2012-01-01ãã1å¹´éã®è¨äºä¸è¦§
ãµã¤ãã¦ãºã§ãå··ã®æµè¡ãã«ä¹ã£ãã£ã¦ãã¢ããã³ãã«ã¬ã³ãã¼ãªããã®ããã£ã¦ã¦ãæ å½ããè¨äºãä»æ¥å ¬éãããã è¨èªå¤å®ã®ä»çµã¿ - Cybozu Inside Out | ãµã¤ãã¦ãºã¨ã³ã¸ãã¢ã®ããã° é常ã®ã¢ããã³ãã«ã¬ã³ãã¼ã¨éã£ã¦ããã¼ãã¯ãæè¡ãªããªãã§ãâ¦
[Wang+ AISTATS 2011] Online Variational Inference for the Hierarchical Dirichlet. Process HDP-LDA ã Online 㪠VB ã§è§£ãã¨ããè«æã Teh ãããã® VB æ¨è«ã®è«æãèªãã ãã ãã©ãå®ã«ã¢ã¯ãããã£ãã¯ãªå解ãåºã¦ãã¦ãããããã¯ã©ããªãã ããâ¦
Kneser-Ney Smoothing ã¯é«æ§è½ãªè¨èªã¢ãã«ã§ãããã¨ãããèãããã¦ç¥ã£ã¦ããã¤ããã ãã©ãã¾ã ä¸åº¦ã試ãããã¨ããªãã£ãã®ã§ã試ãã¦ã¿ãã ã³ã¼ãã¯ããã https://github.com/shuyo/iir/blob/master/ngram/knsmooth.py å®é¨ç¨ã«ã¹ã£ããæ¸ãã¦ãâ¦
11/9 ã«éå¬ãããæ©æ¢°å¦ç¿Ãããã°ã©ãã³ã°åå¼·ä¼ vol.2 ã«ã®ãã®ãåå ï¼çºè¡¨ã主å¬ã®æç²ãããåå è ï¼çºè¡¨è ã®çãããä¼å ´ãæä¾ãã¦ä¸ãã£ã DeNA ãããããã¨ããããã¾ããã æ©æ¢°å¦ç¿Ãããã°ã©ãã³ã°åå¼·ä¼ vol.2 : ATND æç²ããããçºè¡¨ã®ä¾é ¼ãâ¦
ã¸ã¥ã³ã¯å æ± è¢æ¬åºã«ã¦ 10/11 ã«è¡ãããããã¿ã¼ã³èªèã¨æ©æ¢°å¦ç¿ã(PRML) æ好家ã®éã¾ããããããªãã£ãããã¼ã¯ã»ãã·ã§ã³ã«ã®ãã®ãè¡ã£ã¦ãããã°ãããåã§ããã¹ã£ã¦ããããããã¨ããããã¾ããï¼ãç²ãæ§ã§ããï¼åä½ PRMLåäººèª ããã¿ã¼ã³èªâ¦
æ©æ¢°å¦ç¿ã®å®çªæç§æ¸ã®1ã¤ã¨è¨ãããåå°ã§èªæ¸ä¼ãéãããããã¿ã¼ã³èªèã¨æ©æ¢°å¦ç¿ã(PRML)ãèªã¿è§£ãã«ã¯ããç¨åº¦ã®è§£æã¨ç·å½¢ä»£æ°ã®ç¥èãå¿ è¦ãªãããæ°å¦ãè¦æãªå¦çãããÃÃå¹´ã¶ãã«æ°å¼ãç®ã«ããã¨ããã¨ã³ã¸ãã¢ãã¡ã次ã ã¨ãå¼å¤å½¢ã§ããªãâ¦â¦â¦
HDP-LDA ã®æ´æ°å¼ãå®è£ ã§ãããããããæ£ããåãã¦ããã®ããããã¦åæãããã©ããã確èªããããã« perplexity ãæ±ãããã¨ããã ãããã¡ããä¾ã«ãã£ã¦è«æã«ã¯è©³ç´°ãªæ°å¼ã¯æ¸ãä¸ããã¦ããªãã®ã§ãæå¾ã«ããããã£ã¤ãããã è«æã§ã¯ãã¼ãã¬ãâ¦
Dirichlet Process with Mixed Random Measures (Kim+ ICML2012, ä»¥ä¸ DP-MRM) ãå®è£ ãã¦ã¿ãããã¨è«æã®è¨ç®å¼ãã²ã¨éã確ããã¦ããã®ã ããããã¤ãæ°ã«ãªããã¨ãã k, l ã®ãµã³ããªã³ã° è«æã® (9), (10) å¼ã«ããããã¦ãã k, l ã®å ¨æ¡ä»¶ä»ãäºå¾â¦
æ¨æ¥ã® HDP-LDA ã®æ´æ°å¼å°åºã®ã¤ãã§ã«ãICML èªã¿ä¼ã§ç´¹ä»ããæ師ãããã³ãã© LDA ã§ãã (Kim+ ICML2012) Dirichlet Process with Mixed Random Measures (DP-MRM) ã®æ´æ°å¼ãå°åºãã¦ãããã DP-MRM ã®ã¢ãã«ã«ã¤ãã¦ã¯ ICML èªã¿ä¼ã§ã®çºè¡¨è³æåèâ¦
æ¥é±ããµã¤ãã¦ãºã»ã©ãã¦ã¼ã¹å宿ãªãã¦ã®ããã£ãããã(æ¨å¹´ã®æ§åã¡ãè¦)ã ã¡ãã£ã¨ç¼¶è©°ã£ã½ãæãã§ã³ã¼ããã¬ã¸ã¬ã¸æ¸ãæ©ä¼ã«ãªãã®ã§ããã®åã® ICML èªã¿ä¼ã§ç´¹ä»ãã DP-MRM ã§ãå®è£ ãã¦ã¿ããã¼ãã§ã HDP-LDA ã®å®è£ ã®ç´°ããã¨ããå¿ãã¡ãã£â¦
7/28 ã«è¡ããã nokuno ãã主å¬ã® ICML 2012 ã®è«æèªã¿ä¼ã«ã®ãã®ãåå ããç²ãæ§ã§ããï¼ãããã¨ããããã¾ããï¼åä½ ããã¼ã¾ããããã¯ã¢ãã«ãªã®ï¼ï¼ã¶ã¼ã¶ã¼ï¼ãã¨ãã¼ã¤ã³ã°ãæµ´ã³ã¤ã¤ã[Kim+ ICML12] Dirichlet Process with Mixed Random Measâ¦
å®ã¯æè¿ C# ã§ã¡ãã£ã¨åä½ç¢ºèªãããã©ãããã³ã¼ãããºããºãæ¸ãã¦ããã ã¨ããããã§ãçããï¼ï¼ï¼ã§ããã ãæåããã¹ããæ¸ããã¨ãã¦ããã®ã ããå¤é¨ã®ã©ã¤ãã©ãªã¨ããµã¼ãã¹ã¨ã絡ãã§ãã¦ãã¾ãã«ãã¹ãæ³£ãããªç¶æ³ã ãããä»ããã¬ã¬ã·ã¼â¦
Machine Learning that Matters(ICML 2012) èªãã - ç³ãããå¼æ ¶ (Wagstaff ICML2012) Machine Learning that Matters (åè«æ) æ©æ¢°å¦ç¿ã®ãããã«ã³ãã¡ã¬ã³ã¹ ICML ã«ã¦ããæ©æ¢°å¦ç¿ã¯ãã®ã¾ã¾ã§ããã®ï¼ãã¨ããæãã®è«æãçºè¡¨ãããã ç´°ããå 容ã¯â¦
5/14 ã« NAIST(å¥è¯å 端ç§å¦æè¡å¤§å¦é¢å¤§å¦) ã«ã¦ãâ-gramã使ã£ãçæè¨èªå¤å®ãã«ã¤ãã¦çºè¡¨ãã¦ãã¾ãããç´ æ´ãããæ©ä¼ãããã ãã¦ãããã¨ããããã¾ããã ãã®æã®è³æãå ¬éãã¾ãã å 容㯠TokyoNLP #8 ãè¨èªå¦çå¦ä¼2012ï¼ åºå³¶å¸å¤§ã§çºè¡¨ãããâ¦
ãã¾ã«ã¯è¶£å³ã®ã³ã¼ãã¨ããã£ã¦ããã£ã¤ã趣å³ã®ãã¨ããæ¸ãã¦ãªãããã¨ããããã³ãã¯ããã¨ãã¦ã 以åã Erlang ã§å庫çªã½ã«ãã¼ãæ¸ããããææç¡æ§ã«ããºã«ã解ãã³ã¼ããæ¸ããããªããã ãã©ãä»åã¯ã¤ã©ã¹ããã¸ãã¯ã解ãã¦ã¿ã(ãçµµãããã¸â¦
以åããä¸åº¦åå ãã¦ã¿ããã¨æã£ã¦ãã Tokyo.R #22 ã«ã®ãã®ãè¡ã£ã¦ããã主å¬ã® @yokkuns ãããåå è ï¼çºè¡¨è ã®ã¿ãªãããä¼å ´ãæä¾ãã¦ãã ãã£ããããã£ããããããã¨ããããã¾ããï¼ãç²ããã¾ã§ããã atnd ã« LT åéä¸ã¨ãã£ãã®ã§ãååå â¦
id:nokuno ãã主å¬ã® NIPS èªã¿ä¼ã«ã®ãã®ãåå ããããã¨ããããã¾ãï¼ãç²ãæ§ã§ããï¼åä½ [Karger+ NIPS11] Iterative Learning for Reliable Crowdsourcing Systems ã¨ããã¯ã©ã¦ãã½ã¼ã·ã³ã°ãç¹ã« Amazon Mechanical Turk ã«ã¦ spammer ãããå ´åâ¦
宣ä¼ã§ãã ãµã¤ãã¦ãºã»ã©ãã¦ã¼ã¹ã®æçµææå ±åä¼ãæ¥é± 3/26(æ) ã«éããã¾ãã ãã®å ±åä¼ã¯ç¬¬2æåé説æä¼ãå ¼ãã¦ãã¾ãã ãµã¤ãã¦ãºã»ã©ãã¦ã¼ã¹ ããã° http://cybozulabs-youth.blogspot.jp/ 第1æãµã¤ãã¦ãºã»ã©ãã¦ã¼ã¹ æçµææå ±åä¼(è¥å¹²â¦
NLP2012 ã®ãã¹ã¿ã¼çºè¡¨ã«ã¦ãLDA ã®æåãè¦ããããã³ã«æãããã©ãã©ãã¨è¿å¯ãããã¼ã ãã¼ã ç¡è²¬ä»»ãªãã¨ãè¿°ã¹ã¦ãããã決ã¾ã£ã¦ã shuyo ããã§ãããï¼ã ãªãã§ãããè¶ã®æ°´å¤§ã®å°æå çã®ç 究室ã®å¦çããã¯ã¿ãªããæã£ã¦ï¼ããªï¼ï¼ãããã¯ã¢â¦
è¨èªå¦çå¦ä¼ã®ç¬¬18å年次大ä¼(NLP2012)ã§ã®çºè¡¨ã極大é¨åæååã使ã£ã twitter è¨èªå¤å®ãã§ä½¿ç¨ããè³æãå ¬éãã¾ãã 極大é¨åæååã使ã£ã twitter è¨èªå¤å® from Shuyo Nakatani www.slideshare.netè«æã¯ä¸è¨ãã¼ã¸ã«ã¦å ¬éæ¸ã¿ã 極大é¨åæåâ¦
ã使ãã®ããã¤ã¹ã«ãã£ã¦ã¯ï¼ï¼¤ã«è¦ããªããã¨ãããã¾ããããããããäºæ¿ãã ããã
TokyoNLP #9 㧠tkng ããã MCMC ã§æ£è¦åå¸ãããµã³ããªã³ã°ãã¦ã¿ããã©ã1000件ãããã ã¨ãªããªãããããªé£éåã«ãªããªããã¨ãã話ãããã¦ããã éä¸ã®ç³»åãæ¨ã¦ã¦ãªãã¨ãããã¨ã ã£ãã®ã§ããã®ããã§ã¯ï¼ ã¨æã£ã¦ã¡ãã£ã¨è©¦ãã¦ã¿ãããçµè«â¦
æ¥ãã 3/13ã16 ã«åºå³¶ã§è¡ãããè¨èªå¦çå¦ä¼å¹´æ¬¡å¤§ä¼(NLP2012)ã«ã¦çºè¡¨ããã極大é¨åæååã使ã£ã twitter è¨èªå¤å®ãã®è«æãå ¬éã ä¸è°· ç§æ´, 極大é¨åæååã使ã£ã twitter è¨èªå¤å®, è¨èªå¦çå¦ä¼ç¬¬18年次大ä¼, 2012 http://ivoca.31tools.com/â¦