人工ç¥è½ãå®ç¾ããå¦ç¿ã¢ã«ã´ãªãºã ã«å¿ è¦ãªè½å
ä»å¹´ã¯ãDeep Learningãç 究ããäºå®ï¼2014/1/4ï¼ã ã£ãã®ã ããã©ãå¤å±¤ãã¼ã»ãããã³ã¾ã§å°éããï¼2014/2/5ï¼ã¨ããã§å°ã ï¼ï¼ï¼è¶³è¸ã¿ãã¦ãããDeep Learningã®æ§æè¦ç´ ã§ãããã«ããã³ãã·ã³ãç解ããã®ã«æéåã£ã¦ããããã ããã«ããã³ãã·ã³ã®ç解ã«ã¯ããã«ã³ã確çå ´ãMCMCã®ç解ãå¿ è¦ãªãã¨ãããã£ãã®ã§å°ãå»»ãéãã¦ã¢ã³ãã«ã«ãæ³ãå ã«åå¼·ï¼2014/6/20ï¼ãã¦ããã¨ããããã
ãã ãããã°ããã§ã¯å°ã éå±ã«ãªã£ã¦ããã®ã§å°ãå åãã㦠Deep Learning ã®å é§è ã®Bengioãããæ¸ããè«æ Learning Deep Architectures for AI ãåå¼·ãã¦ããã示åã«å¯ãè¦è§£ãå¤ãã®ã§ãã¨ã§æ¯ãè¿ãããããã«è¨é²ãã¦ããããã
ã¾ãã¯ã1.1ç¯ã®Desiderate for Learning AIã®é¨åã人工ç¥è½ãå®ç¾ããå¦ç¿ã¢ã«ã´ãªãºã ã«å¿ è¦ã ã¨æãããè½åãã¾ã¨ãããã¦ãããã¡ãã£ã¨æ¸ãåºãã¦ãããè±èªãé£ããã¦ç¿»è¨³ãæªãããã©ã»ã»ã»
- è¤éã§é常ã«å¤æ§ãªé¢æ°ãå¦ç¿ããè½åããã¨ãã°ãè¨ç·´ãã¼ã¿ã®æ°ããå¤æ§æ§ã®æ°ããã£ã¨å¤§ããé¢æ°ãå¦ç¿ããè½åã
- 人工ç¥è½ã¿ã¹ã¯ã«å¿ è¦ã¨ãããè¤éãªæ©è½ã表ãã®ã«å½¹ç«ã¤ãä½æ¬¡ãä¸æ¬¡ãé«æ¬¡ã®æ½è±¡æ¦å¿µã人éã®æå©ããªãã§å¦ç¿ããè½åã
- é常ã«å¤§éã®ä¾ããå¦ç¿ããè½åãè¨ç·´ã®è¨ç®æéãè¨ç·´æ°ã«å¯¾ãã¦ããï¼ãã¨ãã°ãç·å½¢ã«è¿ããªã¼ãã¼ã§ï¼ã¹ã±ã¼ã«ããã¹ãã
- 主ã«ã©ãã«ãªãã®ãã¼ã¿ããå¦ç¿ããè½åããã¨ãã°ãè¨ç·´ãã¼ã¿ã®ä¸é¨ã«æ£ããã©ãã«ãä»ãã¦ããªãåæ師ããå¦ç¿ã®è¨å®ã§åããã¨ã
- ããããã®ã¿ã¹ã¯ã«åå¨ããç¸ä¹ä½ç¨ãå©ç¨ããè½åããã¨ãã°ããã«ãã¿ã¹ã¯å¦ç¿ããã¹ã¦ã®AIã¿ã¹ã¯ã¯åãåºç¤ããªãäºå®ã«å¯¾ãã¦ç°ãªãè¦æ¹ãæä¾ãããããããã®ç¸ä¹ä½ç¨ã¯åå¨ããã
- ããããã®ã¿ã¹ã¯ãããããã¤å°æ¥ã®ã¿ã¹ã¯ãäºåã«ããããªãã¨ããå¶ç´ã®ä¸ã§ã¯ãå¼·åãªæ師ãªãå¦ç¿ï¼ãã¨ãã°ã観測ãã¼ã¿å ã®çµ±è¨çæ§é ãæ½åºããè½åï¼ãéè¦ãªè¦ç´ ã¨ãªãã
以ä¸ã¯ãDeep Learningã¨ã¯é¢ä¿ãªãããåããããéè¦ãªè½åã¨ãã¦æãããã¦ããã
- å¤æ§ãªé·ãã¨æ§é ã®æèã表ç¾ãããã¨ãå¦ã¶è½åã
- æ©æ¢°ã観測ã®æµãã®ä¸ã§åä½ããè¡åã®æµããçæã§ãããã¨ã
- è¡åãæªæ¥ã®è¦³æ¸¬ã¨æªæ¥ã®å ±é ¬ã«å½±é¿ããç¶æ³ä¸ã§æ±ºå®ãä¸ããè½åããã¶ããå¼·åå¦ç¿ã®ãã¨ã
- ä¸çã«ã¤ãã¦ããé¢é£ã®ããæ å ±ãåéããããã«æªæ¥ã®è¦³æ¸¬ã«å½±é¿ãä¸ããè½åããã¨ãã°ãè½åå¦ç¿ã
åªå 度ã®éãã¯ããããã ãã©ãç¹ã«ç°è«ã¯ãªããããã¤ããéè¦ãªãã¼ã¯ã¼ããå¾ãããã®ã§åå¥ã®ãã¼ãã«ã¤ãã¦ã¯å¾ã§æãä¸ãããããã®è«æã«é¢ããåå¼·ä¼ã®è³æãæ¾å°¾ãããã¢ãããã¼ãããã¦ããã®ã§å 容ãç解ããä¸ã§åèã«ãªã£ãã
ã¡ãªã¿ã«ãç§ã人工ç¥è½ã®éè¦ãªè½åã ã¨èãã¦ããã®ã¯ã
- é£æ³ããè½å
- é¸å¥½ãåµçºãããè½å
- æéã®ã¢ã«ã´ãªãºã ã§ç¡éã®å¤æ§æ§ãçæããè½å
Deep Learningã«èå³ãæã£ãã®ã¯ãåæ£è¡¨ç¾ãå¦ç¿ã§ããã¨ãããã¨ã§1çªç®ã®ãã¼ãã¨é¢ä¿ãããããããã¥ã¼ã©ã«ãããã®æ¡å¼µã¨ãããã¨ã§ä½ã¨ãªãèå¥ããã§ããªãã¨æã£ã¦ããã®ã ããåã«Twitterã§
@mambo_bab DNNã¯ç«ã®ããããã®ç»åâç«ã©ãã«ã¨ããæµãã ã¨æãã®ã§ãããããããã®ã¯ãã®éã§ç«ã¨ããã©ãã«âè²ã
ãªç«ã®ç»åã¨ãããã®ãä½ãããã®ã§ãããã¡ããå
ã®å¦ç¿ã³ã¼ãã¹ã«å«ã¾ããªãç«ã®ç»åãã§ããDNNã®ã¢ãã«ã§ãã®ãã¿ã¼ã³çæãã§ããã®ãç¥ããããªã¨ã
— aidiary (@sylvan5) 2014, 4æ 3
ã¨ã¤ã¶ããã¦ãããåæ£è¡¨ç¾ãããã¿ã¼ã³ãçæã§ããShape Boltzmann Machineï¼PDFï¼ã¨ããã®ããããã¨æãã¦ããã£ããæ¦è¦ãè¦ãã¨
We show that the ShapeBM characterizes a strong model of shape, in that samples from the model look realistic and it can generalize to generate samples that differ from training examples. We find that the ShapeBM learns distributions that are qualitatively and quantitatively better than existing models for this task.
ã¨æ¸ãã¦ãããããã ã¨ããã¨3çªç®ã¨ãé¢ä¿ãåºã¦ããããã§Deep Learningãåå¼·ããã¢ããã¼ã·ã§ã³ããããããã¦ãããã¾ãç´¹ä»ãã¦ããã ããè«æã¯é£ããã¦ã¾ã ç解ã§ããªãã£ãããã§ããï¼ç¬ï¼
ã¾ãå¾ã§æç´ãæ¸ããã¨æãããã©ãå®ãè¨ãã¨ãã¥ã¼ã©ã«ãããã®ã¢ããã¼ãã¯æãããã¾ã好ãã§ã¯ãªãã£ããæ°å¦ãé£ãããããã»ã»ã»ããã©ããã°ã£ã¦ç解ã§ããããã«ãªãããã楽ãããã«æãã¦ããããã