30c7\u30eb\u3092<a href=\"https:\/\/www.mathworks.com\/help\/textanalytics\/ug\/train-bert-document-classifier.html\">\u6587\u66f8\u5206\u985e<\/a>\u3084<a href=\"https:\/\/www.mathworks.com\/help\/textanalytics\/ug\/extract-answers-from-documents-using-BERT.html\">\u62bd\u51fa\u578b\u8cea\u554f\u5fdc\u7b54<\/a>\u306a\u3069\u3001\u3055\u307e\u3056\u307e\u306a NLP \u30bf\u30b9\u30af\u306b\u5bfe\u3057\u3066\u5fae\u8abf\u6574\u3059\u308b\u3053\u3068\u304c\u53ef\u80fd\u3067\u3059\u3002<\/p>\n<p>\u307e\u305f\u3001<a href=\"https:\/\/blogs.mathworks.com\/deep-learning\/2024\/04\/30\/verification-and-validation-for-ai-from-model-implementation-to-requirements-validation\/\">AI \u30e2\u30c7\u30eb\u306e\u691c\u8a3c\u3068\u59a5\u5f53\u6027\u78ba\u8a8d<\/a>\u306b\u304a\u3044\u3066\u91cd\u8981\u306a\u90e8\u5206\u3068\u306a\u308b\u3001\u5206\u5e03\u5916\uff08Out-of-distribution, OOD\uff09\u30c7\u30fc\u30bf\u306e\u691c\u51fa\u3082\u884c\u3048\u307e\u3059\u3002OOD \u30c7\u30fc\u30bf\u691c\u51fa\u3068\u306f\u3001\u30c7\u30a3\u30fc\u30d7\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306b\u5165\u529b\u3055\u308c\u305f\u969b\u306b\u4fe1\u983c\u6027\u306e\u4f4e\u3044\u4e88\u6e2c\u3092\u3082\u305f\u3089\u3059\u53ef\u80fd\u6027\u306e\u3042\u308b\u30c7\u30fc\u30bf\u3092\u7279\u5b9a\u3059\u308b\u30d7\u30ed\u30bb\u30b9\u3067\u3059\u3002OOD \u30c7\u30fc\u30bf\u3068\u306f\u3001\u30e2\u30c7\u30eb\u306e\u8a13\u7df4\u306b\u4f7f\u7528\u3055\u308c\u305f\u30c7\u30fc\u30bf\u3068\u306f\u7570\u306a\u308b\u30c7\u30fc\u30bf\u3092\u6307\u3057\u307e\u3059\u3002\u4f8b\u3048\u3070\u3001\u7570\u306a\u308b\u65b9\u6cd5\u3067\u53ce\u96c6\u3055\u308c\u305f\u30c7\u30fc\u30bf\u3001\u7570\u306a\u308b\u6642\u671f\u3001\u7570\u306a\u308b\u6761\u4ef6\u3001\u307e\u305f\u306f\u30e2\u30c7\u30eb\u304c\u5143\u3005\u8a13\u7df4\u3055\u308c\u305f\u30bf\u30b9\u30af\u3068\u306f\u7570\u306a\u308b\u76ee\u7684\u3067\u53ce\u96c6\u3055\u308c\u305f\u30c7\u30fc\u30bf\u306a\u3069\u304c\u8a72\u5f53\u3057\u307e\u3059\u3002\u5177\u4f53\u4f8b\u306b\u3064\u3044\u3066\u306f\u3001\u300c<a href=\"https:\/\/www.mathworks.com\/help\/textanalytics\/ug\/out-of-distribution-detection-for-bert-document-classifier.html\">BERT \u6587\u66f8\u5206\u985e\u5668\u306e\u305f\u3081\u306e\u5206\u5e03\u5916\uff08OOD\uff09\u691c\u51fa<\/a>\u300d\u3092\u53c2\u7167\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n<h6><\/h6>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16487 \" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/ODD_BERT.png\" alt=\"In-distribution and out-of-distribution scores for energy distribution discriminator\" width=\"558\" height=\"399\" \/><\/p>\n<p><strong>Figure:<\/strong> BERT\u6587\u66f8\u5206\u985e\u5668\u306b\u304a\u3051\u308b\u5206\u5e03\u5916\uff08OOD\uff09\u30c7\u30fc\u30bf\u306e\u691c\u51fa<\/p>\n<h6><\/h6>\n<p>MATLAB \u3067\u306f gpt-4\u3001llama3\u3001mixtral \u306e\u3088\u3046\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\uff08LLM\uff09\u306b API \u3092\u901a\u3058\u3066\u30a2\u30af\u30bb\u30b9\u3057\u305f\u308a\u3001\u30e2\u30c7\u30eb\u3092\u30ed\u30fc\u30ab\u30eb\u306b\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u305f\u308a\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u305d\u306e\u5f8c\u3001\u597d\u307f\u306e\u30e2\u30c7\u30eb\u3092\u7528\u3044\u3066\u30c6\u30ad\u30b9\u30c8\u306e\u5206\u6790\u3084\u751f\u6210\u3092\u884c\u3046\u3053\u3068\u304c\u53ef\u80fd\u3067\u3059\u3002MATLAB \u3067 LLM\u306b\u30a2\u30af\u30bb\u30b9\u3057\u3001\u64cd\u4f5c\u3059\u308b\u305f\u3081\u306e\u30b3\u30fc\u30c9\u306f\u3001<a href=\"https:\/\/www.mathworks.com\/matlabcentral\/fileexchange\/163796-large-language-models-llms-with-matlab\">Large Language (LLMs) with MATLAB<\/a>\u30ea\u30dd\u30b8\u30c8\u30ea\u306b\u3042\u308a\u307e\u3059\u3002<\/p>\n<p>LLM \u306b\u30a2\u30af\u30bb\u30b9\u3059\u308b\u305f\u3081\u306e\u65b9\u6cd5\u306f\u4e09\u3064\u3042\u308a\u307e\u3059\u3002MATLAB \u3092 OpenAI\u00ae \u306e Chat Completions API\uff08ChatGPT\u2122 \u3092\u52d5\u304b\u3059API\uff09\u3001Ollama\u2122\uff08\u30ed\u30fc\u30ab\u30eb LLM \u7528\uff09\u3001\u304a\u3088\u3073 Azure\u00ae OpenAI \u30b5\u30fc\u30d3\u30b9\u306b\u63a5\u7d9a\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30aa\u30d7\u30b7\u30e7\u30f3\u306b\u3064\u3044\u3066\u3055\u3089\u306b\u8a73\u3057\u304f\u77e5\u308a\u305f\u3044\u65b9\u306f\u3001\u4ee5\u4e0b\u306e\u904e\u53bb\u306e\u30d6\u30ed\u30b0\u8a18\u4e8b\u3092\u3054\u89a7\u304f\u3060\u3055\u3044:<\/p>\n<ul>\n<li><a href=\"https:\/\/blogs.mathworks.com\/deep-learning\/2024\/01\/22\/large-language-models-with-matlab\/\">OpenAI LLMs with MATLAB<\/a><\/li>\n<li><a href=\"https:\/\/blogs.mathworks.com\/deep-learning\/2024\/07\/09\/local-llms-with-matlab\/\">Local LLMs with MATLAB<\/a><\/li>\n<\/ul>\n<h6><\/h6>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16490 size-full\" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/llms_with_matlab_repo.png\" alt=\"Screenshot of LLMS with MATLAB repository\" width=\"1329\" height=\"931\" \/><\/p>\n<h6><\/h6>\n<p><strong>Figure:<\/strong> File Exchange: Large Language Models (LLMs) with MATLAB<\/p>\n<h6><\/h6>\n<p style=\"font-size: 18px;\">\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3\u306b\u304a\u3051\u308b\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc<\/p>\n<p>MATLAB \u3068 Computer Vision Toolbox \u3092\u4f7f\u7528\u3059\u308b\u3068\u3001\u4e8b\u524d\u8a13\u7df4\u6e08\u307f Vision Transformer (ViT) \u3092\u30ed\u30fc\u30c9\u3057\u3001\u753b\u50cf\u5206\u985e\u3084<a href=\"https:\/\/www.mathworks.com\/discovery\/object-detection.html\">\u7269\u4f53\u691c\u51fa<\/a>\u3001\u30bb\u30de\u30f3\u30c6\u30a3\u30c3\u30af\u30bb\u30b0\u30e1\u30f3\u30c6\u30fc\u30b7\u30e7\u30f3\u3068\u3044\u3063\u305f<a href=\"https:\/\/www.mathworks.com\/help\/vision\/ug\/transfer-learning-using-pretrained-vit-network.html\">\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3\u30bf\u30b9\u30af\u306e\u305f\u3081\u306b\u30d5\u30a1\u30a4\u30f3\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0<\/a>\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002ViT\u306f\u753b\u50cf\u751f\u6210\u306b\u3082\u5229\u7528\u3067\u304d\u307e\u3059\u3002\u307e\u305f\u3001\u753b\u50cf\u5185\u306e\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u306e\u30bb\u30de\u30f3\u30c6\u30a3\u30c3\u30af\u30bb\u30b0\u30e1\u30f3\u30c6\u30fc\u30b7\u30e7\u30f3\u306b\u306f\u3001<a href=\"https:\/\/www.mathworks.com\/help\/images\/ref\/segmentanythingmodel.html\">Segment Anything Model (SAM)<\/a> \u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u3082\u53ef\u80fd\u3067\u3059\u3002\u8a73\u7d30\u306b\u3064\u3044\u3066\u306f\u3001\u300c<a href=\"https:\/\/www.mathworks.com\/help\/images\/getting-started-with-segment-anything-model.html\">Get Started with SAM for Image Segmentation<\/a>\u300d\u3092\u53c2\u7167\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n<h6><\/h6>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"alignnone wp-image-16493 size-full\" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/TransferLearningUsingPretrainedViTNetworkExample_01.png\" alt=\"Architecture of vision transformer model\" width=\"780\" height=\"441\" \/><\/p>\n<p><strong>Figure:<\/strong> MATLAB\u3092\u7528\u3044\u305fVision Transformer (ViT)\u30e2\u30c7\u30eb\u306e\u30d5\u30a1\u30a4\u30f3\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0<\/p>\n<h6><\/h6>\n<p style=\"font-size: 18px;\"><strong>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u8a2d\u8a08<\/strong><\/p>\n<p>MATLAB \u3068 Deep Learning Toolbox \u3092\u4f7f\u7528\u3059\u308b\u3068\u3001<a href=\"https:\/\/www.mathworks.com\/help\/deeplearning\/ref\/nnet.cnn.layer.attentionlayer.html\">attentionLayer<\/a>\u3001<a href=\"https:\/\/www.mathworks.com\/help\/deeplearning\/ref\/nnet.cnn.layer.selfattentionlayer.html\">selfAttentionLayer<\/a>\u3001<a href=\"https:\/\/www.mathworks.com\/help\/deeplearning\/ref\/nnet.cnn.layer.positionembeddinglayer.html\">positionEmbeddingLayer<\/a>\u3068\u3044\u3063\u305f\u5c64\u3092\u5229\u7528\u3057\u3066\u3001\u30bc\u30ed\u304b\u3089\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u3092\u8a2d\u8a08\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u6b21\u56de\u306e\u30d6\u30ed\u30b0\u8a18\u4e8b\u3067\u306f\u3001\u6642\u7cfb\u5217\u4e88\u6e2c\u306e\u305f\u3081\u306e\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306e\u8a2d\u8a08\u65b9\u6cd5\u3092\u304a\u898b\u305b\u3057\u307e\u3059\u3002\u305d\u308c\u307e\u3067\u306e\u9593\u3001<a href=\"https:\/\/www.mathworks.com\/matlabcentral\/fileexchange\/161016-transformer-networks-for-time-series-prediction\">\u5b9a\u91cf\u7684\u91d1\u878d\u306b\u304a\u3051\u308b\u6642\u7cfb\u5217\u4e88\u6e2c\u306b\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u3092\u4f7f\u7528\u3059\u308b\u30c7\u30e2<\/a>\u3092\u3054\u89a7\u304f\u3060\u3055\u3044\u3002<\/p>\n<h6><\/h6>\n<p style=\"font-size: 22px; color: #c04c0b;\"><b>\u307e\u3068\u3081<\/b><\/p>\n<p>\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30e2\u30c7\u30eb\u306f\u3001\u5b66\u8853\u7684\u306a\u30d6\u30ec\u30fc\u30af\u30b9\u30eb\u30fc\u304b\u3089\u5b9f\u4e16\u754c\u306e\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u306b\u304a\u3044\u3066\u975e\u5e38\u306b\u6709\u7528\u306a\u30c4\u30fc\u30eb\u3078\u3068\u9032\u5316\u3057\u307e\u3057\u305f\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u304c\u6301\u3064\u3001\u9577\u8ddd\u96e2\u4f9d\u5b58\u95a2\u4fc2\u3092\u51e6\u7406\u3057\u3001\u30b7\u30fc\u30b1\u30f3\u30b9\u3092\u4e26\u5217\u306b\u51e6\u7406\u3057\u3001\u5927\u898f\u6a21\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306b\u30b9\u30b1\u30fc\u30eb\u3067\u304d\u308b\u80fd\u529b\u306f\u3001NLP \u3084\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3\u3001\u3055\u3089\u306b\u306f\u305d\u308c\u3092\u8d85\u3048\u308b\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u3001\u6b20\u304b\u305b\u306a\u3044\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3068\u306a\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u3053\u306e\u6280\u8853\u306f\u9032\u5316\u3092\u7d9a\u3051\u3066\u304a\u308a\u3001MATLAB \u3067\u306e\u5229\u7528\u3082\u53ef\u80fd\u3067\u3059\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u306e\u6a5f\u80fd\u3092\u6d3b\u7528\u3057\u3066\u3001\u3042\u306a\u305f\u81ea\u8eab\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u5f37\u5316\u3057\u307e\u3057\u3087\u3046\u3002\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u306b\u3088\u3063\u3066\u5b9f\u73fe\u3057\u305f\u6210\u679c\u306b\u3064\u3044\u3066\u3001\u305c\u3072\u30b3\u30e1\u30f3\u30c8\u6b04\u3067\u6559\u3048\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<div class=\"overview-image\"><img decoding=\"async\"  class=\"img-responsive\" src=\"https:\/\/blogs.mathworks.com\/deep-learning\/files\/2024\/10\/ml_models.png\" onError=\"this.style.display ='none';\" \/><\/div>\n<p>\u203b\u3053\u306e\u6295\u7a3f\u306f 2024 \u5e74 10 \u6708 31 \u65e5\u306bArtificial Intelligence \u3078 \u6295\u7a3f\u3055\u308c\u305f\u3082\u306e\u306e\u6284\u8a33\u3067\u3059\u3002<\/p>\n<p>&nbsp;<br \/>\n\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u4e16\u754c\u3067\u306f\u3001\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\uff08Transformer\uff09\u30e2\u30c7\u30eb\u304c\u5927\u304d\u306a\u6ce8\u76ee\u3092\u96c6\u3081\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306f\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\uff08NLP\uff09\u304b\u3089\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30d3\u30b8\u30e7\u30f3\u306b\u81f3\u308b\u591a\u304f\u306e AI&#8230; <a class=\"read-more\" href=\"https:\/\/blogs.mathworks.com\/japan-community\/2025\/02\/17\/transformer-models-from-hype-to-implementation-jp\/\">read more >><\/a><\/p>\n","protected":false},"author":159,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[182,61],"tags":[],"_links":{"self":[{"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/posts\/12599"}],"collection":[{"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/users\/159"}],"replies":[{"embeddable":true,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/comments?post=12599"}],"version-history":[{"count":8,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/posts\/12599\/revisions"}],"predecessor-version":[{"id":12875,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/posts\/12599\/revisions\/12875"}],"wp:attachment":[{"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/media?parent=12599"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/categories?post=12599"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blogs.mathworks.com\/japan-community\/wp-json\/wp\/v2\/tags?post=12599"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}