Á°Àî/¶µ°é/Neural Networksµ
¤ò¥Æ¥ó¥×¥ì¡¼¥È¤Ë¤·¤ÆºîÀ®
[
¥È¥Ã¥×
] [
¿·µ¬
|
°ìÍ÷
|
ñ¸ì¸¡º÷
|
ºÇ½ª¹¹¿·
|
¥Ø¥ë¥×
|
¥í¥°¥¤¥ó
]
³«»Ï¹Ô:
[[Á°Àî/¶µ°é]]
*1st "Introduction" on April 13 [#h4caa8cb]
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤È¤Ï¡© [#sd0bce1c]
-Ǿµ¡Ç½¤Ë¸«¤é¤ì¤ë¤¤¤¯¤Ä¤«¤ÎÆÃÀ¤ò·×»»µ¡¾å¤Î¥·¥ß¥å¥ì¡¼¥·¥ç¥ó¤Ë¤è¤Ã¤Æɽ¸½¤¹¤ë¤³¤È¤òÌܻؤ·¤¿¿ô³Ø¥â¥Ç¥ë¤Ç¤¢¤ë¡£~
by wikipedia
-Advantage
--ÊÂÎóʬ»¶½èÍý·¿
--¼«¸ÊÁÈ¿¥·¿
--¿®ÍêÀ¤¬¹â¤¤¡£<¡á¥Î¥¤¥º¤äÁû²»¤Ë¤¿¤¤¤·¤Æ¶¯¤¤¡£
--¸Î¾ã¤·¤Ë¤¯¤¤¡£
--Turning machine¤ÎÍýÏÀ¤ò¤â¤È¤Ëºî¤é¤ì¤Æ¤¤¤ë¡£
-Disadvantage
--®¤µ¡¢Àµ³Î¤µ¡¢»»½Ñ·×»»¤Î·«¤êÊÖ¤·¤Ë¼å¤¤¡£
--Ã챤ÎÍý²ò¡¢ºÆÍøÍѤ¬¶ì¼ê¡£
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤ÎÎò»Ë [#d3a2c870]
-1943ǯ
--"¤·¤¤¤ÃÍÁǻҥâ¥Ç¥ë" by W.McCulloch and W.pitts
-1949ǯ
--"The Organization of Behavior" by D,Hebb
--¥·¥Ê¥×¥¹¶¯²½Â§=>¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¤Î³Ø½¬ÊýË¡¤Î´ðÁäˡ£
-1962ǯ
--"¥Ñ¡¼¥»¥×¥È¥í¥ó(perceptron)" by F.Rosenblatt
-1969ǯ
--"¥Ñ¡¼¥»¥×¥È¥í¥ó¤Î¸Â³¦¤ÎÄ輨" by M.Minsky and S.Papert
-1986ǯ
--³¬ÁØ·¿¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Î¸íº¹µÕÅÁÈÂË¡(back propagation)¤ÎÄê¼°²½~
by D.Rumelhart and G,Hinton, R,Williams
-¶áǯ
--Ǿ¤Îµ¡Ç½¤Î¼Â¸½²½¤ò¶¯¤¯°Õ¼±¤·¤¿¸¦µæ¤¬À¹¤ó¡£
**ÂåɽŪ¤Ê¿Í¸ý¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯ [#p445c56c]
-¥Õ¥£¡¼¥É¥Õ¥©¥ï¡¼¥É¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
-RNF¥Í¥Ã¥È¥ï¡¼¥¯
-¼«¸ÊÁÈ¿¥²½¼ÌÁü
-¥ê¥«¥ì¥ó¥È¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
-³ÎΨŪ¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
**¥Û¥Ã¥×¥Õ¥£¡¼¥ë¥É¥Í¥Ã¥È¥ï¡¼¥¯(Hopfield network) 1984ǯ by J.J.Hopfield [#q2b99cc7]
-½ä²ó¥»¡¼¥ë¥¹¥Þ¥óÌäÂê¤Ê¤É¤ÎºÇŬ²½ÌäÂê¤ò¤Ä¤¯¤¿¤á¤Ë¹Í¤¨½Ð¤µ¤ì¤¿¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¡£
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Ë¤è¤ë³Ø½¬ [#q174b2ca]
-¶µ»Õ¤¢¤ê³Ø½¬
--ÆþÎÏÃͤËÂФ·¤Æ½ÐÎÏÃͤò½Ð¤¹¤¿¤á¤Î¿¿ô¤Î¥¢¥ë¥´¥ê¥º¥à¤¬È÷¤ï¤Ã¤Æ¤¤¤ë¡£~
¤½¤ì¤é¤òƱ»þ¤Ë¼Â¸½¤¹¤ë¤è¤¦¤Ë¥â¥Ç¥ë¤ÎÃæ¤Î¥Ñ¥é¥á¡¼¥¿¤òÄ´À᤹¤ë¥â¥Ç¥ë¡£
--»Ï¤á¤«¤éÅú¤¨¤Îµá¤áÊý¤ò¶µ¤ï¤Ã¤Æ¤¤¤ë¡£
--ʬÎà¡¢µÚ¤ÓÈóÀþ·Á²óµ¢¤ÇÍøÍѤµ¤ì¤Æ¤¤¤ë¡£
-¶µ»Õ¤Ê¤·³Ø½¬
--ÆþÎÏ¥Ù¥¯¥È¥ë¤Î¤ßÍ¿¤¨¤é¤ì¤Æ¤¤¤ë¡£¤¢¤È¤Ï¤½¤³¤«¤é¼«Ê¬¤Ç³Ø½¬¤·¤ÆÅú¤¨¤òƳ¤¯¡£
--Åú¤¨¤¬ÍÑ°Õ¤µ¤ì¤Æ¤¤¤Ê¤¤¡£
--¾ðÊ󰵽̡¢´Ø¿ôÌ©Åٴؿô¥â¥Ç¥ë¤òÍѤ¤¤¿¿äÏÀ¤Ê¤É¤Ë±þÍÑ¡£
--ÂåɽÎã
---¥³¥Û¥Í¥ó¤Î¼«¸ÊÁÈ¿¥²½¥Þ¥Ã¥×
---¥¬¥¦¥¹¤Îº®¹ç¥â¥Ç¥ë
---k-means¥¯¥é¥¹¥¿¥ê¥ó¥°
*2nd "Neuron models and basic learning rules" on April 20 [#a51944c2]
>ÌÜɸ "¥Ë¥å¡¼¥í¥ó¤Î¿¶¤ëÉñ¤¤¤Î¿ô³ØŪ¥â¥Ç¥ë²½"
**¥Ë¥å¡¼¥í¥ó¤Î¹½Â¤ [#h5cfe425]
-¼ù¾õÆ͵¯(dendrite)
--ÆþÎϤò¼õ¤±ÉÕ¤±¤ë¤È¤³¤í
-ºÙ˦ÂÎ(soma)
--ÆþÎϤòÅý¹ç¤¹¤ë¤È¤³¤í(¤¿¤Ö¤ó¡¢²èÁü¤Ç¤¤¤¦³Ë¤ÎÊÕ¤ê)
-¼´º÷(axon)
--½ÐÎϤòÁ÷¤ë¤È¤³¤í
-¥·¥Ê¥×¥¹(synapse)
--¼´º÷¤«¤é¤Ç¤¿½ÐÎϤò¼ù¾õÆ͵¯¤ËÁ÷¤ë¤Þ¤Ç¤Î¥Ñ¥¤¥×¤À¤È»×¤¦¡£
**The McCulloch-Pitts neuron model [#cc3096d0]
>¿ÆþÎÏ£±½ÐÎϤʥ·¥¹¥Æ¥à¡£~
ÆþÎϤϤ½¤ì¤¾¤ì½Å¤ß¤ò»ý¤Á¡¢¤½¤Î¹ç·×¤¬¤·¤¤¤Ãͤè¤ê¤âÂ礤¤¤«¾®¤µ¤¤¤«¤Ç½ÐÎϤò·èÄꤹ¤ë¡£
#mimetex(o\; = \;f(\sum_{i=1}^n w_i x_i - T));
#mimetex(f(u)\; = \;\{\array{rcl$1\;\;if\;u\geq\;0 \\ 0 \;\; otherwise});
-ÆþÎÏ '''''x'''''&mimetex(\; = \;(x_1,x_2,\cdots,x_n));
-³ÆÆþÎϤΥ·¥Ê¥×¥¹¤Î¶¯¤µ(weight) '''''w'''''&mimetex(\; = \;(w_1,w_2,\cdots,w_n));
-½ÐÎÏ &mimetex(o);
-ïçÃÍ &mimetex(T);
**ÀìÌçÍѸì(º£¸å»È¤¦¤«¤Ï̤Äê) [#uaf2407d]
-effective input
--ÆþÎϤÎweight¤ÎÁíÏÂ
-activation function
--ºÇ½ªÅª¤Ê½ÐÎϤΤ¿¤á¤Î´Ø¿ô¡£McCulloch-Pitts¤Ë¤ª¤±¤ëStep function
** ¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Î¿Ê¤áÊý[#c51a6241]
+learning
-¾ðÊó¤ò¥Í¥Ã¥È¥ï¡¼¥¯Æâ¤ËÃÖ¤¯
--¶µ»Õ¤¢¤ê³Ø½¬¤È¶µ»Õ¤Ê¤·³Ø½¬
--¥ª¥ó¥é¥¤¥ó³Ø½¬¤È¥ª¥Õ¥é¥¤¥ó³Ø½¬
+recall
-¥Í¥Ã¥È¥ï¡¼¥¯¾å¤ËÃÖ¤«¤ì¤Æ¾ðÊó¤ò¸¡º÷¤¹¤ë
--¼«Æ°Àܳ
**³Ø½¬¥ë¡¼¥ë¡Ê°ìÈÌÊÔ¡Ë [#y4a17b85]
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-³Ø½¬Äê¿ô c
-³Ø½¬¿®¹æ r
-k¤Ë¤ª¤±¤ëweight¥Ù¥¯¥¿ &mimetex(\vec~W);
-ÆþÎÏ¥Ù¥¯¥¿ &mimetex(\vec~x);
***The Hebbian learning rule [#v12daeb6]
>¶µ»Õ¤Ê¤·³Ø½¬
-teaching signal¤¬¤Ê¤¤¤È¤³¤í¤«¤éȽÃÇ~
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;f(u));
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-&mimetex(f(u)\; = \;\{\array{1\;\;if\;u\succ\;0 \\ -1 \;\; if\;u\;\prec\;0});
-&mimetex(\vec~W^0);:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
***Perceptron learning rule [#i7761a73]
>¶µ»Õ¤¢¤ê³Ø½¬
-teaching signal¤¬¤¢¤ë¤È¤³¤í¤«¤éȽÃÇ
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;d\; - \;f(u));
-d : teaching signal
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-&mimetex(f(u)\; = \;\{\array{1\;\;if\;u\succ\;0 \\ -1 \;\; if\;u\;\prec\;0});
-&mimetex(\vec~W^0 );:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
***Delta learnign rule [#i0d88ba0]
>¶µ»Õ¤¢¤ê³Ø½¬
-teaching signal¤¬¤¢¤ë¤È¤³¤í¤«¤éȽÃÇ
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;[d\; - \;f(u)]f'(u));
-d : teaching signal
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-f(u) : ¥·¥°¥â¥¤¥É´Ø¿ô
-&mimetex(\vec~W^0 );:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
½ªÎ»¹Ô:
[[Á°Àî/¶µ°é]]
*1st "Introduction" on April 13 [#h4caa8cb]
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤È¤Ï¡© [#sd0bce1c]
-Ǿµ¡Ç½¤Ë¸«¤é¤ì¤ë¤¤¤¯¤Ä¤«¤ÎÆÃÀ¤ò·×»»µ¡¾å¤Î¥·¥ß¥å¥ì¡¼¥·¥ç¥ó¤Ë¤è¤Ã¤Æɽ¸½¤¹¤ë¤³¤È¤òÌܻؤ·¤¿¿ô³Ø¥â¥Ç¥ë¤Ç¤¢¤ë¡£~
by wikipedia
-Advantage
--ÊÂÎóʬ»¶½èÍý·¿
--¼«¸ÊÁÈ¿¥·¿
--¿®ÍêÀ¤¬¹â¤¤¡£<¡á¥Î¥¤¥º¤äÁû²»¤Ë¤¿¤¤¤·¤Æ¶¯¤¤¡£
--¸Î¾ã¤·¤Ë¤¯¤¤¡£
--Turning machine¤ÎÍýÏÀ¤ò¤â¤È¤Ëºî¤é¤ì¤Æ¤¤¤ë¡£
-Disadvantage
--®¤µ¡¢Àµ³Î¤µ¡¢»»½Ñ·×»»¤Î·«¤êÊÖ¤·¤Ë¼å¤¤¡£
--Ã챤ÎÍý²ò¡¢ºÆÍøÍѤ¬¶ì¼ê¡£
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤ÎÎò»Ë [#d3a2c870]
-1943ǯ
--"¤·¤¤¤ÃÍÁǻҥâ¥Ç¥ë" by W.McCulloch and W.pitts
-1949ǯ
--"The Organization of Behavior" by D,Hebb
--¥·¥Ê¥×¥¹¶¯²½Â§=>¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¤Î³Ø½¬ÊýË¡¤Î´ðÁäˡ£
-1962ǯ
--"¥Ñ¡¼¥»¥×¥È¥í¥ó(perceptron)" by F.Rosenblatt
-1969ǯ
--"¥Ñ¡¼¥»¥×¥È¥í¥ó¤Î¸Â³¦¤ÎÄ輨" by M.Minsky and S.Papert
-1986ǯ
--³¬ÁØ·¿¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Î¸íº¹µÕÅÁÈÂË¡(back propagation)¤ÎÄê¼°²½~
by D.Rumelhart and G,Hinton, R,Williams
-¶áǯ
--Ǿ¤Îµ¡Ç½¤Î¼Â¸½²½¤ò¶¯¤¯°Õ¼±¤·¤¿¸¦µæ¤¬À¹¤ó¡£
**ÂåɽŪ¤Ê¿Í¸ý¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯ [#p445c56c]
-¥Õ¥£¡¼¥É¥Õ¥©¥ï¡¼¥É¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
-RNF¥Í¥Ã¥È¥ï¡¼¥¯
-¼«¸ÊÁÈ¿¥²½¼ÌÁü
-¥ê¥«¥ì¥ó¥È¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
-³ÎΨŪ¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È
**¥Û¥Ã¥×¥Õ¥£¡¼¥ë¥É¥Í¥Ã¥È¥ï¡¼¥¯(Hopfield network) 1984ǯ by J.J.Hopfield [#q2b99cc7]
-½ä²ó¥»¡¼¥ë¥¹¥Þ¥óÌäÂê¤Ê¤É¤ÎºÇŬ²½ÌäÂê¤ò¤Ä¤¯¤¿¤á¤Ë¹Í¤¨½Ð¤µ¤ì¤¿¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¡£
**¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Ë¤è¤ë³Ø½¬ [#q174b2ca]
-¶µ»Õ¤¢¤ê³Ø½¬
--ÆþÎÏÃͤËÂФ·¤Æ½ÐÎÏÃͤò½Ð¤¹¤¿¤á¤Î¿¿ô¤Î¥¢¥ë¥´¥ê¥º¥à¤¬È÷¤ï¤Ã¤Æ¤¤¤ë¡£~
¤½¤ì¤é¤òƱ»þ¤Ë¼Â¸½¤¹¤ë¤è¤¦¤Ë¥â¥Ç¥ë¤ÎÃæ¤Î¥Ñ¥é¥á¡¼¥¿¤òÄ´À᤹¤ë¥â¥Ç¥ë¡£
--»Ï¤á¤«¤éÅú¤¨¤Îµá¤áÊý¤ò¶µ¤ï¤Ã¤Æ¤¤¤ë¡£
--ʬÎà¡¢µÚ¤ÓÈóÀþ·Á²óµ¢¤ÇÍøÍѤµ¤ì¤Æ¤¤¤ë¡£
-¶µ»Õ¤Ê¤·³Ø½¬
--ÆþÎÏ¥Ù¥¯¥È¥ë¤Î¤ßÍ¿¤¨¤é¤ì¤Æ¤¤¤ë¡£¤¢¤È¤Ï¤½¤³¤«¤é¼«Ê¬¤Ç³Ø½¬¤·¤ÆÅú¤¨¤òƳ¤¯¡£
--Åú¤¨¤¬ÍÑ°Õ¤µ¤ì¤Æ¤¤¤Ê¤¤¡£
--¾ðÊ󰵽̡¢´Ø¿ôÌ©Åٴؿô¥â¥Ç¥ë¤òÍѤ¤¤¿¿äÏÀ¤Ê¤É¤Ë±þÍÑ¡£
--ÂåɽÎã
---¥³¥Û¥Í¥ó¤Î¼«¸ÊÁÈ¿¥²½¥Þ¥Ã¥×
---¥¬¥¦¥¹¤Îº®¹ç¥â¥Ç¥ë
---k-means¥¯¥é¥¹¥¿¥ê¥ó¥°
*2nd "Neuron models and basic learning rules" on April 20 [#a51944c2]
>ÌÜɸ "¥Ë¥å¡¼¥í¥ó¤Î¿¶¤ëÉñ¤¤¤Î¿ô³ØŪ¥â¥Ç¥ë²½"
**¥Ë¥å¡¼¥í¥ó¤Î¹½Â¤ [#h5cfe425]
-¼ù¾õÆ͵¯(dendrite)
--ÆþÎϤò¼õ¤±ÉÕ¤±¤ë¤È¤³¤í
-ºÙ˦ÂÎ(soma)
--ÆþÎϤòÅý¹ç¤¹¤ë¤È¤³¤í(¤¿¤Ö¤ó¡¢²èÁü¤Ç¤¤¤¦³Ë¤ÎÊÕ¤ê)
-¼´º÷(axon)
--½ÐÎϤòÁ÷¤ë¤È¤³¤í
-¥·¥Ê¥×¥¹(synapse)
--¼´º÷¤«¤é¤Ç¤¿½ÐÎϤò¼ù¾õÆ͵¯¤ËÁ÷¤ë¤Þ¤Ç¤Î¥Ñ¥¤¥×¤À¤È»×¤¦¡£
**The McCulloch-Pitts neuron model [#cc3096d0]
>¿ÆþÎÏ£±½ÐÎϤʥ·¥¹¥Æ¥à¡£~
ÆþÎϤϤ½¤ì¤¾¤ì½Å¤ß¤ò»ý¤Á¡¢¤½¤Î¹ç·×¤¬¤·¤¤¤Ãͤè¤ê¤âÂ礤¤¤«¾®¤µ¤¤¤«¤Ç½ÐÎϤò·èÄꤹ¤ë¡£
#mimetex(o\; = \;f(\sum_{i=1}^n w_i x_i - T));
#mimetex(f(u)\; = \;\{\array{rcl$1\;\;if\;u\geq\;0 \\ 0 \;\; otherwise});
-ÆþÎÏ '''''x'''''&mimetex(\; = \;(x_1,x_2,\cdots,x_n));
-³ÆÆþÎϤΥ·¥Ê¥×¥¹¤Î¶¯¤µ(weight) '''''w'''''&mimetex(\; = \;(w_1,w_2,\cdots,w_n));
-½ÐÎÏ &mimetex(o);
-ïçÃÍ &mimetex(T);
**ÀìÌçÍѸì(º£¸å»È¤¦¤«¤Ï̤Äê) [#uaf2407d]
-effective input
--ÆþÎϤÎweight¤ÎÁíÏÂ
-activation function
--ºÇ½ªÅª¤Ê½ÐÎϤΤ¿¤á¤Î´Ø¿ô¡£McCulloch-Pitts¤Ë¤ª¤±¤ëStep function
** ¥Ë¥å¡¼¥é¥ë¥Í¥Ã¥È¥ï¡¼¥¯¤Î¿Ê¤áÊý[#c51a6241]
+learning
-¾ðÊó¤ò¥Í¥Ã¥È¥ï¡¼¥¯Æâ¤ËÃÖ¤¯
--¶µ»Õ¤¢¤ê³Ø½¬¤È¶µ»Õ¤Ê¤·³Ø½¬
--¥ª¥ó¥é¥¤¥ó³Ø½¬¤È¥ª¥Õ¥é¥¤¥ó³Ø½¬
+recall
-¥Í¥Ã¥È¥ï¡¼¥¯¾å¤ËÃÖ¤«¤ì¤Æ¾ðÊó¤ò¸¡º÷¤¹¤ë
--¼«Æ°Àܳ
**³Ø½¬¥ë¡¼¥ë¡Ê°ìÈÌÊÔ¡Ë [#y4a17b85]
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-³Ø½¬Äê¿ô c
-³Ø½¬¿®¹æ r
-k¤Ë¤ª¤±¤ëweight¥Ù¥¯¥¿ &mimetex(\vec~W);
-ÆþÎÏ¥Ù¥¯¥¿ &mimetex(\vec~x);
***The Hebbian learning rule [#v12daeb6]
>¶µ»Õ¤Ê¤·³Ø½¬
-teaching signal¤¬¤Ê¤¤¤È¤³¤í¤«¤éȽÃÇ~
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;f(u));
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-&mimetex(f(u)\; = \;\{\array{1\;\;if\;u\succ\;0 \\ -1 \;\; if\;u\;\prec\;0});
-&mimetex(\vec~W^0);:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
***Perceptron learning rule [#i7761a73]
>¶µ»Õ¤¢¤ê³Ø½¬
-teaching signal¤¬¤¢¤ë¤È¤³¤í¤«¤éȽÃÇ
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;d\; - \;f(u));
-d : teaching signal
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-&mimetex(f(u)\; = \;\{\array{1\;\;if\;u\succ\;0 \\ -1 \;\; if\;u\;\prec\;0});
-&mimetex(\vec~W^0 );:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
***Delta learnign rule [#i0d88ba0]
>¶µ»Õ¤¢¤ê³Ø½¬
-teaching signal¤¬¤¢¤ë¤È¤³¤í¤«¤éȽÃÇ
#mimetex(\vec~W^{k+1}\; = \;\vec~W^{k} \; + \;cr\vec~x);
-&mimetex(c\; = \;0\; or \;1);
-&mimetex(r\; = \;[d\; - \;f(u)]f'(u));
-d : teaching signal
-&mimetex(u\; = \; (\vec~W^k , \vec~x) );¡ÊÆâÀÑ¡Ë
-f(u) : ¥·¥°¥â¥¤¥É´Ø¿ô
-&mimetex(\vec~W^0 );:¥é¥ó¥À¥à¤ÇÍ¿¤¨¤é¤ì¤ë¡£
¥Ú¡¼¥¸Ì¾: