2017-01-01ãã1å¹´éã®è¨äºä¸è¦§
ãã®è¨äºã¯ ããå¼ Advent Calendar 2017 - Adventar 24 æ¥ç®ã®è¨äºã§ãã ã¯ããã« joisino.hatenablog.com ååãå¯ã®ç»åãç¡éã«çæãããã¨ã«ï¼é¨åçã«ï¼æåãã訳ã§ãããç»åãã§ãããä»åº¦ã¯å£°ã欲ãããªã£ã¦ãã¾ãã ããã§ã [1710.08969] Effiâ¦
ãã®å Parikh ã®å®çï¼ããªã¼ã¯ã®å®çï¼ãç¥ã£ã¦ã³ã£ããããã®ã§ç´¹ä»ãã¾ãã å½¢å¼è¨èªçéã§ã¯å¸¸èãããã§ãã ä¸è¡ã§èª¬æã㦠æèèªç±è¨èªã¨æ£è¦è¨èªã¯ãåèªãè¨å·ã®åº¦æ°ã§åä¸è¦ï¼ã¤ã¾ãè¨å·ã®é çªãç¡è¦ããï¼ã¨ãåãã¯ã©ã¹ã«ãªãã¨ãããã®ã§ããâ¦
joisino.hatenablog.com â²æã®è¨äº ååã®è©¦ã¿ããäºå¹´ä»¥ä¸çµã¡ã¾ãããã¾ã ããå¼äºæã¯çºè¡¨ããã¾ããã(*1) äºæãçºè¡¨ãããªããããã®è¨äºã®ã¿ã¤ãã«ãã®ãã®ãã³ããããããæåãããã¨ã«ãªãã¾ããã ãã¯ãä»è©±é¡ã®ãã£ã¼ãã©ã¼ãã³ã°ã§ãªãã¨ããâ¦
大å¦ã®å®é¨ã§å¿ è¦ã«ãªã£ã¦å®è£ ããã®ã§ã¡ã¢ãã¦ããã¾ãã Convolutional LSTM ã®èª¬æ ååã§å®å ¨ã«ãã¿ãã¬ãã¦ãæãå¦ããªãã§ãããConvolutional LSTM ã¨ã¯ãLSTM ã®çµåãå ¨çµåããç³ã¿è¾¼ã¿ã«å¤æ´ãããã®ã§ãã ä¾ãã°ç»åã RNN ã«é£ããã¨ãã«ãä½â¦
è²ããªäººããã£ã¦ããã ãã©èªåç¨ã«ä½ã£ã¦ã¿ã¾ããã ãªãã¸ããªã§ãã github.com ä½¿ãæ¹ ãªãã¸ããªãã¯ãã¼ã³ãã¦åæè¨å®ãã¾ã ç¥çµµå¸«ã®ãªã¹ãã twitter ã§ä½ã£ã¦ç»é²ãã¾ã åããã¾ã slack ã«ç¥çµµãæµãã¦ãã¾ã åä½ä¾ 詳ãã åä½ã®æµãã¯ä»¥ä¸ã®â¦
ã¯ããã« IOI 2013 ãªã¼ã¹ãã©ãªã¢å¤§ä¼ã« Art Class ã¨ããåé¡ãããã¾ãã ãã®åé¡ã¯ãç»åãã¼ã¿ãä¸ããããã®ã§ãã®ç»åã æ§å¼ï¼ï¼æ°é 形主義ã®ç¾ä»£è¸è¡ï¼ æ§å¼ï¼ï¼å°è±¡æ´¾ã®é¢¨æ¯ç»ï¼ æ§å¼ï¼ï¼è¡¨ç¾æ´¾ã®ã¢ã¯ã·ã§ã³ã»ãã¤ã³ãã£ã³ã°) æ§å¼ï¼ï¼ã«ã©ã¼ãã£â¦
符å·ä»ãæ´æ°ã 2 åªã§å²ãã¨ãã«å³ç®è¡ã·ããã使ã£ã¦ãã¾ã£ã¦çãç®ã«ãã£ããã¨ã®ããçããããã«ã¡ã¯ãåã¯ããªãã®ä»²éã§ãã ã³ã³ãã¥ã¼ã¿ã§å²ãç®ãè¡ãã®ã¯ä¸è¬ã«éãã®ã§ãã§ããã ãã·ããçã®è»½ãæä½ã§æ¸ã¾ãããã§ãããã æè¿ã®ã³ã³ãã¤ã©ã¯â¦
ãããå°ããæéè¨ç®é ã§è¡åç©ãè¨ç®ããã·ã¥ãã©ãã»ã³ã®ã¢ã«ã´ãªãºã ã¨ãã®å°åºæ¹æ³ãç´¹ä»ãã¾ãã ã·ã¥ãã©ãã»ã³ã®ã¢ã«ã´ãªãºã ã®è¡åç© ãèãã¾ãã ã¾ã ã $$ \begin{pmatrix} A_{11} & A_{12} \\ A_{21} & A_{22} \end{pmatrix} \begin{pmatrixâ¦
ã±ã¤ãªã¼ã®å ¬å¼ã®è¨¼æãã¡ã®ç´¹ä»ã§ãã ã±ã¤ãªã¼ã®å ¬å¼ã¨ã¯ ã±ã¤ãªã¼ã®å ¬å¼ã¨ã¯ é ç¹ã®ã©ãã«ä»ãã®æ¨ã®ç·æ° ã ã§ããã¨ããå ¬å¼ã®ãã¨ã§ãã ããã§ãã©ãã«ä»ãã§ããã¨ã¯ãããããã®é ç¹ãåºå¥ããã¨ãããã¨ã§ãã ãã¨ãã° ã®ã¨ããé ç¹ãåºå¥ããªãâ¦
æéè¨ç®é <O(n), O(1)> ã® LCA(Lowest Common Ancestor) 㨠RMQ(Range Minimum Query) ã C++ ã§å®è£ ãã¾ããã ã¢ã«ã´ãªãºã ã®è§£èª¬ã¯Dããã®ã¹ã©ã¤ã [1] LCA and RMQ ~ç°¡æ½ããããï¼~ ãã¨ã¦ãåãããããã®ã§ãã¡ããåç §ãã¦ãã ããã æ¦è¦ã ã説æãã¾ãã LCA</o(n),>â¦
IOI2017ï¼ã¤ã©ã³å¤§ä¼ï¼ã«å¯å£é·ã¨ãã¦åå ãã¦ãã¾ããã ioi2017.org IOIã®åå è¨ã¯é¸æã»å½¹å¡ã®ææ³ï¼ã¾ã å ¬éããã¦ããªããããªã®ã§ãã¨ã§ãªã³ã¯ãå¼µãã¾ãï¼ã«æ¸ããã®ã§ãã¡ããåç §ãã¦ãã ããã 第29åå½éæ å ±ãªãªã³ãã㯠ã¤ã©ã³å¤§ä¼ éå ± ãé¢ç½â¦
ãã©ã¼ã©ãã³ç´ æ°å¤å®æ³ã«ã¤ã㦠ãã©ã¼ã©ãã³ç´ æ°å¤å®æ³ã¯ãã§ã«ãã¼ãã¹ããæ¡å¼µããæãã§ãã ãã§ã«ãã¼ãã¹ãã§ã¯ ãç´ æ°ã®ã¨ãããã§ã«ãã¼ã®å°å®çãã 㨠ãäºãã«ç´ ãªã¨ã ã¨ãªããã¨ãå©ç¨ãã¦ã ã¨äºãã«ç´ ãªæ´æ° ãåã£ã¦ ã§ããã°åææ°ã¨å¤æâ¦
è¨èªå¦çç³»ã§è¬ç¾©ã§ã¯ãªã¼ãã®ä¸åç¹å®çããã£ãæã«ãã«ãã³ãã©ã¼ãã®è¨¼æã«ã使ãããªãã¨æã£ãã®ã§ç´¹ä»ãã¾ãã ã¯ãªã¼ãã®ä¸åç¹å®ç ãå®ååé åºéåã¨ãã ããã®ä¸ã®ã¹ã³ããé£ç¶ååã¨ããããã®ã¨ãã ã¯æå°ä¸åç¹ãæã¤ã ããã§ãå®ååé åºâ¦
HashedNets [1] ã chainer ã§å®è£ ãã¾ããã HashedNets ã®èª¬æ ãã¥ã¼ã©ã«ãããã¯ã¼ã¯ã®ãã©ã¡ã¼ã¿æ°ã¯é常ã«å¤ãããã©ã¡ã¼ã¿ã¯åé·ãªè¡¨ç¾ã«ãªã£ã¦ãããã¨ãå¤ãã§ãã ããã§ãèªç±ãªãã©ã¡ã¼ã¿ã®æ°ãæ¸ããã¦ãæ£ååã¨è»½éåãéæããããã®ææ³ã®â¦
æ¬è³ªçã«ææ§ãªæèèªç±è¨èªãåå¨ããã¨ããäºå®ã¯è²ããªã¨ããã§ä½åº¦ãç®ã«ãã¦ããã®ã§ããã証æãè¦ããã¨ãç¡ãã£ãã®ã§ããã§ç´¹ä»ãã¾ãã æµãã¯å¤§ä½ [1] ã¨åãã§ãããè£é¡ã®è¨¼æã¯ãªãªã¸ãã«ãæ··ãã£ã¦ã¾ãï¼æ¬è³ªçã«ã¯åãã ã¨ã¯æãã¾ãï¼ï¼éâ¦
Adam [1] ã chainer ã§å®è£ ãã¾ããã Adam ã®èª¬æ å°ãªãæéã»ç©ºéè¨ç®éã§é«ãæ§è½ãåºãã¨ãããã¨ã§ã深層å¦ç¿ã®åéã§æè¿ãã使ããã¦ããæé©åææ³ã§ãã ããã©ã¡ã¼ã¿ã«åã颿° ã®ï¼æå¾ å¤ï¼æå°åãèãã¾ãã ã® ã«å¯¾ããå¾é ã ã¨ãã¾ãã ä¸â¦
大å¦ã®ã½ããã¦ã§ã¢å®é¨ã§MLã®ã¤ã³ã¿ããªã¿ãä½ã£ã¦ããã®ã§ããããã¹ãç¨ã«MLããã°ã©ã ã®èªåçæããã°ã©ã ãæ¸ãã¾ããã generator.sh ã¯ä¸é¨ @basemusi åã«æ¸ãã¦ããã£ã¦ã¾ãã github.com 忍è«å¨ã®ãããã°ç¨ã«ä½ã£ãã®ã§ãã®ååã§ãããã¤ã³ã¿â¦
Dilated Convolution ã chainer ã§å®è£ ãã¾ããã Dilated Convolution ã®èª¬æ Dilated Convolution ã¯ããã£ã«ã¿ã¼ã¨ã®ç©ãåãç¸æã®ééããããç³ã¿è¾¼ã¿ã®ãã¨ã§ãã ä¾ãã°ã以ä¸ã®ãããªç»åã«ããã¦ã 12 ãä¸å¿ã« 3 x 3 ã®æ®éã®ç³ã¿è¾¼ã¿ãã£ã«ã¿ã¼â¦
Batch Normalization [1] ã chainer ã§å®è£ ãã¾ããã è²ã ãªå ´åã«é©ç¨ã§ãã¦ãå¦ç¿é度ãéããªã£ããæ±åæ§è½ãä¸ãã£ããããããããã¯ã§ãã Batch Normalization ã®èª¬æ ä¸å±¤ï¼åºå層ã«è¿ã層ï¼ã®å ¥åã¯ãå½ç¶ä¸å±¤ï¼å ¥å層ã«è¿ãå´ï¼ã®ãã©ã¡ã¼ã¿ã«â¦
Generative Adversarial Nets [1] ã chainer ã§å®è£ ãã¾ããã ãããã GAN ã§ãã æè¿ããããæ´¾çç³»ãåºã¦ãã¾ããç»åã®çæã¢ãã«ã¯ã»ã¨ãã©ããã®æ´¾çãªæ°ããã¾ãã ç»åãçæãã Generator ï¼ä»¥ä¸ Gï¼ã¨ãç»åãæ¬ç©ã G ãçæãããã®ããèå¥â¦
SRM717 ã® Hard ã§åºé¡ããã¦æ°ã«ãªã£ãã®ã§æ¸ãçãã¦ããã¾ãã Chromatic Polynomial ã«ã¤ã㦠ã¾ããç¡åã°ã©ãã®é ç¹ããè²ã¸ã®å¯¾å¿ã彩è²ã飿¥ããé ç¹ãåãè²ã«ãªããªããããªå½©è²ã®ãã¨ãæ£ãã彩è²ã¨ãããã¨ã«ãã¾ããããã¯ãã®è¨äºéå®ç¨èªã§â¦
å æ¥ãé·ãããã§çãã£ãå®é¨ãã¤ãã«çµãã£ãã®ã§è¨é²ãæ®ãã¦ããã¾ãã ã¯ãã㫠京é½å¤§å¦æ å ±å¦ç§ã®è¨ç®æ©ç§å¦ã³ã¼ã¹ã§ã¯è¨ç®æ©ç§å¦å®é¨åæ¼ç¿ï¼ä»¥ä¸ãå®é¨ï¼ã¨ããå¿ ä¿®ç§ç®ã 2 åçåæ / 徿ã 3 åçåæ / 徿ã®è¨ 4 ã¤ããã¾ããå®é¨ 1 ã¯ããâ¦
ææ¥ãKMC ã®çµåãæé©åèªæ¸ä¼ã§ããäºå®ãªã®ã§ãããå°ã調ã¹ãéãã§ã¯æ¥æ¬èªã®è§£èª¬ãè¦ã¤ãããªãã£ãã®ã§æ¸ãã¦ã¿ã¾ããï¼æ³¨ï¼çµåãæé©åæ¬ä½ã«ã¯ O(nm) ã®ã¢ã«ã´ãªãºã ããè¼ã£ã¦ãã¾ãããï¼ ããããã¢ã«ã´ãªãºã ã«æ £ãã¦ããªã人åãã«æ¸ããã¤â¦