Large Concept Models: Language Modeling in a Sentence Representation Space
Large Concept Models: Language Modeling in a Sentence Representation Space 概è¦LLMs have revolutionized the field of artificial intelligence and have emerged as the de-facto tool for many tasks. The current established technology of LLMs is to process input and generate output at the token level. This is in sharp contrast to humans who operate at multiple levels of abstraction, well beyond single wo
オープンソースã®RAG UI「kotaemonã€ã‚’試ã™
2024/11/01追記 GraphRAGã®å®Ÿè£…ã«ã€nano-graphragãŒè¿½åŠ ã•ã‚ŒãŸã‚ˆã†ã§ã™ã€‚ https://x.com/kagamih/status/1852282744694587509 MSã®GraphRAGよりもコンパクトã«ä½¿ãˆã‚‹ã¨æ€ã†ã®ã§ã€å€‹äººçš„ã«ã¯ãã¡ã‚‰ã‚’オススメã—ã¾ã™ï¼ˆã¾ã 試ã—ã¦ã„ã¾ã›ã‚“ãŒï¼‰ nano-graphragã‚’å…ƒã«ã—ãŸLightRAGã«ã¤ã„ã¦ã¯ä»¥ä¸‹ã«ã¾ã¨ã‚ã¦ã¾ã™ã€‚ https://zenn.dev/kun432/scraps/1f28e5d20dfdf5 ã‚ã¨ä¸‹ã«æ›¸ã„ã¦ã‚‹ä¸å…·åˆã‚‚確ã‹ã‚‚ã†ç›´ã£ã¦ã‚‹ã¨æ€ã†ã‘ã©ã€å®Ÿéš›ã«è©¦ã—ã¦ã„ãªã„ã®ã¨ã€nano-graphragã®ã»ã†ãŒã„ã„ã‹ãªã¨æ€ã†ã®ã§ã€ã‚‚ã¯ã‚„本記事ã¯obsoleteã¨ã„ã†ã“ã¨ã§ã€‚ 2024/09/03追記 ã¡ã‚‡ã£ã¨X経由ã§è¦‹ã«æ¥ã¦ã„ãŸã ã„ãŸæ–¹ãŒå¢—ãˆã¦ã„るよã†ãªã®ã§ã‚らã‹ã˜ã‚注æ„。 kotaemonã§æ™®é€šã®RAG
2024年度 第38回人工知能å¦ä¼šå…¨å›½å¤§ä¼šï¼ˆJSAI2024）å‚åŠ ãƒ¬ãƒãƒ¼ãƒˆ - Insight Edge Tech Blog
ã¯ã˜ã‚ã¾ã—ã¦ï¼2024å¹´5月よりInsight Edgeã«ã‚¸ãƒ§ã‚¤ãƒ³ã—ãŸData Scientistã®å¸‚å·ã§ã™ã€‚ ã¾ã 入社ã—ã¦é–“ã‚‚ãªã„ã§ã™ãŒã€ã‚ªãƒ«ã‚¿ãƒŠãƒ†ã‚£ãƒ–データを活用ã—ãŸæ¡ˆä»¶ã‚„ã€é‡‘èžå¸‚å ´ã®ã‚ªãƒ—ションç‰ã‚’活用ã—ãŸåˆ†æžãªã©ã«æºã‚ã£ã¦ãŠã‚Šã¾ã™ã€‚ 今回ã¯ã€å…ˆæ—¥äººå·¥çŸ¥èƒ½å¦ä¼šï¼ˆJSAI2024）ã«è¡Œã£ã¦ãã¾ã—ãŸã®ã§ã€ãã®ãƒ¬ãƒãƒ¼ãƒˆã‚’ã•ã›ã¦é ‚ãã¾ã™ã€‚ イベントã®æ¦‚è¦ ç™ºè¡¨ã®æ¦‚è¦ [2J1-KS-19] 金èžåˆ†é‡Žã«ãŠã‘る大è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã®æ´»ç”¨ [2A2-PS-2] 進化ã™ã‚‹å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ« [2O4-OS-25a] ä¸å‹•ç”£ã¨AI [2O4-OS-25a-01] ä½å®…ä¾¡æ ¼äºˆæ¸¬ãƒ¢ãƒ‡ãƒ«ã®çµŒæ™‚çš„ãªç²¾åº¦ã®å¤‰åŒ–åˆ†æž [2O4-OS-25a-02] 地ç†ç©ºé–“ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ‡ãƒ¼ã‚¿ã¨æ©Ÿæ¢°å¦ç¿’を用ã„ãŸèª¬æ˜Žå¯èƒ½æ€§ã®é«˜ã„賃料予測 [2O4-OS-25a-03] 機械å¦ç¿’を用ã„ãŸç‰©ä»¶è¨å‚™ã‚¹ã‚³ã‚¢ã®æŽ¨å®šï¼šä¸å‹•ç”£ãƒ‡ãƒ¼ã‚¿ã‚’使用ã—ãŸã‚±ãƒ¼ã‚¹ã‚¹ã‚¿ãƒ‡
T-LEAF: Taxonomy Learning and EvaluAtion Framework
How we applied qualitative learning, human labeling and machine learning to iteratively develop Airbnb’s Community Support Taxonomy. By: Mia Zhao, Peggy Shao, Maggie Hanson, Peng Wang, Bo Zeng BackgroundTaxonomies are knowledge organization systems used to classify and organize information. Taxonomies use words to describe things — as opposed to numbers or symbols — and hierarchies to group things
RecBole
A unified, comprehensive and efficient recommendation library General and extensible data structure We design general and extensible data structures to unify the formatting and usage of various recommendation datasets. Comprehensive benchmark models and datasets We implement more than 100 commonly used recommendation algorithms, and provide the formatted copies of 43 recommendation datasets. Exten
KAN: Kolmogorov-Arnold Networks
Inspired by the Kolmogorov-Arnold representation theorem, we propose Kolmogorov-Arnold Networks (KANs) as promising alternatives to Multi-Layer Perceptrons (MLPs). While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights"). KANs have no linear weights at all -- every weight parameter is replaced by a univariate function parametriz
Scikit-learnã§ã‚ˆã使ã†å›žå¸°ãƒ¢ãƒ‡ãƒ«ã¾ã¨ã‚
回帰モデルã®è©•ä¾¡ MSE:å¹³å‡äºŒä¹—誤差 from sklearn.metrics import mean_squared_error scores = mean_squared_error(val_y, pred) RMSE import numpy as np from sklearn.metrics import mean_squared_error scores = np.sqrt(mean_squared_error(val_y, pred)) R^2:決定係数 from sklearn.metrics import r2_score scores = r2_score(val_y, pred) å‚考 sklearnã®å›žå¸°ãƒ¢ãƒ‡ãƒ« LinearRegression from sklearn.linear_model import LinearRegression lr = Linea
Sakana AI
更新（2025å¹´1月28日） 論文「Evolutionary Optimization of Model Merging Recipesã€ãŒè«–文誌「Nature Machine Intelligenceã€ã«æŽ¡æŠžã•ã‚Œæœ¬æ—¥æŽ²è¼‰ã•ã‚Œã¾ã—ãŸã€‚最新ãƒãƒ¼ã‚¸ãƒ§ãƒ³ã§ã¯æœ¬ã‚¢ãƒ—ãƒãƒ¼ãƒã‚’ã•ã‚‰ã«å®Ÿè¨¼ã™ã‚‹æ–°ãŸãªå®Ÿé¨“çµæžœã‚’å«ã‚“ã§ã„ã¾ã™ã€‚ãœã²ä»¥ä¸‹ã‹ã‚‰ã”覧ãã ã•ã„。 https://www.nature.com/articles/s42256-024-00975-8 Sakana AIã¯2024å¹´3月ã«ã€Œé€²åŒ–的モデルマージã€ã‚’公開ã—ã€å¤§ããªå響を呼ã³ã¾ã—ãŸã€‚公開時ã«ã¯å›½å†…外ã®å¤šãã®ãƒ¡ãƒ‡ã‚£ã‚¢ã«å–り上ã’られニュースã«ãªã‚Šã¾ã—ãŸã€‚進化的モデルマージã¯mergekitã‚„Optuna Hubã¨ã„ã£ãŸè‘—åãªOSSフレームワークã«ã‚‚実装ã•ã‚Œã€å¤šæ§˜ãªãƒ¦ãƒ¼ã‚¶ãƒ¼ãŒãれを活用ã—ã€æ•°ã€…ã®å€‹æ€§çš„ãªãƒ¢ãƒ‡ãƒ«ãŒä½œæˆãƒ»å…¬é–‹ã•ã‚Œã¦ãã¾ã—ãŸã€‚ã¾ãŸã€ç¤¾å†…外ã®è¤‡
Transformerã«ã‚ˆã‚‹æ™‚系列データ予測ã®ã”紹介 - Platinum Data Blog by BrainPad ブレインパッド
本記事ã¯ã€å½“社オウンドメディア「Doorsã€ã«ç§»è»¢ã—ã¾ã—ãŸã€‚ ç´„5秒後ã«è‡ªå‹•çš„ã«ãƒªãƒ€ã‚¤ãƒ¬ã‚¯ãƒˆã—ã¾ã™ã€‚ 発展を続ã‘る「自然言語処ç†ã€æŠ€è¡“。ãã®ä¸ã§ã‚‚幅広ã使用ã•ã‚Œã‚‹æ·±å±¤å¦ç¿’モデルTransformerã¯ã€è‡ªç„¶è¨€èªžå‡¦ç†ä»¥å¤–ã®éŸ³æ¥½ç”Ÿæˆã‚„物体検出ãªã©ã®é ˜åŸŸã§ã‚‚é©ç”¨ã§ãã¾ã™ã€‚本ブãƒã‚°ã§ã¯ã€ Transformerを時系列データã«é©ç”¨ã™ã‚‹æ–¹æ³•ã‚’ã”紹介ã—ã¾ã™ã€‚ ã“ã‚“ã«ã¡ã¯ã€AIソリューションサービス部ã®äº•å‡ºã¨ç”³ã—ã¾ã™ã€‚ ã“ã®è¨˜äº‹ã§ã¯ã€ç‰¹ã«è‡ªç„¶è¨€èªžå‡¦ç†åˆ†é‡Žã§å¹…広ã使用ã•ã‚Œã‚‹æ·±å±¤å¦ç¿’モデルTransformerを時系列データã¸é©ç”¨ã™ã‚‹æ–¹æ³•ã«é–¢ã—ã¦ã”紹介ã—ã¾ã™ã€‚ 以å‰ã®è¨˜äº‹ã§ã¯ã€Transformerã®æ§‹é€ や特徴ãªã©ã«ã¤ã„ã¦ã€è‡ªç„¶è¨€èªžå‡¦ç†åˆ†é‡Žã®æ©Ÿæ¢°ç¿»è¨³ã‚’例ã¨ã—ã¦ã”紹介ã—ã¦ãŠã‚Šã¾ã™ã€‚ã¯ã˜ã‚ã«ã€ã“ã¡ã‚‰ã®è¨˜äº‹ã‚’ã”一èªã—ã¦ã„ãŸã ãã“ã¨ã§ã€ã‚ˆã‚Šæœ¬è¨˜äº‹ã§ã”紹介ã™ã‚‹å†…容ã«å¯¾ã™ã‚‹ç†è§£ãŒæ·±ã¾ã‚‹ã‹ã¨æ€ã„ã¾ã™ã€‚ Transform
Zero-shot Learning網羅的サーベイ：CLIPãŒåˆ‡ã‚Šé–‹ã„ãŸVision & Languageã®æ–°ã—ã„世界 - エクサウィザーズ Engineer Blog
ã“ã‚“ã«ã¡ã¯ï¼ã€€ç”»åƒã‚·ã‚¹ãƒ†ãƒ グループã§æ©Ÿæ¢°å¦ç¿’エンジニアをやã£ã¦ã„ã‚‹å°å³¶ã§ã™ã€‚ ã“ã®è¨˜äº‹ã§ã¯ã€ä»Šãƒ›ãƒƒãƒˆãªã€ŒZero-shot Learningã€ã¨ã€ŒVision & Languageã€ã«é–¢ã™ã‚‹æœ€æ–°æƒ…å ±ã‚’ã€CLIPã¨ã„ã†ç ”究を起点ã¨ã—ã¦ç¶²ç¾…çš„ã«ã‚µãƒ¼ãƒ™ã‚¤ã‚’ã—ã¦ã„ãã¾ã™ã€‚ã“ã®ãŸã‚ã«è«–æ–‡1000本ã«ç›®ã‚’通ã—ã€70本程度を記事ã«ã—ã¾ã—ãŸã€‚ Zero-shotã‚„Vision & Languageã¯ã€Stable Diffusionã«ä»£è¡¨ã•ã‚Œã‚‹ç”»åƒç”ŸæˆAIã¨ã‚‚密接ã«é–¢é€£ã—ã¦ã„る技術ã§ã™ã€‚ã“ã®è¨˜äº‹ã‚’通ã—ã¦ã€Vision & Languageã®å¥¥æ·±ã„世界を体感ã§ãã‚‹ã§ã—ょã†ã€‚ 注æ„äº‹é … ã“ã®è¨˜äº‹ã¯éžå¸¸ã«é•·ã„ãŸã‚ã€å…¨éƒ¨èªã‚€ã®ã«1時間以上ã‹ã‹ã‚‹å¯èƒ½æ€§ãŒã‚ã‚‹ã®ã§ã€ä¼‘憩をå–ã‚ŠãªãŒã‚‰ã€ã¾ãŸã¯å¿…è¦ãªéƒ¨åˆ†ã ã‘èªã‚“ã§ãã ã•ã„。å„セクションを個別ã«èªã‚“ã§ã‚‚å•é¡Œã‚ã‚Šã¾ã›ã‚“。 ã¾ãŸã€æ–‡ç« ä¸ã®ç”»åƒã¯ã€ç‰¹åˆ¥ãªè¨˜è¼‰ãŒãªã„é™ã‚Šã€å¼•ç”¨å…ƒã®è«–
Google Colab 㧠Moore-AnimateAnyone を試ã™ï½œnpaka
「Google Colabã€ã§ã€ŒMoore-AnimateAnyoneã€ã‚’試ã—ãŸã®ã§ã€ã¾ã¨ã‚ã¾ã—ãŸã€‚ 1. Moore-AnimateAnyone「Moore-AnimateAnyoneã€ã¯ã€ã€ŒAnimateAnyoneã€ã®å†ç¾å®Ÿè£…ã§ã™ã€‚å…ƒã®è«–æ–‡ã§å®Ÿè¨¼ã•ã‚ŒãŸçµæžœã‚’一致ã•ã›ã‚‹ãŸã‚ã«ã€ã•ã¾ã–ã¾ãªã‚¢ãƒ—ãƒãƒ¼ãƒã‚„トリックを採用ã—ã¦ã„ã¾ã™ãŒã€ãれらã¯è«–文や別ã®å®Ÿè£…ã¨ã¯å¤šå°‘ç•°ãªã‚‹å ´åˆãŒã‚ã‚Šã¾ã™ã€‚ ã“ã‚Œã¯éžå¸¸ã«æš«å®šçš„ãªãƒãƒ¼ã‚¸ãƒ§ãƒ³ã§ã€AnimateAnyone ã§ç¤ºã•ã‚ŒãŸãƒ‘フォーマンス (テストã§ã¯ç´„ 80%) ã«è¿‘ä¼¼ã™ã‚‹ã“ã¨ã‚’目的ã¨ã—ã¦ã„ã¾ã™ã€‚ 2. ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«æ‰‹é †ã¯ã€æ¬¡ã®ã¨ãŠã‚Šã§ã™ã€‚ (1) パッケージã®ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«ã€‚ # パッケージã®ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ« !git clone https://github.com/MooreThreads/Moore-AnimateAnyone %cd Moor
プãƒé©šã屋AIã‚’ãƒãƒ¼ãƒ ã®Slackã«æ‹›å¾…ã—タイムラインをè’ら㙠- Qiita
20XX å¹´ã€æˆ‘々人類ã¯é€²åŒ–è‘—ã—ã„ AI ã«é©šãã—ã‹ãªã„日々ãŒç¶šã„ã¦ã„ã¾ã™ã€‚ソーシャルメディアã«é©šããŒã‚ãµã‚Œã‚¤ãƒ³ãƒ—レッションを競ã†æ§˜ã¯ã¾ã•ã«å¤§æµ·è³Šæ™‚代ã€ã„ã„ãã®æµ·è³ŠçŽ‹ã«ä¿ºã¯ãªã‚‹ã€ã¨ã°ã‹ã‚Šé£›ã³ã¬ã‘ã¦è€³ç›®ã‚’引ã超新星 ( スーパールーã‚ー ) ãŒé 角をç¾ã—ã¦ã„ã¾ã™ã€‚ 「プãƒé©šã屋ã€ã¨ã—ã°ã—ã°ç§°ã•ã‚Œã‚‹ãƒ«ãƒ¼ã‚ーãŸã¡ã¯ X ( æ—§ Twitter ) ã®ã‚¿ã‚¤ãƒ ラインã«ç¾ã‚Œã¦ã¯æƒ…å ±ã®æ£ç¢ºæ€§ã‚’é‡ã‚“ã˜ã‚‹ã‚¨ãƒ³ã‚¸ãƒ‹ã‚¢ã‚„ç ”ç©¶è€…ã‚’æˆ¸æƒ‘ã‚ã›ã¾ã™ã€‚ãƒãƒ¼ãƒ やコミュニティ内㮠Slack ã¯ãã†ã—ãŸå–§å™ªã‹ã‚‰é›¢ã‚ŒãŸã‚ªã‚¢ã‚·ã‚¹ã¨ã„ãˆã‚‹ã‹ã‚‚ã—ã‚Œã¾ã›ã‚“。ãã‚“ãªã‚ªã‚¢ã‚·ã‚¹ã«ãƒ—ãƒé©šã屋をå¬å–šã—タイムラインをè’らã™ã®ãŒä»Šå›žã®ç›®çš„ã§ã™ã€‚次ã«å®Ÿéš›ã®ä¾‹ã‚’示ã—ã¾ã™ã€‚ ãªã‹ãªã‹æ¨¡å€£ã§ãã¦ã„ã‚‹ã®ã§ã¯ãªã„ã§ã—ょã†ã‹ã€‚オアシスã¯ä»Šã€ã‚¸ãƒ£ãƒ³ã‚°ãƒ«ã«å¤‰ã‚ã‚Šã¾ã—ãŸã€‚ç§ãŸã¡ãŒç”Ÿãã¦ã„る世界ã§ã¯æ£ç¢ºã§åŸå‘³ã•ã‚ŒãŸæƒ…å ±ã ã‘ã«ã‚¢ã‚¯ã‚»ã‚¹ã—ãŸã„ã¨ã„ã†é¡˜ã„ã¯å®Ÿç¾ã•ã‚Œãªã„ã®ã§ã€ç¾
scikit-learn ã§ã‚¯ãƒ©ã‚¹ã‚¿åˆ†æž (K-means 法)
本ページã§ã¯ã€Python ã®æ©Ÿæ¢°å¦ç¿’ライブラリ㮠scikit-learn を用ã„ã¦ã‚¯ãƒ©ã‚¹ã‚¿åˆ†æžã‚’è¡Œã†æ‰‹é †ã‚’紹介ã—ã¾ã™ã€‚ クラスタ分æžã¨ã¯ ã‚¯ãƒ©ã‚¹ã‚¿åˆ†æž (クラスタリング, Clustering) ã¨ã¯ã€ãƒ©ãƒ™ãƒ«ä»˜ã‘ãŒãªã•ã‚Œã¦ã„ãªã„データã«å¯¾ã—ã¦ã€è¿‘ã—ã„属性をæŒã¤ãƒ‡ãƒ¼ã‚¿ã‚’グループ化ã™ã‚‹æ‰‹æ³•ã§ã™ã€‚例をã‚ã’ã‚‹ã¨ã€ä»¥ä¸‹ã®ã‚ˆã†ãªæ´»ç”¨æ–¹æ³•ãŒã‚ã‚Šã€ãƒžãƒ¼ã‚±ãƒ†ã‚£ãƒ³ã‚°æ–½ç–や商å“ã®ä¼ç”»é–‹ç™ºãªã©ã«æ´»ç”¨ã™ã‚‹ã“ã¨ã¾ã™ã€‚ 製å“ã”ã¨ã®ç‰¹å¾´ (自動車ã§ã‚ã‚Œã°ã€ä¾¡æ ¼ã‚„定員ã€ç‡ƒè²»ã€æŽ’æ°—é‡ã€ç›´è¿‘ã®è²©å£²å°æ•°) を用ã„ã¦é¡žä¼¼ã®è£½å“をグループ化 店舗ã®ç‰¹å¾´ (スーパーã§ã‚ã‚Œã°ã€å£²ä¸Šã‚„é¢ç©ã€å¾“æ¥å“¡æ•°ã€æ¥å®¢æ•°ã€é§è»Šå ´ã®æ•°) ã‹ã‚‰é¡žä¼¼ã®åº—舗をグループ化 顧客ã®ç‰¹å¾´ (銀行ã§ã‚ã‚Œã°ã€æ€§åˆ¥ã€å¹´é½¢ã€è²¯è“„残高ã€æ¯Žæœˆã®æ”¯å‡ºã€ä½å®…ãƒãƒ¼ãƒ³ã®åˆ©ç”¨æœ‰ç„¡ãªã©) を用ã„ã¦ä¼¼ãŸã‚ˆã†ãªåˆ©ç”¨å‚¾å‘ã®é¡§å®¢ã‚’グループ化 クラスタ分æžã«ã¯å¤§åˆ¥ã—ã¦ã€K-Means ã«ä»£è¡¨ã•ã‚Œ
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
OCR Benchmark - Omni AI
Reinforcement Learning: An Overview
https://hirokatsukataoka.net/temp/presen/240621CVPR2024Report_finalized.pdf
AlphaFold 3 predicts the structure and interactions of all of life’s molecules
Attention Mechanisms in Computer Vision: CBAM | DigitalOcean
ã€scikit-learn】ニューラルãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ã‚ˆã‚‹å›žå¸°åˆ†æžã€MLPRegressor】
ã€DL輪èªä¼šã€‘How Much Can CLIP Benefit Vision-and-Language Tasks?
{{#tags}}- {{label}}
{{/tags}}