ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ـ ﻣﺼﻨﻮﻋﻰ« )Artificial Neural Networks (ANN ﺗﺄﻟﻴﻒ :ﻧﻴﻤﺎ ﻗﺎﺳﻢﻧﮋﺍﺩ ﻣﻘﺪﻡ ﺩﺍﻧﺸﮕﺎﻩ ﺁﺯﺍﺩ ﺍﺳﻼﻣﻲ ـ ﻭﺍﺣﺪ ﺍﻳﻠﺨﭽﻲ E-mail: Nima175@yahoo.com
ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒــﻰ ﻣﺼﻨﻮﻋــﻰ )Artificial ﻧﮕﺎﻫﻲ ﺑﻪ ﻛﻞ ﻣﻄﻠﺐ (Neural Networks=ANNﺩﺭ ﺩﻫﻪ 1940 ﺑﺮﺍﻯ ﺍﻭﻟﻴﻦ ﺑﺎﺭ ﻣﻌﺮﻓﻰ ﺷــﺪﻧﺪ ﻭ ﺗــﺎ ﺑﻪ ﺍﻣﺮﻭﺯ ﻧﻴﺰ ﻛﺎﺭﺑﺮﺩﻫﺎﻯ ﻓﺮﺍﻭﺍﻧﻰ ﺩﺍﺷــﺘﻪﺍﻧﺪ؛ ﻭﻟﻰ ﺑﺎ ﺍﻳﻦ ﻫﻤﻪ ﺩﺭ ﺣﻮﺯﻩ ﻣﺪﻳﺮﻳﺖ ﻭ ﻣﻬﻨﺪﺳﻰ ﺍﺯ ﻋﻠﻮﻡ ﻧﻮﭘﺎ ﺑﻪ ﺣﺴﺎﺏ ﻣﻰﺁﻳﺪ .ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻗﺎﺑﻠﻴﺖ ﺍﺳﺘﺨﺮﺍﺝ ﺍﻃﻼﻋﺎﺕ ﻣﻔﻴﺪ ﻭ ﻛﺎﺭﺑﺮﺩﻯ ﺍﺯ ﺩﺍﺩﻩﻫﺎﻯ ﺧﺎﻡ ﺭﺍ ﺩﺍﺭﺍ ﻣﻰﺑﺎﺷﻨﺪ .ﺍﻳﻦ ﺷﺒﻜﻪﻫﺎ ﺍﺑﺰﺍﺭﻯ ﻗﺪﺭﺗﻤﻨﺪ ﻫﺴﺘﻨﺪ ﻛﻪ ﻣﻰﺗﻮﺍﻧﻨﺪ ﺍﻟﮕﻮﻫﺎﻯ ﻣﺘﻔﺎﻭﺕ ﺭﺍ ﺷﻨﺎﺳﺎﻳﻰ ﻛﺮﺩﻩ ﻭ ﻳﺎ ﺍﺯ ﺍﻃﻼﻋﺎﺕ ﭘﻴﭽﻴﺪﻩ ﻭ ﮔﺎﻩ ﻣﺨﺪﻭﺵ ﻧﺘﺎﻳﺠﻰ ﻧﺰﺩﻳﻚ ﺑﻪ ﻭﺍﻗﻌﻴﺖ ﺑﻪ ﺩﺳﺖ ﺁﻭﺭﻧﺪ. ﻗــﺪﺭﺕ ﻭﺍﻗﻌﻰ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺗﻮﺍﻥ ﺁﻣﻮﺯﺵﭘﺬﻳﺮﻯ ﺁﻧﻬﺎﺳــﺖ .ﺑﻪ ﺍﻳﻦ ﻣﻔﻬﻮﻡ ﻛﻪ ﺍﻳﻦ ﺷــﺒﻜﻪﻫﺎ ﻗﺎﺩﺭﻧﺪ ﺗﺎ ﺍﺯ ﺭﻭﻯ ﺍﻟﮕﻮﻫﺎﻯ ﺁﻣﻮﺯﺷﻰ )ﻭﺭﻭﺩﻯﻫﺎ ﻭ ﺧﺮﻭﺟﻰﻫﺎﻯ ﻣﺘﻨﺎﺳــﺐ( ،ﺑﺎ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺍﻟﮕﻮﺭﻳﺘﻢﻫﺎﻯ ﻣﺨﺘﻠﻒ ﻳﺎﺩﮔﻴﺮﻯ ،ﺭﺍﺑﻄﻪ ﺑﻴﻦ ﻣﺘﻐﻴﺮﻫﺎ ﺭﺍ ﺷﻨﺎﺳﺎﻳﻰ ﻧﻤﺎﻳﻨﺪ. ﺩﺭ ﺍﻳﻦ ﻧﻮﺷﺘﺎﺭ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ ﺭﺍ ﺑﻪ ﻃﻮﺭ ﺧﻼﺻﻪ ﻣﻌﺮﻓﻰ ﻧﻤﻮﺩﻩﺍﻳﻢ ﻭ ﺑﻪ ﺍﻧﻮﺍﻉ ﺷــﺒﻜﻪﻫﺎ ،ﻳﺎﺩﮔﻴﺮﻯﺷﺎﻥ ،ﻣﺰﺍﻳﺎ ﻭ ﻣﺤﺪﻭﺩﻳﺖﻫﺎ ﻭ ﻧﻴﺰ ﭼﮕﻮﻧﮕﻰ ﻋﻤﻠﻜﺮﺩ ﺁﻧﻬﺎ ﭘﺮﺩﺍﺧﺘﻪ ﺷﺪﻩ ﺍﺳﺖ.
ﻣﻘﺪﻣﻪ
ﺩﺭ ﺳــﺎﻟﻴﺎﻥ ﺍﺧﻴﺮ ﺷــﺎﻫﺪ ﺣﺮﻛﺘﻲ ﻣﺴــﺘﻤﺮ ﺍﺯ ﺗﺤﻘﻴﻘــﺎﺕ ﺻﺮﻓﺎً ﻧﻈﺮﻯ ﺑــﻪ ﺗﺤﻘﻴﻘﺎﺕ ﻛﺎﺭﺑﺮﺩﻱ ﺑﻮﺩﻩﺍﻳﻢ ،ﺑﺨﺼﻮﺹ ﺩﺭ ﺯﻣﻴﻨﻪ ﭘﺮﺩﺍﺯﺵ ﺍﻃﻼﻋﺎﺕ ،ﺑﺮﺍﻱ ﻣﺴﺎﺋﻠﻲ ﻛﻪ ﺑﺮﺍﻱ ﺁﻧﻬﺎ ﺭﺍﻩﺣﻠﻲ ﻣﻮﺟﻮﺩ ﻧﻴﺴﺖ ﻭ ﻳﺎ ﺑﻪ ﺭﺍﺣﺘﻲ ﻗﺎﺑﻞ ﺣﻞ ﻧﻴﺴﺘﻨﺪ .ﺑﺎ ﻋﻨﺎﻳﺖ ﺑﻪ ﺍﻳﻦ ﺍﻣﺮ ،ﻋﻼﻗﻪﺍﻱ ﻓﺰﺍﻳﻨﺪﻩ ﺩﺭ ﺗﻮﺳﻌﻪ ﻧﻈﺮﻯ ﺳﻴﺴﺘﻢﻫﺎﻱ ﺩﻳﻨﺎﻣﻴﻜﻲ ﻫﻮﺷﻤﻨﺪ ﻣﺪﻝ ﺁﺯﺍﺩ ﺍﻳﺠﺎﺩ ﺷﺪﻩ ﺍﺳﺖ ﻛﻪ ﻣﺒﺘﻨﻲ ﺍﺳﺖ ﺑﺮ ﭘﺮﺩﺍﺯﺵ ﺩﺍﺩﻩﻫﺎﻱ ﺗﺠﺮﺑﻲ .ANNﺍﻳﻦ ﺳﻴﺴﺘﻢﻫﺎ ﺩﺍﻧﺶ ﻳﺎ ﻗﺎﻧﻮﻥ ﻧﻬﻔﺘﻪ ﺩﺭ ﻭﺭﺍﻱ ﺩﺍﺩﻩﻫﺎ ﺭﺍ ﺑﻪ ﺳــﺎﺧﺘﺎﺭ ﺷــﺒﻜﻪ ﻣﻨﺘﻘﻞ ﻣﻲﻛﻨﻨﺪ .ﺑﻪ ﻫﻤﻴﻦ ﺩﻟﻴﻞ ﺑﻪ ﺍﻳﻦ ﺳﻴﺴﺘﻢﻫﺎ ،ﻫﻮﺷﻤﻨﺪ ﮔﻔﺘﻪ ﻣﻲﺷﻮﺩ؛ ﺯﻳﺮﺍ ﺑﺮ ﺍﺳﺎﺱ ﻣﺤﺎﺳﺒﺎﺕ ﺭﻭﻱ ﺩﺍﺩﻩﻫﺎﻱ ﻋﺪﺩﻱ ﻳﺎ ﻣﺜﺎﻝﻫﺎ ،ﻗﻮﺍﻧﻴﻦ ﻛﻠﻲ ﺭﺍ ﻓﺮﺍ ﻣﻲﮔﻴﺮﻧﺪ ﻛﻪ ﺍﻳﻦ ﻋﻤﻞ ﻧﺸــﺄﺕ ﮔﺮﻓﺘﻪ ﺍﺯ ﺳﻴﺴﺘﻢ ﻋﻤﻠﻜﺮﺩ ﻣﻐﺰ ﻭ ﺍﻋﺼﺎﺏ ﻣﻐﺰﻯ ﺍﻧﺴﺎﻥ ﺍﺳﺖ. ﻣﻐﺰ ﺍﻧﺴﺎﻥ ﻣﻴﻠﻴﻮﻥﻫﺎ ﺳﻠﻮﻝ ﻋﺼﺒﻰ ﺩﺍﺭﺩ ﻛﻪ ﻭﻇﻴﻔﻪ ﺫﺧﻴﺮﻩ ﻛﺮﺩﻥ ﻭ ﭘﺮﺩﺍﺯﺵ ﺍﻃﻼﻋﺎﺕ ﺭﺍ ﺑﻪ ﻋﻬﺪﻩ ﺩﺍﺭﻧﺪ .ﻳﻜﻰ ﺍﺯ ﺳــﻠﻮﻝﻫﺎﻯ ﻋﺼﺒﻰ ،ﻣﻌﺮﻭﻑ ﺑﻪ ﻧﺮﻭﻥ 1ﺍﺳــﺖ ﻛﻪ ﻓﻘﻂ ٪10 ﺣﺠﻢ ﻣﻐﺰ ﺭﺍ ﺗﺸــﻜﻴﻞ ﻣﻰﺩﻫﺪ .ﺳــﻠﻮﻝﻫﺎﻯ ﻋﺼﺒــﻰ ﻗﺎﺩﺭﻧﺪ ﺗﺎ ﺑﺎ ﺍﺗﺼــﺎﻝ ﺑﻪ ﻳﻜﺪﻳﮕﺮ ﺗﺸــﻜﻴﻞ ﺷﺒﻜﻪﻫﺎﻯ ﻋﻈﻴﻢ ﺑﺪﻫﻨﺪ .ﮔﻔﺘﻪ ﻣﻰﺷﻮﺩ ﻛﻪ ﻫﺮ ﻧﺮﻭﻥ ﻣﻰﺗﻮﺍﻧﺪ ﺑﻪ ﻫﺰﺍﺭ ﺗﺎ ﺩﻩ ﻫﺰﺍﺭ ﻧﺮﻭﻥ ﺩﻳﮕﺮ ﺍﺗﺼﺎﻝ ﻳﺎﺑﺪ ﻭ ﮔﻤﺎﻥ ﻣﻰﺭﻭﺩ ﻣﻐﺰ ﺍﻧﺴﺎﻥ ﺍﺯ ﺗﻌﺪﺍﺩ 1011ﻧﺮﻭﻥ ﺗﺸﻜﻴﻞ 1 . Neural
16 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127
ﺷﺪﻩ ﺍﺳﺖ. ﺩﺭ ﺗﺸــﺮﻳﺢ ﺷــﺒﻜﻪ ﻋﺼﺒﻰ ﺩﺍﺭﻳﻢ :ﻋﺼﺐ ﻳﻚ ﻭﺍﺣﺪ ﺳــﻠﻮﻟﻲ ﺍﺳﺎﺳــﻲ ﺍﺯ ﺳﻴﺴﺘﻢ ﻣﻐﺰ ﺍﺳــﺖ .ﻋﺼﺐ ﻳﻚ ﻋﻨﺼﺮ ﭘﺮﺩﺍﺯﺷﮕﺮ ﺳﺎﺩﻩ ﺍﺳﺖ ﻛﻪ ﺍﺯ ﻃﺮﻳﻖ ﻣﺴﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺑﻪ ﻧﺎﻡ ﺩﻧﺪﺭﻳﺖﻫﺎ 2ﻋﻼﺋﻤﻲ ﺭﺍ ﺍﺯ ﺳــﺎﻳﺮ ﺍﻋﺼﺎﺏ ﺩﺭﻳﺎﻓﺖ ﻛﺮﺩﻩ ﻭ ﺁﻧﻬﺎ ﺭﺍ ﺑﺎ ﻫﻢ ﺗﺮﻛﻴﺐ ﻣﻲﻛﻨﺪ. ﺍﮔﺮ ﺍﻳﻦ ﻋﻼﻣﺖ ﻭﺭﻭﺩﻱ ﻣﺮﻛﺐ ﺑﻪ ﺍﻧﺪﺍﺯﻩ ﻛﺎﻓﻲ ﻗﻮﻱ ﺑﺎﺷــﺪ ،ﻋﺼﺐ ﺑﻪ ﺍﺻﻄﻼﺡ ﺷــﻠﻴﻚ ﻣﻲﻛﻨﺪ ﻭ ﻳﻚ ﻋﻼﻣﺖ ﺧﺮﻭﺟﻲ ﺭﺍ ﺩﺭ ﻃﻮﻝ ﺁﻛﺴــﻮﻧﻲ 3ﻛﻪ ﺩﺭ ﺩﻧﺪﺭﻳﺖ ﺳــﺎﻳﺮ ﺍﻋﺼﺎﺏ ﻣﺘﺼﻞ ﺍﺳــﺖ ﺍﺭﺳــﺎﻝ ﻣﻲﻛﻨﺪ .ﺷﻜﻞ ) (1ﻃﺮﺣﻲ ﺍﺯ ﻗﺴــﻤﺖﻫﺎﻱ ﻣﺨﺘﻠﻒ ﻳﻚ ﻋﺼﺐ ﺍﺳــﺖ .ﻫﺮ ﻋﻼﻣﺘﻲ ﻛﻪ ﺩﺭ ﻃﻮﻝ ﺩﻧﺪﺭﻳﺖ ﻳﻚ ﻋﺼﺐ ﻓﺮﺳــﺘﺎﺩﻩ ﻣﻲﺷﻮﺩ ،ﺍﺯ ﺳﻴﻨﺎﭘﺲ4 ﻳﺎ ﺍﺗﺼﺎﻝ ﺳﻴﻨﺎﭘﺴــﻲ ﻋﺒﻮﺭ ﻣﻲﻛﻨﺪ .ﺍﻳﻦ ﺍﺗﺼﺎﻝ ،ﻳﻚ ﺷﻜﺎﻑ ﺑﺴﻴﺎﺭ ﻛﻮﭼﻚ ﺩﺭ ﺩﻧﺪﺭﻳﺖ ﺍﺳﺖ ﻛﻪ ﺑﺎ ﻧﻮﻋﻲ ﻣﺎﻳﻊ ﻫﺎﺩﻱ ﻋﺼﺒﻲ 5ﭘﺮ ﺷﺪﻩ ﺍﺳﺖ. ﺍﻳــﻦ ﻣﺎﻳــﻊ ﻫﺎﺩﻱ ﻋﺼﺒــﻲ ﻋﻼﺋﻤﻲ ﺍﻟﻜﺘﺮﻳﻜﻲ ﺗﻮﻟﻴﺪ ﻣﻲﻛﻨﺪ ﻭ ﺑﻪ ﺳــﻮﻱ ﻫﺴــﺘﻪ ﻳﺎ ﺳــﻮﻣﺎﻱ 6ﻋﺼﺐ ﻣﻲﻓﺮﺳــﺘﺪ .ﺗﻨﻈﻴﻢ ﻣﻴﺰﺍﻥ ﻣﻘﺎﻭﻣﺖ ﻳﺎ ﻫﺪﺍﻳﺖ ﺍﻟﻜﺘﺮﻳﻜﻲ ﺍﻳﻦ ﻓﺎﺻﻠﻪ ﺳﻴﻨﺎﭘﺴــﻲ ،ﻳﻚ ﻓﺮﺍﻳﻨﺪ ﺑﺴﻴﺎﺭ ﻣﻬﻢ ﺍﺳﺖ .ﺑﻪ ﻋﻼﻭﻩ ﻫﻤﻴﻦ ﺗﻨﻈﻴﻢ ﻫﺪﺍﻳﺖ ﺑﺎﻋﺚ ﺣﻔﻆ ﻛﺮﺩﻥ ﻣﻄﺎﻟﺐ ﻭ ﻳﺎﺩﮔﻴﺮﻱ ﻣﻲﺷﻮﺩ. ﻭﻗﺘﻲ ﻗﺪﺭﺕ ﺳﻴﻨﺎﭘﺴــﻲ ﺍﻋﺼﺎﺏ ﺗﻨﻈﻴﻢ ﺷــﻮﺩ ،ﻣﻐﺰ ﻳﺎﺩﮔﻴﺮﻱ ﻣﻲﻧﻤﺎﻳﺪ ﻭ ﺍﻃﻼﻋﺎﺕ ﺭﺍ ﺫﺧﻴﺮﻩ ﻣﻲﻛﻨﺪ.
ﻳﻚ ﺷﺒﻜﻪ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻲ ﺭﺍ ﻣﻲﺗﻮﺍﻥ ﺑﻪ ﺻﻮﺭﺕ ﺯﻳﺮ ﺗﻌﺮﻳﻒ ﻛﺮﺩ: »ﻳﻚ ﺳﻴﺴــﺘﻢ ﭘﺮﺩﺍﺯﺵ ﺩﺍﺩﻩﻫﺎ ﻛﻪ ﺍﺯ ﺗﻌﺪﺍﺩ ﺯﻳﺎﺩﻱ ﻋﻨﺎﺻﺮ ﭘﺮﺩﺍﺯﺷــﮕﺮ ﺳﺎﺩﻩ ﻭ ﺑﺴﻴﺎﺭ ﻣﺮﺗﺒﻂ ﺑﺎ ﻫﻢ )ﻳﻌﻨﻲ ﻫﻤﺎﻥ ﺍﻋﺼﺎﺏ ﻣﺼﻨﻮﻋﻲ( ﺗﺸــﻜﻴﻞ ﺷﺪﻩ ﺍﺳﺖ ﻭ ﺩﺭ ﺳﺎﺧﺘﺎﺭ ﺁﻥ ﺍﺯ ﭘﻮﺳﺘﻪ ﺩﻣﺎﻏﻲ ﻣﻐﺰ ﺍﻟﻬﺎﻡ ﮔﺮﻓﺘﻪ ﺷﺪﻩ ﺍﺳﺖ«. ﺩﺭ ﻳﻚ ﺗﻌﺮﻳﻒ ﺳــﻨﺘﻰ ،ﻫﺎﻳﻜﻴﻦ ﻣﻰﮔﻮﻳﺪ :ﺷــﺒﻜﻪ ﻋﺼﺒﻰ ﻋﺒﺎﺭﺕ ﺍﺳﺖ ﺍﺯ ﻣﺠﻤﻮﻋﻪﺍﻯ ﻋﻈﻴﻢ ﺍﺯ ﭘﺮﺩﺍﺯﺷــﮕﺮﻫﺎﻯ ﻣﻮﺍﺯﻯ ﻛﻪ ﺍﺳﺘﻌﺪﺍﺩ ﺫﺍﺗﻰ ﺑﺮﺍﻯ ﺫﺧﻴﺮﻩ ﺍﻃﻼﻋﺎﺕ ﺗﺠﺮﺑﻰ ﻭ ﺑﻪ ﻛﺎﺭﮔﻴﺮﻯ ﺁﻥ ﺩﺍﺭﻧﺪ ﻭ ﺍﻳﻦ ﺷــﺒﻜﻪ ﺩﺳــﺘﻜﻢ ﺍﺯ ﺩﻭ ﺑﺎﺑﺖ ﺷﺒﻴﻪ ﻣﻐﺰ ﺍﺳﺖ1 :ـ ﻣﺮﺣﻠﻪﺍﻯ ﻣﻮﺳــﻮﻡ ﺑــﻪ ﻳﺎﺩﮔﻴــﺮﻯ ﺩﺍﺭﺩ2 .ـ ﻭﺯﻥﻫﺎﻯ ﺳﻴﻨﺎﭘﺴــﻰ ﺟﻬﺖ ﺫﺧﻴﺮﻩ ﺩﺍﻧــﺶ ﺑﻪ ﻛﺎﺭ ﻣﻰﺭﻭﻧﺪ.
ﺷﻜﻞ :2ﻃﺮﺡ ﺷﻤﺎﺗﻴﻚ ﻋﺼﺐ ﻣﺼﻨﻮﻋﻰ n
* ﺟﻤﻊ ﻭﺭﻭﺩﻯﻫﺎﻯ ﻭﺯﻥﺩﺍﺭ ﺑﺮﺍﺑﺮ ﺍﺳﺖ ﺑﺎ:
I j = ∑ wij xi i =1
) yi = f ( I j
ﺷﻜﻞ :1ﻗﺴﻤﺖﻫﺎﻱ ﻣﺨﺘﻠﻒ ﻳﻚ ﻋﺼﺐ
ﺍﻋﺼﺎﺏ ﻣﺼﻨﻮﻋﻲ ﻳﻚ ﻋﺼﺐ ﻣﺼﻨﻮﻋﻲ ﻣﺪﻟﻲ ﺍﺳــﺖ ﻛﻪ ﺍﺟﺰﺍﻱ ﺁﻥ ﺷﺒﺎﻫﺖ ﻣﺴﺘﻘﻴﻤﻲ ﺑﻪ ﺍﺟﺰﺍﻯ ﻭﺍﻗﻌﻲ ﺩﺍﺭﻧﺪ .ﺷﻜﻞ ) (2ﻧﻤﺎﻳﻲ ﺍﺯ ﻳﻚ ﻋﺼﺐ ﻣﺼﻨﻮﻋﻲ ﺍﺳﺖ .ﻋﻼﺋﻢ ﻭﺭﻭﺩﻱ ﺑﺎ ,X2 ,X1 , X0 … Xn ,ﻣﺸــﺨﺺ ﺷــﺪﻩﺍﻧﺪ .ﺍﻳﻦ ﻋﻼﺋﻢ ،ﻣﺘﻐﻴﺮﻫﺎﻳﻲ ﭘﻴﻮﺳﺘﻪ ﻫﺴﺘﻨﺪ ﻭ ﻧﻪ ﭘﺎﻟﺲﻫﺎﻱ ﺍﻟﻜﺘﺮﻳﻜــﻲ ﻛﻪ ﺩﺭ ﻣﻐﺰ ﺭﺥ ﻣﻲﺩﻫﻨﺪ .ﻫﺮ ﻳــﻚ ﺍﺯ ﺍﻳﻦ ﻣﻘﺎﺩﻳﺮ ﻭﺭﻭﺩﻱ ﺗﺤﺖ ﺗﺄﺛﻴﺮ ﻭﺯﻧﻲ )ﻛﻪ ﮔﺎﻩ ﻭﺯﻥ ﺳﻴﻨﺎﭘﺴﻲ ﻧﺎﻣﻴﺪﻩ ﻣﻲﺷﻮﺩ( ﻗﺮﺍﺭ ﻣﻲﮔﻴﺮﻧﺪ ﻛﻪ ﺗﺎﺑﻊ ﺍﻳﻦ ﻭﺯﻥ ﺷﺒﻴﻪ ﺍﺗﺼﺎﻝ ﺳﻴﻨﺎﭘﺴــﻲ ﺩﺭ ﻳﻚ ﻋﺼﺐ ﻭﺍﻗﻌﻲ ﺍﺳــﺖ .ﺑﺴــﺘﻪ ﺑﻪ ﻣﻴﺰﺍﻥ ﻫﺪﺍﻳﺖ ﻳﺎ ﻣﻘﺎﻭﻣﺖ ﺟﺮﻳﺎﻥ ﻋﻼﺋﻢ ﺍﻟﻜﺘﺮﻳﻜﻲ ،ﺍﻳﻦ ﻭﺯﻥﻫﺎ ﻣﻲﺗﻮﺍﻧﻨﺪ ﻣﺜﺒﺖ ﻳﺎ ﻣﻨﻔﻲ ﺑﺎﺷــﻨﺪ .ﺍﻳﻦ ﻋﻨﺎﺻﺮ ﭘﺮﺩﺍﺯﺷــﮕﺮ ﺍﺯ ﺩﻭ ﻗﺴﻤﺖ ﺗﺸﻜﻴﻞ ﺷﺪﻩﺍﻧﺪ؛ ﻗﺴﻤﺖ ﺍﻭﻝ ﻭﺭﻭﺩﻱﻫﺎﻱ ﻭﺯﻥﺩﺍﺭ ﺭﺍ ﺑﺎ ﻫﻢ ﺟﻤﻊ ﻣﻲﺯﻧﺪ ﻭﻛﻤﻴﺘﻲ ﺑﻪ ﻧﺎﻡ Iﺑﻪ ﺩﺳــﺖ ﻣﻲﺁﻭﺭﺩ؛ ﻗﺴــﻤﺖ ﺩﻭﻡ ﻳﻚ ﺻﺎﻓﻲ ﻏﻴﺮ ﺧﻄﻲ ﺍﺳــﺖ ﻛﻪ ﻣﻌﻤﻮﻻً ﺗﺎﺑﻊ ﻓﻌﺎﻟﺴــﺎﺯﻱ 7ﻧﺎﻣﻴﺪﻩ ﻣﻲﺷــﻮﺩ ﻭ ﺍﺯ ﻃﺮﻳــﻖ ﺁﻥ ﺟﺮﻳﺎﻥﻫﺎﻱ ﻋﻼﺋﻢ ﻭﺭﻭﺩﻱ ﺗﺮﻛﻴﺐ ﻣﻲﺷﻮﻧﺪ. 2 .Dendrites 3 .Axon 4 .Synapse 5 .Neurotransmitter Fluid 6 .Soma 7 .Activation Function
* ﺗﺎﺑﻊ ﻓﻌﺎﻟﺴﺎﺯﻯ ﺑﺮﺍﺑﺮ ﺍﺳﺖ ﺑﺎ: ﺍﻳﻦ ﻋﻨﺎﺻﺮ ﭘﺮﺩﺍﺯﺷــﮕﺮ ﻣﻌﻤﻮﻻً ﺩﺭ ﻻﻳﻪﻫﺎ ﻳﺎ ﺻﻔﺤﺎﺕ ﻣﻨﻈﻤﻰ ﻗﺮﺍﺭ ﮔﺮﻓﺘﻪﺍﻧﺪ؛ ﺑﻪ ﻃﻮﺭﻯ ﻛــﻪ ﺑﻴﻦ ﻻﻳﻪﻫﺎ ﺍﺭﺗﺒﺎﻃــﺎﺕ ﻛﺎﻣﻞ ﻭ ﻳﺎ ﺗﺼﺎﺩﻓــﻰ ﻭﺟﻮﺩ ﺩﺍﺭﺩ .ﻻﻳــﻪ ﻭﺭﻭﺩﻯ ﺑﻪ ﻋﻨﻮﺍﻥ ﭘﺮﺩﺍﺯﺷﮕﺮﻯ ﺍﺳﺖ ﻛﻪ ﭘﺲ ﺍﺯ ﭘﺮﺩﺍﺯﺵ ﺩﺍﺩﻩﻫﺎﻯ ﻭﺭﻭﺩﻯ ﺁﻧﻬﺎ ﺭﺍ ﺑﻪ ﺷﺒﻜﻪ ﺍﺭﺍﺋﻪ ﻣﻰﺩﻫﺪ. ﻻﻳﻪ ﺑﺎﻻﻳﻰ ،ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﺍﺳــﺖ ﻛﻪ ﺧﺮﻭﺟﻰ ﺷﺒﻜﻪ ﺭﺍ ﺩﺭ ﭘﺎﺳﺦ ﺑﻪ ﻳﻚ ﻭﺭﻭﺩﻯ ﻧﺸﺎﻥ ﻣﻰﺩﻫﺪ ﻭ ﺳﺎﻳﺮ ﻻﻳﻪﻫﺎﻯ ﺑﻴﻦ ﺍﻳﻦ ﺩﻭ ﻻﻳﻪ ،ﻻﻳﻪﻫﺎﻯ ﻣﻴﺎﻧﻰ ﻳﺎ ﭘﻨﻬﺎﻥ ﻧﺎﻣﻴﺪﻩ ﻣﻰﺷﻮﻧﺪ. ﺯﻣﺎﻧﻰ ﻛﻪ ﻣﻰﮔﻮﻳﻴﻢ ﺷــﺒﻜﻪ ﺍﺯ nﻻﻳﻪ ﺗﺸﻜﻴﻞ ﺷﺪﻩ ﺍﺳــﺖ ،ﻣﺎ ﺗﻨﻬﺎ ﻻﻳﻪﻫﺎﻯ ﻣﻴﺎﻧﻰ ﻭ ﻻﻳﻪ ﺧﺎﺭﺟﻰ ﺭﺍ ﻣﻰﺷــﻤﺎﺭﻳﻢ ﻭ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﺷــﻤﺎﺭﺵ ﻧﻤﻰﺷــﻮﺩ؛ ﭼﺮﺍ ﻛﻪ ﺍﻳﻦ ﻧﺮﻭﻥﻫﺎ ﻣﺤﺎﺳــﺒﻪﺍﻯ ﺭﺍ ﺍﻧﺠﺎﻡ ﻧﻤﻰﺩﻫﻨﺪ .ﺑﻨﺎﺑﺮﺍﻳﻦ ﺷــﺒﻜﻪ ﺗﻚﻻﻳﻪ ﺷﺎﻣﻞ ﺷﺒﻜﻪﺍﻯ ﺑﺎ ﺗﻨﻬﺎ ﻳﻚ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﻣﻰﺑﺎﺷﺪ.
ﺷﻜﻞ :3ﺳﺎﺧﺘﺎﺭ ﺷﺒﻜﻪ ﻋﺼﺒﻰ
17 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127
ﭘﻴﺸــﺮﻓﺖ ﻋﻠﻢ ﻋﺼﺐﺷﻨﺎﺳــﻰ ﺑﻪ ﻣﺤﻘﻘﺎﻥ ﺍﻳﻦ ﻓﺮﺻﺖ ﺭﺍ ﺩﺍﺩﻩ ﻛــﻪ ﻣﺪﻝﻫﺎﻯ ﺭﻳﺎﺿﻰ ﺍﻋﺼﺎﺏ ﺭﺍ ﺑﺮﺍﻯ ﺷﻨﺎﺳﺎﻳﻰ ﺭﻓﺘﺎﺭﺷﺎﻥ ﺑﻪ ﺩﺳﺖ ﺁﻭﺭﻧﺪ .ﺍﻳﻦ ﺍﻳﺪﻩ ﺑﻪ ﺩﻫﻪ 1940ﻣﻴﻼﺩﻯ ﺑﺮﻣﻰﮔﺮﺩﺩ )ﺳــﺎﻝ ،(1943ﺯﻣﺎﻧﻰ ﻛﻪ ﻳﻚ ﻣﺪﻝ ﺍﻧﺘﺰﺍﻋﻰ ﺗﻮﺳﻂ ﻣﻚﻛﺎﻟﻨﺪ ﻭ ﭘﻴﺘﺲ ﺩﺭ ﺳــﺎﻝ 1943ﻣﻌﺮﻓﻰ ﺷــﺪ .ﺭﻭﺯﻧﺒﺎﻟﺖ ) (1958ﺍﻟﮕﻮﺭﻳﺘﻢ ﻳﺎﺩﮔﻴﺮﻯ ﭘﺮﺳﭙﺘﺮﻭﻥ ﺭﺍ ﺍﺭﺍﺋﻪ ﺩﺍﺩ .ﺩﺭ ﺍﻳﻦ ﺯﻣﺎﻥ ،ﻭﻳﺪﺭﻭ ﻭ ﻫﻮﻑ ﻳﻚ ﺗﻐﻴﻴﺮ ﻣﻬﻤﻰ ﺩﺭ ﺍﻟﮕﻮﺭﻳﺘﻢ ﻳﺎﺩﮔﻴﺮﻯ ﭘﺮﺳــﭙﺘﺮﻭﻥ ﺍﻳﺠﺎﺩ ﻧﻤﻮﺩﻧﺪ ﻛﻪ ﺑﻪ ﻧﺎﻡ ﻗﺎﻧﻮﻥ ﻭﻳﺪﺭﻭ ـ ﻫﻮﻑ ﺷــﻨﺎﺧﺘﻪ ﺷــﺪ .ﺳﭙﺲ ﻣﻴﻨﺴﻜﻰ ﻭ ﭘﺎﭘﺮﺕ ﻣﺤﺪﻭﺩﻳﺖﻫﺎﻯ ﻣﺪﻝ ﺷﺒﻜﻪ ﻋﺼﺒﻰ ﻳﻚ ﻻﻳﻪ ﺭﺍ ﻧﺸﺎﻥ ﺩﺍﺩﻧﺪ .ﻛﻮﻫﻨﻦ ) (1977ﻣﺪﻝﻫﺎﻯ ﺣﺎﻓﻈﻪ ﺍﻧﺠﻤﻨﻰ ﺭﺍ ﺗﻮﺳﻌﻪ ﺩﺍﺩ. ﺩﺭ ﺍﻭﺍﺳــﻂ ﺩﻫﻪ 1980ﺍﻟﮕﻮﺭﻳﺘﻢ ﻳﺎﺩﮔﻴﺮﻯ ﭘﺲﺍﻧﺘﺸﺎﺭ ﺗﻮﺳــﻂ ﺭﻭﻣﻠﻬﺎﺭﺕ ،ﻫﻴﻨﺘﻮﻥ ﻭ ﻭﻳﻠﻴﺎﻡ ) (1986ﺍﺭﺍﺋﻪ ﺷــﺪ ﻛﻪ ﺭﺍﻩﺣﻞ ﻗﺪﺭﺗﻤﻨﺪﻯ ﺑﺮﺍﻯ ﻳﺎﺩﮔﻴﺮﻯ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﭼﻨﺪﻻﻳﻪ ﻣﻰﺑﺎﺷﺪ.
ﺍﺟﺰﺍﻯ ﻳﻚ ﺷﺒﻜﻪ ﻋﺼﺒﻰ ﻋﺒﺎﺭﺗﻨﺪ ﺍﺯ:
ﻭﺭﻭﺩﻯﻫﺎ :ﻭﺭﻭﺩﻯﻫﺎ ﻣﻰﺗﻮﺍﻧﻨﺪ ﺧﺮﻭﺟﻰ ﺳﺎﻳﺮ ﻻﻳﻪﻫﺎ ﺑﻮﺩﻩ ﻭ ﻳﺎ ﺁﻧﻜﻪ ﺑﻪ ﺣﺎﻟﺖ ﺧﺎﻡ ﺩﺭ ﺍﻭﻟﻴﻦ ﻻﻳﻪ ﻭ ﺑﺪﻳﻦ ﺻﻮﺭﺕﻫﺎ ﺑﺎﺷﻨﺪ :ﺩﺍﺩﻩﻫﺎﻯ ﻋﺪﺩﻯ ،ﻣﺘﻮﻥ ﺍﺩﺑﻰ ،ﻓﻨﻰ ،ﺗﺼﻮﻳﺮ ﻭ ﻳﺎ ﺷﻜﻞ. ﻭﺯﻥﻫﺎ :ﻣﻴﺰﺍﻥ ﺗﺄﺛﻴﺮ ﻭﺭﻭﺩﻯ xiﺑﺮ ﺧﺮﻭﺟﻰ yﺗﻮﺳﻂ ﻭﺯﻥ ﺍﻧﺪﺍﺯﻩﮔﻴﺮﻯ ﻣﻰﺷﻮﺩ. ﺗﺎﺑﻊ ﺟﻤﻊ :ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﺗﻚﻧﺮﻭﻧﻰ ،ﺗﺎﺑﻊ ﺟﻤﻊ ﺩﺭ ﻭﺍﻗﻊ ﺧﺮﻭﺟﻰ ﻣﺴﺌﻠﻪ ﺭﺍ ﺗﺎ ﺣﺪﻭﺩﻯ ﻣﺸﺨﺺ ﻣﻰﻛﻨﺪ ﻭ ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﭼﻨﺪﻧﺮﻭﻧﻰ ﻧﻴﺰ ﺗﺎﺑﻊ ﺟﻤﻊ ﻣﻴﺰﺍﻥ ﺳﻄﺢ ﻓﻌﺎﻟﻴﺖ ﻧﺮﻭﻥ jﺩﺭ ﻻﻳﻪﻫﺎﻯ ﺩﺭﻭﻧﻰ ﺭﺍ ﻣﺸﺨﺺ ﻣﻰﺳﺎﺯﺩ. ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ )ﺗﺎﺑﻊ ﻓﻌﺎﻟﺴــﺎﺯﻯ( :ﺑﺪﻳﻬﻰ ﺍﺳــﺖ ﻛﻪ ﺗﺎﺑﻊ ﺟﻤﻊ ﭘﺎﺳﺦ ﻣﻮﺭﺩ ﺍﻧﺘﻈﺎﺭ ﺷﺒﻜﻪ ﻧﻴﺴــﺖ .ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ ﻋﻀﻮﻯ ﺿﺮﻭﺭﻯ ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺤﺴﻮﺏ ﻣﻰﮔﺮﺩﺩ .ﺍﻧﻮﺍﻉ ﻭ ﺍﻗﺴﺎﻡ ﻣﺘﻔﺎﻭﺗﻰ ﺍﺯ ﺗﻮﺍﺑﻊ ﺗﺒﺪﻳﻞ ﻭﺟﻮﺩ ﺩﺍﺭﺩ ﻛﻪ ﺑﻨﺎ ﺑﻪ ﺍﻫﻤﻴﺖ ﻭ ﻧﻮﻉ ﻣﺴﺌﻠﻪ ﻛﺎﺭﺑﺮﺩ ﺩﺍﺭﻧﺪ. ﺍﻳﻦ ﺗﺎﺑﻊ ﺗﻮﺳﻂ ﻃﺮﺍﺡ ﻣﺴﺌﻠﻪ ﺍﻧﺘﺨﺎﺏ ﻣﻰﮔﺮﺩﺩ ﻭ ﺑﺮ ﺍﺳﺎﺱ ﺍﻧﺘﺨﺎﺏ ﺍﻟﮕﻮﺭﻳﺘﻢ ﻳﺎﺩﮔﻴﺮﻯ، ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻯ ﻣﺴﺌﻠﻪ )ﻭﺯﻥﻫﺎ( ﺗﻨﻈﻴﻢ ﻣﻰﮔﺮﺩﺩ. ﺍﻧﻮﺍﻉ ﺗﻮﺍﺑﻊ ﻓﻌﺎﻟﻴﺖ ﺧﺮﻭﺟــﻰ ﻧﻬﺎﻳــﻰ ﻧــﺮﻭﻥ ﻣﺼﻨﻮﻋﻰ ﺑﺎ ﺍﺳــﺘﻔﺎﺩﻩ ﺍﺯ ﺗﻮﺍﺑﻊ ﻓﻌﺎﻟﻴﺖ ﻣﺤﺎﺳــﺒﻪ ﻣﻰﺷــﻮﺩ. ﻣﻌﻤﻮﻝﺗﺮﻳــﻦ ﺗﻮﺍﺑــﻊ ﻓﻌﺎﻟﻴﺘﻰ ﻛﻪ ﺩﺭ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒــﻰ ﻣﺼﻨﻮﻋﻰ ﺍﺯ ﺁﻧﻬﺎ ﺍﺳــﺘﻔﺎﺩﻩ ﻣﻰﻛﻨﻨﺪ ،ﺑﻪ ﺷﺮﺡ ﺯﻳﺮ ﺍﺳﺖ:
ﺩﺭ ﻫﻤﻪ ﺭﺍﺑﻄﻪﻫﺎﻯ ﻓﻮﻕ netﻣﻘﺪﺍﺭ ﺧﺮﻭﺟﻰ ﺍﻭﻟﻴﻪ ﻧﺮﻭﻥ ﺭﺍ ﻧﺸــﺎﻥ ﻣﻰﺩﻫﺪ .ﻣﻨﻈﻮﺭ ﺍﺯ ﺧﺮﻭﺟﻰ ،ﭘﺎﺳﺦ ﻣﺴﺌﻠﻪ ﻫﺴﺖ. ﻣﺪﻝ ﻧﺮﻭﻥ ﻫﺮ ﺷــﺒﻜﻪ ﻋﺼﺒﻰ ﺷﺎﻣﻞ ﻳﻚ ﺳــﺮﻯ ﻧﺮﻭﻥﻫﺎﻳﻰ ﻣﻰﺷﻮﺩ ﻛﻪ ﺑﺎ ﻫﻢ ﻣﺮﺗﺒﻄﻨﺪ .ﻫﺮ ﻧﺮﻭﻥ ﺭﺍ ﻣﻰﺗــﻮﺍﻥ ﺑﻪ ﻋﻨﻮﺍﻥ ﻳﻚ ﺟﺰء ﻛﻮﭼﻚ ﻭ ﻫــﺮ ﺍﺭﺗﺒﺎﻁ ﺑﻴﻦ ﻧﺮﻭﻥ ﺭﺍ ﺑﻪ ﻋﻨﻮﺍﻥ ﻳﻚ ﻻﻳﻪ ﺗﺼﻮﺭ ﻛﺮﺩ .ﺑﻪ ﻋﻼﻭﻩ ﻫﺮ ﻻﻳﻪ ،ﻭﺯﻧﻰ ﺩﺍﺭﺩ ﻛﻪ ﺑﻴﺎﻧﮕﺮ ﺁﻥ ﺍﺳﺖ ﻛﻪ ﺩﻭ ﻧﺮﻭﻥ ﺗﺎ ﭼﻪ ﻣﻴﺰﺍﻥ ﺭﻭﻯ ﻳﻜﺪﻳﮕﺮ ﺗﺄﺛﻴﺮ ﻣﻰﮔﺬﺍﺭﻧﺪ .ﺑﻨﺎﺑﺮﺍﻳﻦ ﺍﮔﺮ ﻭﺯﻧﻰ ﺯﻳﺎﺩﺗﺮ ﺑﺎﺷﺪ ،ﺩﻭ ﻧﺮﻭﻥ ﺭﻭﻯ ﻳﻜﺪﻳﮕﺮ ﺑﻴﺸــﺘﺮﻳﻦ ﺗﺄﺛﻴﺮ ﺭﺍ ﻣﻰﮔﺬﺍﺭﻧﺪ ﻭ ﺳــﻴﮕﻨﺎﻝ ﻗﻮﻯﺗﺮﻯ ﻣﻰﺗﻮﺍﻧﺪ ﺍﺯ ﺍﻳــﻦ ﻻﻳﻪ ﻋﺒﻮﺭ ﻛﻨﺪ. ﺑﻪ ﻃﻮﺭ ﻛﻠﻰ ﻧﺮﻭﻥ ﻛﻮﭼﻚﺗﺮﻳﻦ ﻭﺍﺣﺪ ﭘﺮﺩﺍﺯﺷــﮕﺮ ﺍﻃﻼﻋﺎﺕ ﺍﺳــﺖ ﻛﻪ ﺍﺳﺎﺱ ﻋﻤﻠﻜﺮﺩ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺭﺍ ﺗﺸﻜﻴﻞ ﻣﻰﺩﻫﺪ.
ﺍﻧﻮﺍﻉ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺍﺯ ﻧﻈﺮ ﺑﺮﮔﺸﺖﭘﺬﻳﺮﻯ
ﺷــﺒﻜﻪﻫﺎﻯ ﭘﻴﺶﺧــﻮﺭ :8ﺩﺭ ﻳﻚ ﺷــﺒﻜﻪ ﭘﻴﺶﺧﻮﺭ ﺟﺮﻳﺎﻥ ﺍﻃﻼﻋــﺎﺕ ﺑﻪ ﺻﻮﺭﺕ ﺧﻂ ﻣﺴــﺘﻘﻴﻢ ﺍﺯ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﻳﺎ ﺍﺯ ﻻﻳﻪ ﻣﻴﺎﻧﻰ )ﭘﻨﻬﺎﻧﻰ( ﺑﻪ ﻃﺮﻑ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﻣﻰﺑﺎﺷــﺪ ﻭ ﺑﺎﺯﺧﻮﺭﻯ ﺑﻴﻦ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﻭ ﺩﻳﮕﺮ ﻻﻳﻪﻫﺎ ﻭﺟﻮﺩ ﻧﺪﺍﺭﺩ ﻭ ﻧﻴﺰ ﺑﻴﻦ ﺍﻋﺼﺎﺏ ﻫﺮ ﻻﻳﻪ ﻧﻴﺰ ﺍﺭﺗﺒﺎﻃﻰ ﻭﺟﻮﺩ ﻧﺪﺍﺭﺩ .ﺳﺎﺩﻩﺗﺮﻳﻦ ﺍﻳﻦ ﺷﺒﻜﻪﻫﺎ ﺷﺒﻜﻪﻫﺎﻯ ﭘﺮﺳﭙﺘﺮﻭﻥ ﻫﺴﺘﻨﺪ.
ﺷﻜﻞ :4ﺷﺒﻜﻪ ﭘﻴﺶﺧﻮﺭ 9
ﺷــﺒﻜﻪﻫﺎﻯ ﭘﺲﺧﻮﺭ )ﺑﺮﮔﺸﺘﻰ( :ﺗﻔﺎﻭﺕ ﺷﺒﻜﻪﻫﺎﻯ ﭘﺲﺧﻮﺭ ﺑﺎ ﺷﺒﻜﻪﻫﺎﻯ ﭘﻴﺶﺧﻮﺭ ﺩﺭ ﺁﻥ ﺍﺳﺖ ﻛﻪ ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﭘﺲﺧﻮﺭ ﺣﺪﺍﻗﻞ ﻳﻚ ﺳﻴﮕﻨﺎﻝ ﺑﺮﮔﺸﺘﻰ ﺍﺯ ﻳﻚ ﻧﺮﻭﻥ ﺑﻪ ﻫﻤﺎﻥ ﻧﺮﻭﻥ ﻳﺎ ﻧﺮﻭﻥﻫﺎﻯ ﻫﻤﺎﻥ ﻻﻳﻪ ﻳﺎ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪﻫﺎﻯ ﻗﺒﻞ ﻭﺟﻮﺩ ﺩﺍﺭﺩ .ﺷﺒﻜﻪﻫﺎﻯ ﺑﺮﮔﺸــﺘﻰ ﺑﻬﺘﺮ ﻣﻰﺗﻮﺍﻧﻨﺪ ﺭﻓﺘﺎﺭ ﻣﺮﺑﻮﻁ ﺑﻪ ﻭﻳﮋﮔﻰﻫﺎﻯ ﺯﻣﺎﻧﻰ ﻭ ﭘﻮﻳﺎﻳﻰ ﺳﻴﺴــﺘﻢﻫﺎ ﺭﺍ ﻧﺸﺎﻥ ﺩﻫﻨﺪ .ﺳﺎﺩﻩﺗﺮﻳﻦ ﺍﻳﻦ ﺷﺒﻜﻪﻫﺎ ،ﺷﺒﻜﻪ ﻫﺎﭘﻔﻴﻠﺪ ﻣﻰﺑﺎﺷﺪ.
ﺷﻜﻞ :5ﺷﺒﻜﻪ ﭘﺲﺧﻮﺭ 8 .Feedforward Networks 9 .Recurrent Networks
18 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127
ﭘﺮﺳﭙﺘﺮﻭﻥ ﻛﻪ ﻓﺮﺍﻧــﻚ ﺭﻭﺯﻧﺒﻼﺕ ﺍﺯ ﻳﻜﻰ ﺍﺯ ﺍﻧﻮﺍﻉ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ، ﺁﺯﻣﺎﻳﺸﮕﺎﻩ ﭘﺮﻭﺍﺯ ﻛﺮﻧﻞ ،ﺍﻳﻦ ﻣﺪﻝ ﻣﺤﺎﺳﺒﺎﺗﻰ ﺭﺍ ﺑﻪ ﻧﺎﻡ »ﭘﺮﺳﭙﺘﺮﻭﻥ« ﺍﻳﺠﺎﺩ ﻧﻤﻮﺩ .ﺍﻳﻨﻬﺎ ﺑﻪ ﺻﻮﺭﺕ ﭘﺮﺳﭙﺘﺮﻭﻥﻫﺎﻯ ﺗﻚﻻﻳﻪ ﻭ ﭼﻨﺪﻻﻳﻪ ﻣﻮﺟﻮﺩ ﻫﺴﺘﻨﺪ .ﭘﺮﺳﭙﺘﺮﻭﻥ ﺗﻚﻻﻳﻪ ﺗﻨﻬﺎ ﻣﻰﺗﻮﺍﻧﺪ ﻣﺴﺎﺋﻞ ﻣﺠﺰﺍﻯ ﺧﻄﻰ ﺭﺍ ﺩﺳﺘﻪﺑﻨﺪﻯ ﻛﻨﺪ ﻭ ﺑﺮﺍﻯ ﻣﺴﺎﺋﻞ ﭘﻴﭽﻴﺪﻩﺗﺮ ﻻﺯﻡ ﺍﺳﺖ ﻛﻪ ﺍﺯ ﺗﻌﺪﺍﺩ ﺑﻴﺸﺘﺮﻯ ﻻﻳﻪ ﺍﺳﺘﻔﺎﺩﻩ ﻛﻨﻴﻢ. ﻋﻤﻮﻣﻰﺗﺮﻳﻦ ﺷــﻜﻞ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ،ﺷﺒﻜﻪﻫﺎﻯ ﭘﺮﺳــﭙﺘﺮﻭﻥ ﭼﻨﺪﻻﻳﻪ )(MPL ﻫﺴﺘﻨﺪ ﻛﻪ ﻳﻚ ﭘﺮﺳﭙﺘﺮﻭﻥ ﭼﻨﺪﻻﻳﻪ: -1ﺍﺯ ﺗﻌﺪﺍﺩﻯ ﭼﻨﺪ ﻭﺭﻭﺩﻯ ﺗﺸﻜﻴﻞ ﻳﺎﻓﺘﻪ ﺍﺳﺖ. -2ﻳﻚ ﻳﺎ ﭼﻨﺪ ﻻﻳﻪ ﻣﻴﺎﻧﻰ )ﭘﻨﻬﺎﻥ( ﺩﺍﺭﺩ. -3ﺍﺯ ﺗﻮﺍﺑﻊ ﺗﺮﻛﻴﺒﻰ ﺧﻄﻰ ﺩﺭ ﻻﻳﻪﻫﺎﻯ ﻭﺭﻭﺩﻯ ﺍﺳﺘﻔﺎﺩﻩ ﻣﻰﻛﻨﺪ. -4ﺩﺭ ﻻﻳﻪﻫﺎﻯ ﻣﻴﺎﻧﻰ )ﭘﻨﻬﺎﻥ( ﻣﻌﻤﻮﻻً ﺍﺯ ﺗﻮﺍﺑﻊ ﻓﻌﺎﻟﺴــﺎﺯﻯ ﺳــﻴﮕﻤﻮﺋﻴﺪ 11ﺍﺳــﺘﻔﺎﺩﻩ ﻣﻰﻛﻨﺪ. -5ﺩﺍﺭﺍﻯ ﺗﻌﺪﺍﺩﻯ ﺧﺮﻭﺟﻰ ﺑﺎ ﺗﻮﺍﺑﻊ ﻓﻌﺎﻟﺴﺎﺯﻯ ﻣﺨﺘﻠﻒ ﺍﺳﺖ. -6ﺑﻴﻦ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﻭ ﺍﻭﻟﻴﻦ ﻻﻳﻪ ﻣﻴﺎﻧﻰ ﻭ ﺑﻴﻦ ﻻﻳﻪﻫﺎﻯ ﭘﻨﻬﺎﻥ ﻭ ﻧﻴﺰ ﺑﻴﻦ ﺁﺧﺮﻳﻦ ﻻﻳﻪ ﭘﻨﻬﺎﻥ ﻭ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﺍﺭﺗﺒﺎﻁ ﻭﺟﻮﺩ ﺩﺍﺭﺩ. ﺷﻜﻞ ﻳﻚ ﻋﺼﺐ ﭘﺮﺳﭙﺘﺮﻭﻥ ﻫﻤﺎﻧﻨﺪ ﺷﻜﻞ 4ﻣﻰﺑﺎﺷﺪ. ﺷﺒﻜﻪ ﻫﺎﭘﻔﻴﻠﺪ ﺷــﺒﻜﻪ ﻫﺎﭘﻔﻴﻠﺪ12 ﻳﻚ ﺳﻴﺴﺘﻢ ﺑﺎﺯﺧﻮﺭﺩ ﭼﻨﺪ ﺣﻠﻘﻮﻯ ﺍﺳــﺖ ﻭ ﺷﺒﻜﻪﺍﻯ ﭘﺲﺧﻮﺭ ﻳﺎ ﺑﺮﮔﺸــﺘﻰ ﻣﻰﺑﺎﺷﺪ .ﺷــﺒﻜﻪ ﻫﺎﭘﻔﻴﻠﺪ ﺑﻪ ﺗﻌﺪﺍﺩ ﻭ ﺣﺠﻢ ﺍﻃﻼﻋﺎﺕ ﺟﺬﺏ ﺷﺪﻩ ﻭ ﺣﺎﻓﻈﻪ ﺍﻧﺠﻤﻨــﻰ ﺗﻮﺟﻪ ﺩﺍﺭﺩ ﻛﻪ ﮔﻔﺘﻪ ﻣﻰﺷــﻮﺩ :ﻣﻌﻤﺎﺭﻯ ﺣﺎﻓﻈﻪ ﻣﻰﺗﻮﺍﻧــﺪ ﺑﺎ ﺣﺠﻢ ﻛﻤﻰ ﺍﺯ ﺍﻃﻼﻋﺎﺕ ﺷﻨﺎﺳﺎﻳﻰ ﺷﻮﺩ ﻛﻪ ﺍﻳﻦ ﺷﺒﻴﻪ ﺑﻪ ﻛﺎﺭ ﻣﻐﺰ ﺩﺭ ﺯﻣﺎﻥ ﻳﺎﺩﺁﻭﺭﻯ ﻣﻰﺑﺎﺷﺪ .ﻣﺎ ﺍﻓﺮﺍﺩ ﺭﺍ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻣﻮ ،ﭼﺸﻢ ،ﺻﺪﺍ ﻭ ﺩﻳﮕﺮ ﻋﻼﺋﻢ ﺑﻪ ﺧﺎﻃﺮ ﻣﻰﺳﭙﺎﺭﻳﻢ. ﺷــﻜﻞ ﺯﻳﺮ ﭘﻴﻮﺳﺘﮕﻰ ﺳﺎﺧﺘﻤﺎﻥ ﺷــﺒﻜﻪ ﻫﺎﭘﻔﻴﻠﺪ ﺭﺍ ﻧﺸــﺎﻥ ﻣﻰﺩﻫﺪ .ﻫﻤﻪ ﻭﺍﺣﺪﻫﺎ ﺑﻪ ﻭﺍﺣﺪﻫــﺎﻯ ﺩﻳﮕﺮ ﻣﺮﺗﺒﻂ ﺍﺳــﺖ ﻭ ﻫﻴﭻ ﻻﻳﻪﺍﻯ ﻣﻘﺪﻡ ﺑﺮ ﻻﻳﻪ ﺩﻳﮕﺮ ﻧﻴﺴــﺖ .ﺑﻪ ﻋﻼﻭﻩ، ﺍﺭﺗﺒﺎﻃﺎﺕ ،ﺩﻭ ﻃﺮﻓﻪ ﻭ ﻣﺘﻘﺎﺭﻥ ﻫﺴــﺘﻨﺪ .ﻳﻚ ﻭﺯﻥ ﻫﻢ ﺑﺮﺍﻯ ﻫﺮ ﺍﺗﺼﺎﻝ ﺗﻌﻴﻴﻦ ﻣﻰﺷﻮﺩ ﻛﻪ ﺩﺭ ﻫﺮ ﺩﻭ ﺟﻬﺖ ﻳﻜﺴﺎﻥ ﻣﻰﺑﺎﺷﺪ. ﭘﺮﺳــﭙﺘﺮﻭﻥ 10ﻣﻰﺑﺎﺷــﺪ
ﺷﺒﻜﻪﻫﺎﻯ 13 RBF
ﺷــﺒﻜﻪﻫﺎﻯ RBFﺍﺯ ﻧﻮﻉ ﺷﺒﻜﻪﻫﺎﻯ ﭘﻴﺶﺧﻮﺭ ﻫﺴﺘﻨﺪ ،ﻟﻴﻜﻦ ﻓﻘﻂ ﺑﺎ ﻳﻚ ﻻﻳﻪ ﻣﻴﺎﻧﻰ. ﻳﻚ ﺷﺒﻜﻪ :RBF ﺩﺍﺭﺍﻯ ﺗﻌﺪﺍﺩﻯ ﻭﺭﻭﺩﻯ ﺍﺳﺖ. ﻓﻘﻂ ﺩﺍﺭﺍﻯ ﻳﻚ ﻻﻳﻪ ﻣﻴﺎﻧﻰ ﻣﻰﺑﺎﺷﺪ. ﺍﺯ ﺗﺎﺑﻊ ﺗﺮﻛﻴﺒﻰ ﺭﺍﺩﻳﺎﻟﻰ )ﺷﻌﺎﻋﻰ( ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻰ ﺍﺳﺘﻔﺎﺩﻩ ﻣﻰﻛﻨﺪ. ﺩﺭ ﻻﻳﻪ ﭘﻨﻬﺎﻥ ﺍﺯ ﺗﺎﺑﻊ ﻓﻌﺎﻟﺴــﺎﺯﻯ ﻧﻤﺎﻳﻰ ﻳﺎ softmaxﺍﺳــﺘﻔﺎﺩﻩ ﻣﻰﻛﻨﺪ ﻛﻪ ﺷﺒﻜﻪ RBFﻳﻚ ﺷﺒﻜﻪ ﮔﺎﻭﺳﻰ ﺍﺳﺖ. 10 .Percepetron 11 .Sigmoidal Function 12 .Hopfield Network 13 .Radial Basis Function Networks
ﺩﺍﺭﺍﻯ ﺗﻌﺪﺍﺩﻯ ﺧﺮﻭﺟﻰ ﺑﺎ ﺍﻧﻮﺍﻉ ﺗﻮﺍﺑﻊ ﻓﻌﺎﻟﺴﺎﺯﻯ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻣﻮﺭﺩ ﻣﺴﺌﻠﻪ ﻣﻰﺑﺎﺷﺪ. ﻳﻚ ﺭﺍﺑﻄﻪ ﭘﻴﺶﺧﻮﺭ ﺑﻴﻦ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﻭ ﺧﺮﻭﺟﻰ ﻭﺟﻮﺩ ﺩﺍﺭﺩ.ﺷﺒﻜﻪ ﻋﺼﺒﻰ ﻛﻮﻫﻦ ﺷــﺒﻜﻪﻫﺎﻯ ﺗﺮﺳــﻴﻢ ﻧﻘﺸﻪ ﺧﻮﺩﺳــﺎﺯﻣﺎﻧﺪﻫﻰ )ﺷــﺒﻜﻪﻫﺎﻯ ﺧﻮﺩ ﺳــﺎﺯﻣﺎﻧﺪﻩ ﻳﺎ ﻛﻮﻫﻦ ﻳــﺎ (SOFMﻛﺎﻣ ً ﻼ ﻣﺘﻔﺎﻭﺕ ﺍﺯ ﺍﻧﻮﺍﻉ ﺩﻳﮕﺮ ﺷــﺒﻜﻪﻫﺎ ﻣﻰﺑﺎﺷــﻨﺪ .ﺍﺯ ﺁﻧﺠﺎﻳﻰ ﻛﻪ ﺩﻳﮕﺮ ﺷــﺒﻜﻪﻫﺎ ﻳﺎﺩﮔﻴﺮﻳﺸﺎﻥ ﺑﻪ ﺻﻮﺭﺕ ﻧﻈﺎﺭﺕ ﺷﺪﻩ ﻣﻰﺑﺎﺷــﺪ ،ﺷﺒﻜﻪﻫﺎﻯ SOFMﺗﺤﺖ ﻳﺎﺩﮔﻴﺮﻯ ﺑﺪﻭﻥ ﻧﻈﺎﺭﺕ ﻃﺮﺍﺣﻰ ﺷﺪﻩﺍﻧﺪ ).(1996 ,Patterson ﺩﺭ ﺍﻭﻟﻴﻦ ﻧﮕﺎﻩ ﻋﺠﻴﺐ ﺑﻪ ﻧﻈﺮ ﻣﻰﺭﺳﺪ؛ ﺑﺪﻭﻥ ﺩﺍﺷﺘﻦ ﺧﺮﻭﺟﻰ ﺷﺒﻜﻪ ﭼﮕﻮﻧﻪ ﻳﺎﺩﮔﻴﺮﻯ ﻣﻰﻛﻨﺪ؟ ﺟﻮﺍﺏ ﺍﻳﻦ ﺍﺳﺖ ﻛﻪ ﺷﺒﻜﻪ SOFMﺳﻌﻰ ﺩﺭ ﻳﺎﺩﮔﻴﺮﻯ ﺳﺎﺧﺘﺎﺭ ﺩﺍﺩﻩﻫﺎ ﺩﺍﺭﺩ. ﺍﻭﻟﻴﻦ ﺍﺣﺘﻤﺎﻝ ﻣﻮﺟﻮﺩ ﺑﺮﺍﻯ ﺍﺳــﺘﻔﺎﺩﻩ ﺍﺯ ﺍﻳﻨﻬﺎ ،ﺗﺠﺰﻳﻪ ﻭ ﺗﺤﻠﻴﻞ ﺍﻛﺘﺸﺎﻓﻰ ﺩﺍﺩﻩﻫﺎﺳﺖ ﻭ ﺩﻭﻣﻴﻦ ﺍﺣﺘﻤﺎﻝ ﻛﺸــﻒ ﻭ ﻳﺎﻓﺘﻦ ﭼﻴﺰﻫﺎﻯ ﺟﺪﻳﺪ ﻣﻰﺑﺎﺷﺪ .ﻳﻚ ﺷﺒﻜﻪ SOFMﻓﻘﻂ ﺩﻭ ﻻﻳﻪ ﺩﺍﺭﺩ؛ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﻭ ﻻﻳﻪ ﺧﺮﻭﺟﻰ. ﺣﺎﻓﻈــﻪ ﺍﻧﺠﻤﻨــﻰ ﻭ ﻗﺎﺑﻠﻴﺖ ﺣــﻞ ﻣﺴــﺎﻳﻞ ﺑﻬﻴﻨﻪﺳــﺎﺯﻯ ﺩﻭ ﻣﺰﻳﺖ ﻋﻤــﺪﻩ ﺍﻳﻦ ﻧﻮﻉ ﺷﺒﻜﻪﻫﺎﺳﺖ .ﺣﺎﻓﻈﻪ ﺍﻧﺠﻤﻨﻰ ﺑﻪ ﺍﻳﻦ ﻣﻌﻨﺎﺳﺖ ﻛﻪ ﻣﺎ ﺧﺼﻮﺻﻴﺎﺕ ﻭ ﻭﻳﮋﮔﻰﻫﺎﻯ ﭼﻴﺰﻯ ﺭﺍ ﺑﮕﻮﻳﻴﻢ ﻭ ﺳﻴﺴــﺘﻢ ﺑﺘﻮﺍﻧﺪ ﺩﻳﮕﺮ ﻭﻳﮋﮔﻰﻫﺎﻯ ﺁﻥ ﺭﺍ ﺑﻪ ﻣﺎ ﻧﺸــﺎﻥ ﺩﻫﺪ .ﻫﺎﭘﻔﻴﻠﺪ ﻧﺸﺎﻥ ﺩﺍﺩ ﻛﻪ ﺍﻳﻦ ﺷﺒﻜﻪ ﺑﻪ ﺳﻤﺖ ﻭﺿﻌﻴﺘﻰ ﭘﺎﻳﺪﺍﺭ ﺳﻴﺮ ﻣﻰﻛﻨﺪ ﻭ ﻫﻢﮔﺮﺍﺳﺖ. ﺷــﺒﻜﻪ ﻛﻮﻫﻦ ،ﻳﻜﻰ ﺍﺯ ﺍﻧﻮﺍﻉ ﺷــﺒﻜﻪﻫﺎﻯ ﺭﻗﺎﺑﺘﻰ ﻫﺴــﺘﻨﺪ .ﺷــﺒﻜﻪ ﻛﻮﻫﻦ ﺗﻨﻬﺎ ﺷﺎﻣﻞ ﺩﻭ ﻻﻳﻪ ﻣﻰﺷــﻮﺩ ،ﻻﻳــﻪ ﻭﺭﻭﺩﻯ ﻭ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﺭﻗﺎﺑﺘﻰ .ﻫﺮ ﻧــﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﺑﻪ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪ ﺭﻗﺎﺑﺘﻰ ﻭﺻﻞ ﺷــﺪﻩ ﺍﺳــﺖ .ﺩﺭ ﺿﻤﻦ ،ﻫﺮ ﻧﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﺭﻗﺎﺑﺘﻰ ﻣﻤﻜﻦ ﺍﺳﺖ ﺑﻪ ﻫﻤﻪ ﻧﺮﻭﻥﻫﺎﻯ ﺭﻗﺎﺑﺘﻰ ﺩﻳﮕﺮ ﻣﺘﺼﻞ ﺷﺪﻩ ﺑﺎﺷﺪ.
ﻳﺎﺩﮔﻴﺮﻯ ﻳﺎﺩﮔﻴﺮﻯ ﻋﺒﺎﺭﺕ ﺍﺳﺖ ﺍﺯ ﻓﺮﺍﻳﻨﺪ ﺗﻌﺪﻳﻞ ﺍﻭﺯﺍﻥ ﺍﺭﺗﺒﺎﻃﻰ ﺩﺭ ﻳﻚ ﺷﺒﻜﻪ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ، ﺑﻪ ﮔﻮﻧﻪﺍﻯ ﻛﻪ ﺷﺒﻜﻪ ﺑﺘﻮﺍﻧﺪ ﺑﻪ ﻫﻨﮕﺎﻡ ﺩﺭﻳﺎﻓﺖ ﺑﺮﺩﺍﺭ ﻣﺤﺮﻙ ﺗﻮﺳﻂ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ،ﺑﺮﺩﺍﺭ ﺧﺮﻭﺟﻰ ﺩﻟﺨﻮﺍﻩ ﺭﺍ ﺑﻪ ﻋﻨﻮﺍﻥ ﭘﺎﺳﺦ ﺗﻮﻟﻴﺪ ﻛﻨﺪ .ﺍﻧﻮﺍﻉ ﻳﺎﺩﮔﻴﺮﻯ ﻋﺒﺎﺭﺗﻨﺪ ﺍﺯ: ﻳﺎﺩﮔﻴﺮﻯ ﺑﺎ ﻧﺎﻇﺮ )ﺑﺎ ﻧﻈﺎﺭﺕ( ﺩﺭ ﻳﺎﺩﮔﻴﺮﻯ ﺑﺎ ﻧﺎﻇﺮ ،14ﻫﻨﮕﺎﻣﻰ ﻛﻪ ﻭﺭﻭﺩﻯ ﺑﻪ ﺷــﺒﻜﻪ ﺍﻋﻤﺎﻝ ﻣﻰﺷــﻮﺩ ﺟﻮﺍﺏ ﺷﺒﻜﻪ ﺑﺎ ﺟﻮﺍﺏ ﻫﺪﻓﻰ ﻛﻪ ﻣﺎ ﺑﺮﺍﻯ ﺷــﺒﻜﻪ ﺗﻌﻴﻴﻦ ﻛﺮﺩﻩﺍﻳﻢ ﻣﻘﺎﻳﺴــﻪ ﻣﻰﺷﻮﺩ ﻭ ﺳﭙﺲ ﺧﻄﺎﻯ ﻳﺎﺩﮔﻴﺮﻯ ﻣﺤﺎﺳﺒﻪ ﺷﺪﻩ ﻭ ﺍﺯ ﺁﻥ ﺑﺮﺍﻯ ﺗﻨﻈﻴﻢ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻯ ﺷﺒﻜﻪ ﺍﺳﺘﻔﺎﺩﻩ ﻣﻰﺷﻮﺩ. ﻳﺎﺩﮔﻴﺮﻯ ﺷﺒﻜﻪﻫﺎﻯ ﻫﺎﭘﻔﻴﻠﺪ ،ﻳﺎﺩﮔﻴﺮﻯ ﺑﺮ ﭘﺎﻳﻪ ﺍﺻﻮﻝ ﻫﺐ ﻭ ﻧﻘﺸﻪ ﺧﻮﺩﺳﺎﺯﻣﺎﻧﺪﻫﻰ ﺍﺯ ﺟﻤﻠﻪ ﻳﺎﺩﮔﻴﺮﻯﻫﺎﻯ ﺑﺎ ﻧﺎﻇﺮ ﻫﺴﺘﻨﺪ. ﻳﺎﺩﮔﻴﺮﻯ ﺗﻘﻮﻳﺘﻰ ﻛﻪ ﻧﻮﻉ ﺧﺎﺻﻰ ﺍﺯ ﻳﺎﺩﮔﻴﺮﻯ ﺑﺎ ﻧﺎﻇﺮ ﺍﺳــﺖ ﻛﻪ ﺳﻴﺴﺘﻢ ﺭﺍ ﺑﻪ ﻭﺳﻴﻠﻪ ﺍﺭﺍﺋﻪ ﻣﻴﺰﺍﻥ ﺻﺤﺖ ﭘﺎﺳﺦﻫﺎﻯ ﺳﻴﺴــﺘﻢ ﺁﻣﻮﺯﺵ ﻣﻰﺩﻫﺪ )ﺍﻣﺎ ﺟﻮﺍﺏ ﺻﺤﻴﺢ ﺭﺍ ﺍﺭﺍﺋﻪ ﻧﻤﻰﺩﻫﺪ( .ﺍﻳﻦ ﺭﻭﺵ ﺩﺭ ﻭﺍﻗﻊ ﻧﻤﺮﻩ ﺩﺍﺩﻥ ﺑﻪ ﭘﺎﺳﺦ ﺳﻴﺴﺘﻢ ﺍﺳﺖ. ﻳﺎﺩﮔﻴﺮﻯ ﺑﺪﻭﻥ ﻧﺎﻇﺮ 15 ﺍﻳﻦ ﻧﻮﻉ ﻳﺎﺩﮔﻴﺮﻯ ﻫﻴﭻ ﺍﻃﻼﻋﺎﺗﻰ ﺭﺍ ﺑﻪ ﺳﻴﺴﺘﻢ ﻧﻤﻰﺩﻫﺪ .ﺩﺭ ﻳﺎﺩﮔﻴﺮﻯ ﺑﺪﻭﻥ ﻧﺎﻇﺮ ، ﻫﺪﻑ ﺍﻳﻦ ﻧﻴﺴــﺖ ﻛﻪ ﺷﺒﻜﻪ ﭘﺎﺳــﺦ ﺻﺤﻴﺢ ﺭﺍ ﻳﺎﺩ ﺑﮕﻴﺮﺩ ،ﺑﻠﻜﻪ ﭘﺎﺳﺦﻫﺎ ﺭﺍ ﻃﺒﻘﻪﺑﻨﺪﻯ ﻣﻰﻛﻨﺪ ﻭ ﭘﺎﺳــﺦ ﺑﻪ ﺗﻮﺍﻧﺎﻳﻰ ﺷــﺒﻜﻪ ﺑﺮﺍﻯ ﺳﺎﺯﻣﺎﻧﺪﻫﻰ ﺧﻮﻳﺶ ﺑﺴــﺘﮕﻰ ﺩﺍﺭﺩ .ﺑﻪ ﻃﻮﺭ ﻣﺜﺎﻝ ،ﺷــﻤﺎ ﻣﻰﺧﻮﺍﻫﻴﺪ ﺳﻴﺴﺘﻤﻰ ﺍﻳﺠﺎﺩ ﻛﻨﻴﺪ ﻛﻪ ﻭﺭﻭﺩﻯﻫﺎ ﺭﺍ ﺑﻪ ﺷﺶ ﺩﺳﺘﻪ ﺗﻘﺴﻴﻢ ﻛﻨﺪ؛ ﭼﮕﻮﻧﮕﻰ ﺩﺳﺘﻪﺑﻨﺪﻯ ﻭﺭﻭﺩﻯﻫﺎ ﺑﻪ ﻳﺎﺩﮔﻴﺮﻯ ﺳﻴﺴﺘﻢ ﺑﺮﻣﻰﮔﺮﺩﺩ. ﻳﺎﺩﮔﻴﺮﻯ ﺷــﺒﻜﻪ ﭘﺮﺳﭙﺘﺮﻭﻥ ،ﻗﺎﻧﻮﻥ ﺩﻟﺘﺎ ﻭ ﻳﺎﺩﮔﻴﺮﻯ ﭘﺲﺍﻧﺘﺸﺎﺭ ﺍﺯ ﺟﻤﻠﻪ ﻳﺎﺩﮔﻴﺮﻯﻫﺎﻯ ﺑﺪﻭﻥ ﻧﺎﻇﺮ ﻫﺴﺘﻨﺪ.
14 .Supervised Learning 15 .Unsupervised Learning
19 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127
ﻣﺰﺍﻳﺎ ﻭ ﻭﻳﮋﮔﻰﻫﺎﻯ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺑﻪ ﺩﻟﻴﻞ ﭘﺮﺩﺍﺯﺵ ﻣﻮﺍﺯﻯ ،ﺍﺯ ﺳﺮﻋﺖ ﺑﺎﻻﻳﻰ ﺑﺮﺧﻮﺭﺩﺍﺭﻧﺪ .ﭘﺮﺩﺍﺯﺵﻣﻮﺍﺯﻯ ﺍﻃﻼﻋﺎﺕ ﺩﺭ ﻣﻐﺰ ﺑﺪﻳﻦ ﺻﻮﺭﺕ ﺍﺳــﺖ ﻛﻪ ﻫﺮ ﻛﺪﺍﻡ ﺍﺯ ﺍﻋﻤﺎﻝ ﺩﻳﺪﻥ ،ﺷــﻨﻴﺪﻥ، ﻟﻤــﺲ ﻛــﺮﺩﻥ ﻭ ﻏﻴﺮﻩ ﻣﻰﺗﻮﺍﻧﻨﺪ ﻣﺴــﺘﻘﻞ ﻭ ﻫﻤﺰﻣــﺎﻥ ﺍﻧﺠﺎﻡ ﺷــﻮﻧﺪ .ﻛﺎﻣﭙﻴﻮﺗﺮﻫﺎ ﻫﻢ ﻣﻰﺗﻮﺍﻧﻨﺪ ﺁﻧﻘﺪﺭ ﺳﺮﻳﻊ ﺷﻮﻧﺪ ﺗﺎ ﺑﻪ ﺭﻭﺵ ﺳﺮﻳﺎﻝ ﺍﻋﻤﺎﻝ ﺩﻳﺪﻥ ،ﻟﻤﺲ ﻛﺮﺩﻥ ،ﻓﻜﺮ ﻛﺮﺩﻥ ﻭ ﻏﻴﺮﻩ ﺭﺍ ﺑﻪ ﺗﺮﺗﻴﺐ ﺍﻧﺠﺎﻡ ﺩﻫﻨﺪ ﻭ ﭼﻮﻥ ﺳــﺮﻋﺖ ﻛﺎﻣﭙﻴﻮﺗﺮﻫﺎ ﺑﺴــﻴﺎﺭ ﺑﺎﻻﺳﺖ ،ﺗﺼﻮﺭ ﻣﺎ ﺑﺮ ﺍﻳﻦ ﺍﺳﺖ ﻛﻪ ﺗﻤﺎﻡ ﺍﻋﻤﺎﻝ ﻫﻤﺰﻣﺎﻥ ﺍﻧﺠﺎﻡ ﻣﻰﮔﻴﺮﺩ. ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺗﻮﺍﻥ ﺑﺎﻟﻘﻮﻩﺍﻯ ﺑﺮﺍﻯ ﺣﻞ ﻣﺴــﺎﺋﻠﻰ ﺩﺍﺭﻧﺪ ﻛﻪ ﺷﺒﻴﻪﺳﺎﺯﻯ ﺁﻧﻬﺎ ﺍﺯﻃﺮﻳﻖ ﻣﻨﻄﻘﻰ ﻭ ﻳﺎ ﺳﺎﻳﺮ ﺭﻭﺵﻫﺎ ﻣﺸﻜﻞ ﻭ ﻳﺎ ﻏﻴﺮ ﻣﻤﻜﻦ ﺍﺳﺖ. ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻫﻤﺎﻧﻨﺪ ﻣﻐﺰ ﺍﻧﺴﺎﻥ ﺑﻪ ﻃﻮﺭ ﭘﻴﻮﺳﺘﻪ ﺩﺭ ﺣﺎﻝ ﻳﺎﺩﮔﻴﺮﻯ ﻭ ﺍﻧﻄﺒﺎﻕ ﺑﺎﻣﺤﻴﻂ ﻫﺴــﺘﻨﺪ .ﺑﻪ ﺍﻳﻦ ﻣﻌﻨﻰ ﻛﻪ ﺍﮔﺮ ﺷــﺒﻜﻪ ﺑﺮﺍﻯ ﻳﻚ ﻭﺿﻌﻴﺖ ﺧﺎﺹ ﺁﻣﻮﺯﺵ ﺑﺒﻴﻨﺪ ﻭ ﺗﻐﻴﻴﺮ ﻛﻮﭼﻜﻰ ﺩﺭ ﺷــﺮﺍﻳﻂ ﻣﺤﻴﻄﻰ ﺁﻥ ﺭﺥ ﺩﻫﺪ ،ﻣﻰﺗﻮﺍﻧﺪ ﺑﺎ ﺁﻣﻮﺯﺵ ﻣﺨﺘﺼﺮ ،ﺑﺮﺍﻯ ﺷﺮﺍﻳﻂ ﺟﺪﻳﺪ ﻧﻴﺰ ﻛﺎﺭﺁﻣﺪ ﺑﺎﺷﺪ. ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ،ﻋﺪﻡ ﻋﻤﻠﻜﺮﺩ ﺻﺤﻴﺢ ﻗﺴﻤﺘﻰ ﺍﺯ ﻧﺮﻭﻥﻫﺎ ﻣﻮﺟﺐ ﺍﺯ ﻛﺎﺭﺍﻓﺘﺎﺩﮔﻰﻛﺎﻣﻞ ﻣﻐﺰ ﻧﻤﻰﺷﻮﺩ ﻭ ﺍﻣﻜﺎﻥ ﺍﺗﺨﺎﺫ ﺗﺼﻤﻴﻢ ﺻﺤﻴﺢ ﻧﻴﺰ ﻭﺟﻮﺩ ﺩﺍﺭﺩ. ﺍﻳﻦ ﺭﻭﺵ ﻗﺎﺩﺭ ﺍﺳﺖ ﺑﺮﺍﻯ ﺩﺍﺩﻩﻫﺎ ﺩﺭ ﺷﺮﺍﻳﻂ ﻋﺪﻡ ﺍﻃﻤﻴﻨﺎﻥ )ﺍﻋﻢ ﺍﺯ ﺁﻧﻜﻪ ﻓﺎﺯﻯ ﺑﺎﺷﻨﺪﻭ ﻳــﺎ ﺑﻪ ﻃﻮﺭ ﻧﺎﻗــﺺ ﻭ ﺗﻮﺃﻡ ﺑﺎ ﺩﺭﻳﺎﻓﺖ ) noiseﺩﺍﺩﻩﻫﺎﻯ ﺩﺍﺭﺍﻯ ﺧﻄﺎ ﺭﺍ ﮔﻮﻳﻨﺪ( ﺟﻮﺍﺏ ﻣﻨﻄﻘﻰ ﺍﺭﺍﺋﻪ ﺩﻫﺪ.
ﻣﺤﺪﻭﺩﻳﺖﻫﺎﻯ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ
ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ ﻗﺎﺩﺭ ﺑﻪ ﺗﻮﺿﻴﺢ ﻣﻨﻄﻖ ﻭ ﻗﺎﻋﺪﻩ ﻛﺎﺭ ﻧﻴﺴــﺘﻨﺪ ﻭ ﺍﺛﺒﺎﺕ ﺩﺭﺳﺘﻰ ﻧﺘﺎﻳﺞ ﺁﻧﻬﺎ ﺑﺴﻴﺎﺭ ﺩﺷﻮﺍﺭ ﺍﺳﺖ. ﻣﺤﺎﺳــﺒﺎﺕ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﻌﻤﻮﻻً ﻣﺤﺘﺎﺝ ﻣﻘﺎﺩﻳﺮ ﺯﻳﺎﺩﻯ ﺩﺍﺩﻩ ﺑﺮﺍﻯ ﺁﻣﻮﺯﺵ ﻭ ﺁﺯﻣﻮﻥ ﻣﺪﻝ ﺍﺳﺖ. ﺩﺭ ﺣﺎﻟﺖ ﻛﻠﻰ ،ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺑﺮﺍﻯ ﺑﺮﺧﻰ ﺍﺯ ﻣﺴﺎﺋﻞ ﻛﺎﺭﺁﻳﻰ ﻧﺪﺍﺭﻧﺪ .ﺑﺮﺍﻯ ﻣﺜﺎﻝ، ﺑﺮﺍﻯ ﺣﻞ ﻣﺴﺎﺋﻞ ﻭ ﭘﺮﺩﺍﺯﺵ ﺩﺍﺩﻩﻫﺎ ﺑﺎ ﺭﻭﺵ ﻣﺴﺘﺪﻝ ﻣﻨﺎﺳﺐ ﻧﻴﺴﺘﻨﺪ.
ﺗﺸﺮﻳﺢ ﺳﺎﺩﻩ ﻋﻤﻠﻜﺮﺩ ﻳﻚ ﺳﻴﺴﺘﻢ ﻋﺼﺒﻰ
ﻭﻇﻴﻔﻪ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻳﺎﺩﮔﻴﺮﻯ ﺍﺳﺖ .ﺗﻘﺮﻳﺒﺎً ﭼﻴﺰﻯ ﺷﺒﻴﻪ ﻳﺎﺩﮔﻴﺮﻯ ﻳﻚ ﻛﻮﺩﻙ
ﺧﺮﺩﺳﺎﻝ ﻣﻰﺑﺎﺷﺪ .ﻳﺎﺩﮔﻴﺮﻯ ﺩﺭ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺭﺍﻳﺞ ﺑﻪ ﺻﻮﺭﺕ ﺗﺤﺖ ﻧﻈﺎﺭﺕ ﺍﺳﺖ. ﻭﺍﻟﺪﻳــﻦ ﺗﺼﺎﻭﻳﺮ ﺣﻴﻮﺍﻧﺎﺕ ﻣﺨﺘﻠﻒ ﺭﺍ ﺑﻪ ﻛﻮﺩﻙ ﻧﺸــﺎﻥ ﻣﻰﺩﻫﻨــﺪ ﻭ ﻧﺎﻡ ﻫﺮ ﻛﺪﺍﻡ ﺭﺍ ﺑﻪ ﻛــﻮﺩﻙ ﻣﻰﮔﻮﻳﻨﺪ .ﻣﺎ ﺭﻭﻯ ﻳﻚ ﺣﻴﻮﺍﻥ ،ﻣﺜ ً ﻼ ﺳــﮓ ،ﺗﻤﺮﻛﺰ ﻣﻰﻛﻨﻴﻢ .ﻛﻮﺩﻙ ﺗﺼﺎﻭﻳﺮ ﺍﻧﻮﺍﻉ ﻣﺨﺘﻠﻒ ﺳــﮓ ﺭﺍ ﻣﻰﺑﻴﻨﺪ ﻭ ﺩﺭ ﻛﻨــﺎﺭ ﺍﻃﻼﻋﺎﺕ ﻭﺭﻭﺩﻯ )ﺗﺼﺎﻭﻳﺮ ﻭ ﺻﺪﺍ( ﺑﺮﺍﻯ ﻫﺮ ﻧﻤﻮﻧﻪ ،ﺑﻪ ﺍﻭ ﮔﻔﺘﻪ ﻣﻰﺷﻮﺩ ﻛﻪ ﺍﻳﻦ ﺍﻃﻼﻋﺎﺕ ﻣﺮﺑﻮﻁ ﺑﻪ ﻳﻚ ﻧﻮﻉ »ﺳﮓ« ﻫﺴﺖ ﻳﺎ ﺧﻴﺮ. ﺑﺪﻭﻥ ﺍﻳﻨﻜﻪ ﺑﻪ ﺍﻭ ﮔﻔﺘﻪ ﺷــﻮﺩ ،ﺳﻴﺴــﺘﻢ ﻣﻐﺰ ﺍﻭ ﺍﻃﻼﻋــﺎﺕ ﻭﺭﻭﺩﻯ ﺭﺍ ﺗﺠﺰﻳﻪ ﻭ ﺗﺤﻠﻴﻞ ﻣﻰﻛﻨﺪ ﻭ ﺑﻪ ﻳﺎﻓﺘﻪﻫﺎﻳﻰ ﺩﺭ ﺯﻣﻴﻨﻪ ﻫﺮ ﻳﻚ ﺍﺯ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻯ ﻭﺭﻭﺩﻯ ﺍﺯ ﻗﺒﻴﻞ »ﺭﻧﮓ ،ﺍﻧﺪﺍﺯﻩ، ﺻﺪﺍ ،ﺩﺍﺷﺘﻦ ﭘﻨﺠﻪ ﻳﺎ ﺳﻢ ﻳﺎ ﺷﺎﺥ« ﻣﻰﺭﺳﺪ .ﭘﺲ ﺍﺯ ﻣﺪﺗﻰ ﺍﻭ ﻗﺎﺩﺭ ﺧﻮﺍﻫﺪ ﺑﻮﺩ ﻳﻚ »ﻧﻮﻉ ﺟﺪﻳﺪ« ﺍﺯ ﺳﮓ ﺭﺍ ﻛﻪ ﻗﺒ ً ﻼ ﻫﺮﮔﺰ ﻧﺪﻳﺪﻩ ﺍﺳﺖ ﺷﻨﺎﺳﺎﻳﻰ ﻛﻨﺪ. ﺗﻌﺪﺍﺩ ﺳــﻠﻮﻝﻫﺎﻯ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﺑﺴــﺘﻪ ﺑﻪ ﺗﻌﺪﺍﺩ ﻭﺭﻭﺩﻯﻫﺎ ﺍﺳﺖ .ﺩﺭ ﻋﻤﻞ ﺳﻌﻰ ﺑﺮ ﺍﻳﻦ ﺍﺳــﺖ ﻛﻪ ﻛﻠﻴﻪ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻳﻰ ﻛﻪ ﺩﺭ ﭘﺎﺳــﺦ ﺗﺄﺛﻴﺮ ﺩﺍﺭﻧﺪ ﺩﺭ ﻧﻈﺮ ﮔﺮﻓﺘﻪ ﺷﻮﻧﺪ .ﺍﻟﺒﺘﻪ ﺑﺎﻳﺪ ﺩﺭ ﻧﻈﺮ ﮔﺮﻓﺖ ﻛﻪ ﺍﻃﻼﻋﺎﺕ ﺑﻰﺍﺳــﺘﻔﺎﺩﻩ ،ﻭﺭﻭﺩﻯ ﻛﺎﺭ ﺷــﺒﻜﻪ ﺭﺍ ﻣﺸــﻜﻞﺗﺮ ﻣﻰﻛﻨﻨﺪ، ﺯﻳﺮﺍ ﺍﮔﺮ ﭼﻪ ﺷــﺒﻜﻪ ﻋﺼﺒﻰ ﺑﻪ ﻧﻮﻳﺰ) noise-ﺩﺍﺩﻩﻫﺎﻯ ﺩﺍﺭﺍﻯ ﺧﻄﺎ( ﻣﻘﺎﻭﻡ ﺍﺳــﺖ ،ﺍﻣﺎ ﺩﺭ ﻫﺮ ﺻﻮﺭﺕ ﺍﮔﺮ ﻣﻴﺰﺍﻥ ﻧﻮﻳﺰ ﺑﻴﺶ ﺍﺯ ﺣﺪ ﺯﻳﺎﺩ ﺑﺎﺷــﺪ ﻣﻤﻜﻦ ﺍﺳــﺖ ﺷــﺒﻜﻪ ﻧﺘﻮﺍﻧﺪ ﺑﻪ ﺣﺪﺍﻗﻞﺗﺮﻳﻦ ﻣﻘﺪﺍﺭ ﺧﻄﺎ )ﺻﻔﺮ( ﺑﺮﺳﺪ. ﺗﻌﺪﺍﺩ ﮔﺮﻩﻫﺎ ﺩﺭ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﺑﻪ ﭘﻴﺸﮕﻮﻳﻰ ﻣﻮﺭﺩ ﻧﻈﺮ ﻣﺎ ﺑﺴﺘﮕﻰ ﺩﺍﺭﺩ. ﻣﺜ ً ﻼ ﺍﮔﺮ ﻗﺮﺍﺭ ﺍﺳــﺖ ﻛﻪ ﺷــﺒﻜﻪ ﺍﻧﺠﺎﻡ ﻛﻨﺘﺮﻝ ﻛﻴﻔﻴﺖ ﺑﺮ ﺭﻭﻯ ﻣﺤﺼﻮﻝ ﺭﺍ ﭘﻴﺸــﮕﻮﻳﻰ ﻛﻨﺪ ،ﭘﺲ ﺩﺭ ﺍﺯﺍﻯ ﺩﺍﺩﻩﻫﺎﻯ ﻫﺮ ﻣﺤﺼﻮﻝ ،ﺩﺭ ﻣﺮﺣﻠﻪ ﻳﺎﺩﮔﻴﺮﻯ ﻳﻚ ﺳﺘﻮﻥ ﺣﺎﻭﻯ ﺻﻔﺮ ﻳﺎ ﻳﻚ ﺑﻪ ﺷــﺒﻜﻪ ﺩﺍﺩﻩ ﻣﻰﺷــﻮﺩ .ﺻﻔﺮ ﺑﻪ ﻣﻌﻨﺎﻯ ﻛﻨﺘﺮﻝ ﺍﻧﺠﺎﻡ ﻧﺸﺪﻩ ﻭ ﻳﻚ ﺑﻪ ﻣﻌﻨﺎﻯ ﻛﻨﺘﺮﻝ ﻛﻴﻔﻴﺖ ﺍﻧﺠﺎﻡ ﺷــﺪﻩ ﺧﻮﺍﻫﺪ ﺑﻮﺩ .ﺑﻪ ﺍﻳﻦ ﺗﺮﺗﻴﺐ ،ﻳﻚ ﺳــﻠﻮﻝ ﺩﺭ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﻛﺎﻓﻰ ﺧﻮﺍﻫﺪ ﺑﻮﺩ ﻛﻪ ﻓﻌﺎﻟﻴﺖ ﺁﻥ ﺑﻪ ﻣﻌﻨﻰ ﻳﻚ )ﻛﻨﺘﺮﻝ ﻛﻴﻔﻴﺖ ﺍﻧﺠﺎﻡ ﺷﺪﻩ( ﻭ ﺧﺎﻣﻮﺵ ﺑﻮﺩﻥ ﺁﻥ ﺑﻪ ﻣﻌﻨﻰ ﺻﻔﺮ )ﻋﺪﻡ ﻛﻨﺘﺮﻝ ﻛﻴﻔﻴﺖ( ﺧﻮﺍﻫﺪ ﺑﻮﺩ.
ﺑﺮﺧﻰ ﺍﺯ ﻛﺎﺭﺑﺮﺩﻫﺎﻯ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ ﺩﺭ ﻣﺒﺎﺣﺚ ﻣﺪﻳﺮﻳﺖ ﻭ ﺻﻨﺎﻳﻊ
ﺩﺳــﺘﻪﺑﻨﺪﻯ ﻭ ﺷﻨﺎﺳــﺎﻳﻰ ﺍﻟﮕﻮ :ﺍﻳﻦ ﺷــﺒﻜﻪﻫﺎ ﺑﻪ ﮔﻮﻧﻪﺍﻯ ﻃﺮﺍﺣﻰ ﺷﺪﻩﺍﻧﺪ ﻛﻪ ﻗﺎﺩﺭ ﻫﺴﺘﻨﺪ ﺍﻧﻮﺍﻉ ﺍﻟﮕﻮﻫﺎ ﺭﺍ ﺩﺳﺘﻪﺑﻨﺪﻯ ﻭ ﺍﺯ ﻫﻢ ﺗﻔﻜﻴﻚ ﻛﻨﻨﺪ . ﺩﺳﺘﻪﺑﻨﺪﻯ ﻧﻘﺎﻁ ﺧﺎﺭﺝ ﺍﺯ ﻛﻨﺘﺮﻝ ﺩﺭ ﻛﻨﺘﺮﻝ ﻛﻴﻔﻴﺖ؛ ) (DSSﺗﻔﻜﻴــﻚ ﻭ ﺩﺳــﺘﻪﺑﻨﺪﻯ ﻧﻈﺮﺍﺕ ﺧﺒﺮﮔﺎﻥ ﺍﺯ ﻋﺎﻣﻪ ﺩﺭ ﺳﻴﺴــﺘﻢ ﭘﺸــﺘﻴﺒﺎﻥ ﺗﺼﻤﻴﻢﮔﻴﺮﻯ؛ ﺩﺳﺘﻪﺑﻨﺪﻯ ﺑﻬﻴﻨﻪ ﻣﺎﺷﻴﻦﺁﻻﺕ. ﭘﻴﺶﺑﻴﻨﻰ :ﺍﻳﻦ ﮔﻮﻧﻪ ﺷــﺒﻜﻪﻫﺎ ﺑﻪ ﮔﻮﻧﻪﺍﻯ ﺁﻣﻮﺯﺵ ﺩﻳﺪﻩﺍﻧﺪ ﻛﻪ ﺑﺮ ﺍﺳﺎﺱ ﻳﺎﺩﮔﻴﺮﻯ ﻭ ﺣﻔﻆ ﺗﺠﺎﺭﺏ ،ﻗﺎﺩﺭ ﺑﻪ ﭘﻴﺶﺑﻴﻨﻰ ﺁﻳﻨﺪﻩ ﻫﺴﺘﻨﺪ. ﺷﺒﻜﻪﻫﺎﻯ ﭘﻴﺶﺑﻴﻨﻰ ﻗﻴﻤﺖ ﻧﻔﺖ ﻭ ﺑﺎﺯﺍﺭ ﺑﻮﺭﺱ؛ ﺷــﺒﻜﻪﻫﺎﻯ ﭘﻴﺶﺑﻴﻨﻰﻛﻨﻨــﺪﻩ ﺩﺭ ﻣﺒﺎﺣــﺚ ﻛﻨﺘــﺮﻝ ﻣﻮﺟﻮﺩﻯ ،ﻛﻨﺘــﺮﻝ ﻛﻴﻔﻴﺖ ﻭ ﺑﺮﻧﺎﻣﻪﺭﻳﺰﻯ ﺗﻌﻤﻴﺮﺍﺕ. ﻣﺪﻟﺴـﺎﺯﻯ :ﺍﻳﻦ ﮔﻮﻧﻪ ﺷــﺒﻜﻪﻫﺎ ﺑﻪ ﻃﻮﺭ ﮔﺴﺘﺮﺩﻩﺍﻯ ﺩﺭ ﻣﺴــﺎﺋﻞ ﺑﺮﻧﺎﻣﻪﺭﻳﺰﻯ ﺗﻮﻟﻴﺪ ﻭ Job-shop Scheduleﻭ TPSﻃﺮﺍﺣﻰ ﺷﺪﻩﺍﻧﺪ ﻛﻪ ﻗﺎﺩﺭ ﺑﻪ ﺟﺴﺘﺠﻮﻯ ﻧﻘﻄﻪ ﺑﻬﻴﻨﻪ ﺳﺮﺍﺳﺮﻯ ﺩﺭ ﺯﻣﻴﻨﻪ ﺑﺮﻧﺎﻣﻪﺭﻳﺰﻯ ﻭ ﻏﻴﺮﻩ ﻫﺴﺘﻨﺪ.
ﻣﺸﺨﺼﻪ
ﺭﻭﺵ ﻣﺤﺎﺳﺒﺎﺗﻲ ﻣﺘﺪﺍﻭﻝ )ﺷﺎﻣﻞ ﺳﻴﺴﺘﻢﻫﺎﻱ ﺧﺒﺮﻩ(
ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻲ
ﺭﻭﺵ ﭘﺮﺩﺍﺯﺵ
ﺗﺮﺗﻴﺒﻲ
ﻣﻮﺍﺯﻱ
ﺗﻮﺍﺑﻊ
ﻣﻨﻄﻘﻲ )(left brained
ﻫﻮﺵ ﻭ ﻓﺮﺍﺳﺖ (estault (right brained
ﺭﻭﺵ ﻓﺮﺍﮔﻴﺮﻱ
ﺑﻪ ﻛﻤﻚ ﻗﻮﺍﻋﺪ )(didactically
ﺑﺎ ﻣﺜﺎﻝ )(Socratically
ﻛﺎﺭﺑﺮﺩ
ﺣﺴﺎﺑﺪﺍﺭﻱ ،ﻭﺍژﻩﭘﺮﺩﺍﺯﻱ ،ﺭﻳﺎﺿﻴﺎﺕ ،ﺍﺭﺗﺒﺎﻃﺎﺕ ﺩﻳﺠﻴﺘﺎﻝ
ﭘﺮﺩﺍﺯﺵ ﺣﺴﮕﺮﻫﺎ ،ﺗﺸﺨﻴﺺ ﮔﻔﺘﺎﺭ ،ﻧﻮﺷﺘﺎﺭ ،ﺍﻟﮕﻮ
20 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127
ﺑﻬﻴﻨﻪﺳﺎﺯﻯ: ﺩﺭ ﺳﻴﺴﺘﻢﻫﺎﻯ ﻛﻨﺘﺮﻟﻰ؛ ﺩﺭ ﺳﻴﺴﺘﻢﻫﺎﻯ ﻣﺪﻳﺮﻳﺖ ،ﺗﺨﺼﻴﺺ ﻭ ﺗﺴﻬﻴﻢ ﻣﻨﺎﺑﻊ؛ ﺩﺭ ﺳﻴﺴﺘﻢﻫﺎﻯ ﻣﺎﻟﻰ ،ﻣﻰﺗﻮﺍﻥ ﺍﺯ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ )ﺑﺨﺼﻮﺹ ﺷﺒﻜﻪﻫﺎﻯ ﺑﺮﮔﺸﺘﻰ( ﺍﺳﺘﻔﺎﺩﻩ ﻛﺮﺩ.
ﺗﻔﺎﻭﺕﻫﺎﻱ ﺷـﺒﻜﻪﻫﺎﻱ ﻋﺼﺒـﻲ ﺑﺎ ﺭﻭﺵﻫﺎﻱ ﻣﺤﺎﺳـﺒﺎﺗﻲ ﻣﺘﺪﺍﻭﻝ ﻭ ﺳﻴﺴﺘﻢﻫﺎﻱ ﺧﺒﺮﻩ
ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻰ ﺗﻮﺍﻥ ﺑﺎﻻﻳﻰ ﺩﺭ ﭘﺮﺩﺍﺯﺵ ﻭ ﺁﻧﺎﻟﻴﺰ ﺍﻃﻼﻋﺎﺕ ﺩﺍﺭﻧﺪ .ﺍﻣﺎ ﻧﺒﺎﻳﺪ ﺗﺼﻮﺭ ﺷــﻮﺩ ﻛﻪ ﺷــﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ ﻣﻲﺗﻮﺍﻧﻨﺪ ﺑﺮﺍﻱ ﺣﻞ ﺗﻤﺎﻡ ﻣﺴﺎﺋﻞ ﻣﺤﺎﺳﺒﺎﺗﻲ ﻣﻮﺭﺩ ﺍﺳﺘﻔﺎﺩﻩ ﻭﺍﻗﻊ ﺷﻮﻧﺪ .ﺭﻭﺵﻫﺎﻱ ﻣﺤﺎﺳﺒﺎﺗﻲ ﻣﺘﺪﺍﻭﻝ ﻫﻤﭽﻨﺎﻥ ﺑﺮﺍﻱ ﺣﻞ ﮔﺮﻭﻩ ﻣﺸﺨﺼﻲ ﺍﺯ ﻣﺴــﺎﺋﻞ ﻣﺎﻧﻨﺪ ﺍﻣﻮﺭ ﺣﺴﺎﺑﺪﺍﺭﻱ ،ﺍﻧﺒﺎﺭﺩﺍﺭﻱ ﻭ ﻣﺤﺎﺳﺒﺎﺕ ﻋﺪﺩﻱ ﻣﺒﺘﻨﻲ ﺑﺮ ﻓﺮﻣﻮﻝﻫﺎﻱ ﻣﺸﺨﺺ ،ﺑﻬﺘﺮﻳﻦ ﮔﺰﻳﻨﻪ ﻣﺤﺴــﻮﺏ ﻣﻲﺷﻮﻧﺪ .ﺟﺪﻭﻝ ،1ﺗﻔﺎﻭﺕﻫﺎﻱ ﺑﻨﻴﺎﺩﻱ ﺩﻭ ﺭﻭﺵ ﻣﺤﺎﺳﺒﺎﺗﻲ ﺭﺍ ﻧﺸﺎﻥ ﻣﻲﺩﻫﺪ.
ﺍﺩﻋﺎ ﺷﺪﻩ ﺍﺳﺖ ﻛﻪ ﻓﺮﻛﺎﻧﺲ ﺁﺗﺶ )ﺷﻠﻴﻚ( ﻧﺮﻭﻥ ﻃﺒﻴﻌﻰ ﺑﻪ ﺻﻮﺭﺕ ﺗﺎﺑﻌﻰ ﺷﺒﻴﻪ ﺑﻪ ﺍﻳﻦ ﺗﺎﺑﻊ ﺍﺳﺖ .ﺍﻣﺎ ﺍﺯ ﺩﻻﻳﻞ ﻋﻤﺪﻩ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺍﻳﻦ ﺗﺎﺑﻊ ﺍﻳﻦ ﺍﺳﺖ ﻛﻪ ﺗﻘﺮﻳﺒﺎً ﺧﻄﻰ ،ﺍﻓﺰﺍﻳﺸﻰ ﻭ ﻣﺸــﺘﻖﭘﺬﻳﺮ ﺍﺳﺖ ﻭ ﺩﺭ ﻓﺮﻡ ﺑﺴﺘﻪ ﻗﺎﺑﻞ ﻧﻤﺎﻳﺶ ﺍﺳﺖ .ﻣﺸﺘﻖﮔﻴﺮﻯ ﺍﺯ ﺁﻥ ﺳﺎﺩﻩ ﺍﺳﺖ ﻭ ﻣﺤﺪﻭﺩﻩ ﻭﺭﻭﺩﻯ )∞ (−∞,+ﺭﺍ ﺑﻪ ﺧﺮﻭﺟﻰ ] [ 0,1ﻓﺸﺮﺩﻩﺳﺎﺯﻯ ﻣﻰﻛﻨﺪ. ﻣﺮﺣﻠﻪ .3ﺁﻣﻮﺯﺵ ﺷﺒﻜﻪ ﺍﻟﮕﻮﺭﻳﺘﻢﻫﺎﻯ ﻳﺎﺩﮔﻴﺮﻯ ،ﺭﻭﻧﺪﻫﺎﻳﻰ ﻫﺴــﺘﺪ ﻛﻪ ﺗﻮﺳــﻂ ﺁﻧﻬﺎ ﻭﺯﻥﻫﺎﻯ ﺷﺒﻜﻪ ﺗﻨﻈﻴﻢ ﻭ ﺗﻌﺪﻳﻞ ﻣﻰﺷــﻮﺩ .ﻫﺪﻑ ﺍﺯ ﺁﻣﻮﺯﺵ ﺷــﺒﻜﻪ ﺍﻳﻦ ﺍﺳﺖ ﻛﻪ ﺷﺒﻜﻪ ﻗﺎﻧﻮﻥ ﻛﺎﺭ ﺭﺍ ﻳﺎﺩ ﺑﮕﻴﺮﺩ ﻭ ﭘﺲ ﺍﺯ ﺁﻣﻮﺯﺵ ﺑﻪ ﺍﺯﺍﻯ ﻫﺮ ﻭﺭﻭﺩﻯ ،ﺧﺮﻭﺟﻰ ﻣﻨﺎﺳــﺐ ﺭﺍ ﺍﺭﺍﺋﻪ ﺩﻫﺪ .ﺗﺎﻛﻨﻮﻥ ﺑﻴﺶ ﺍﺯ 100ﻧﻮﻉ ﺍﻟﮕﻮﺭﻳﺘﻢ ﻳﺎﺩﮔﻴﺮﻯ ﺑﻪ ﻭﺟﻮﺩ ﺁﻣﺪﻩ ﺍﺳﺖ ﻛﻪ ﻣﻬﻢﺗﺮﻳﻦ ﺁﻧﻬﺎ ﺫﻛﺮ ﺷﺪ.
ﻣﺮﺍﺣﻞ ﻃﺮﺍﺣﻰ ﻳﻚ ﺷﺒﻜﻪ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ
ﻣﺮﺣﻠﻪ .1ﻃﺮﺍﺣﻰ ﻣﻌﻤﺎﺭﻯ ﺷﺒﻜﻪ ﺍﻳﻦ ﻣﺮﺣﻠﻪ ﺷــﺎﻣﻞ ﺗﻌﻴﻴﻦ ﺗﻌﺪﺍﺩ ﻻﻳﻪﻫﺎﻯ ﻣﻮﺟﻮﺩ ﺩﺭ ﺷﺒﻜﻪ ،ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥﻫﺎﻯ ﻫﺮ ﻻﻳﻪ، ﺗﻌﻴﻴﻦ ﺑﺮﮔﺸــﺖﭘﺬﻳﺮ ﺑﻮﺩﻥ ﻳﺎ ﻧﺒﻮﺩﻥ ﺷــﺒﻜﻪ ﻭ ﻏﻴﺮﻩ ﺍﺳــﺖ ﻛﻪ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻧﻮﻉ ﻣﺴﺌﻠﻪ ﺗﻌﻴﻴﻦ ﻣﻰﮔﺮﺩﺩ )ﺑﺮﺍﻯ ﻣﺜﺎﻝ ﺷــﺒﻜﻪﻫﺎﻯ ﺑﺮﮔﺸــﺘﻰ ﺩﺭ ﺍﻏﻠﺐ ﻣﻮﺍﺭﺩ ﺑﺮﺍﻯ ﻣﺴﺎﺋﻞ ﭘﻮﻳﺎ ﻛﺎﺭﺑﺮﺩ ﺩﺍﺭﻧﺪ ﻭ ﻳﺎ ﺍﻳﻨﻜﻪ ﺷﺒﻜﻪﻫﺎﻯ ﭘﺮﺳﭙﺘﺮﻭﻥ ﭘﻴﺶﺧﻮﺭ ،ﺑﺮﺍﻯ ﻧﮕﺎﺷﺖﻫﺎﻯ ﻏﻴﺮﺧﻄﻰ ﻛﺎﺭﺑﺮﺩ ﺩﺍﺭﻧﺪ(. ﻧﻜﺘﻪ ﻗﺎﺑﻞ ﺗﻮﺟﻪ ﺍﻳﻨﻜﻪ ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪ ﻭﺭﻭﺩﻯ ﺍﺯ ﺻﻮﺭﺕ ﻣﺴــﺌﻠﻪ ﻣﻮﺭﺩ ﺑﺮﺭﺳــﻰ ﻣﺸــﺨﺺ ﻣﻰﮔﺮﺩﺩ .ﺑﻪ ﻋﺒﺎﺭﺕ ﺩﻳﮕﺮ ،ﺗﺤﺖ ﺍﻧﺘﺨﺎﺏ ﻃﺮﺍﺡ ﻣﺴﺌﻠﻪ ﻧﻴﺴﺖ ﺑﻠﻜﻪ ﺑﺴﺘﮕﻰ ﺑﻪ ﺭﻭﺵ ﺣﻞ ﻣﺴــﺄﻟﻪ ﻣﻮﺭﺩ ﻧﻈﺮ ﺩﺍﺭﺩ .ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪ ﺧﺮﻭﺟﻰ ﺑﺴــﺘﮕﻰ ﺑﻪ ﻧﻮﻉ ﺟﻮﺍﺏ ﻣﺎ ﺩﺍﺭﺩ .ﺑﺮﺍﻯ ﻣﺜﺎﻝ ،ﭼﻨﺎﻧﭽﻪ ﭘﺎﺳﺦ ﻣﺎ ﺑﻪ ﺻﻮﺭﺕ ﻳﻚ ﻋﺪﺩ ﺑﺎﺷﺪ ،ﻳﻚ ﻧﺮﻭﻥ ﻛﺎﻓﻰ ﺍﺳــﺖ .ﺗﻌﺪﺍﺩ ﻻﻳﻪﻫﺎ ﻭ ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪ ﭘﻨﻬﺎﻥ ﺗﻮﺳــﻂ ﻛﺎﺭﺑﺮ ﺗﻌﻴﻴﻦ ﻣﻰﮔﺮﺩﺩ؛ ﺍﻣﺎ ﺩﺭ ﺍﻛﺜﺮ ﻣﺴــﺎﺋﻞ ﺍﺯ ﻳﻚ ﺗﺎ ﺳــﻪ ﻻﻳﻪ ﻣﻴﺎﻧﻰ ﻛﻔﺎﻳﺖ ﻣﻰﻛﻨﺪ .ﻫﻤﭽﻨﻴﻦ ﻳﻚ ﺭﻭﺵ ﻋﻤﻠﻰ ﺑﺮﺍﻯ ﺗﺨﻤﻴﻦ ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥﻫﺎﻯ ﻻﻳﻪ ﭘﻨﻬﺎﻥ ﻭﺟﻮﺩ ﻧﺪﺍﺭﺩ .ﺑﻪ ﻫﻤﻴﻦ ﺩﻟﻴﻞ ﺑﺮﺍﻯ ﺩﺳﺘﻴﺎﺑﻰ ﺑــﻪ ﻣﻘﺪﺍﺭ ﻣﻴﺎﻧﮕﻴﻦ ﺧﻄﺎﻯ ﻣﻄﻠﻮﺏ ،ﺍﺯ ﺭﻭﺵﻫﺎﻯ ﺳــﻌﻰ ﻭ ﺧﻄــﺎ )ﺩﺭ ﺣﻴﻦ ﺁﻣﻮﺯﺵ( ﺍﺳﺘﻔﺎﺩﻩ ﻣﻰﺷﻮﺩ. ﻣﺮﺣﻠﻪ .2ﺗﻌﻴﻴﻦ ﻧﻮﻉ ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ ﻣﻰﺗﻮﺍﻥ ﺑﺮﺍﻯ ﺍﻳﻨﻜﻪ ﺧﺮﻭﺟﻰ ﺧﺎﺻﻰ ﺗﻮﻟﻴﺪ ﺷﻮﺩ ،ﺍﺯ ﻳﻚ ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ ﺍﺳﺘﻔﺎﺩﻩ ﻛﺮﺩ .ﺍﻳﻦ ﺗﺎﺑﻊ ﻣﺤﺪﻭﺩﻩ ﻭﺳﻴﻌﻰ ﺍﺯ ﻣﻘﺎﺩﻳﺮ ﻭﺭﻭﺩﻯ ﺭﺍ ﺑﻪ ﻣﻘﺪﺍﺭ ﺧﺎﺻﻰ ﻧﮕﺎﺷﺖ ﻣﻰﻛﻨﺪ .ﺑﻪ ﻋﻨﻮﺍﻥ ﻣﺜﺎﻝ ﻣﻰﺗﻮﺍﻥ ﻫﺮ ﻣﻘﺪﺍﺭ ﺧﺮﻭﺟﻰ ﺭﺍ ﺑﻪ ﻣﻘﺪﺍﺭ ﺑﺎﻳﻨﺮﻯ 0ﻭ 1ﻧﮕﺎﺷﺖ ﻛﺮﺩ .ﺍﻧﻮﺍﻉ ﻣﺨﺘﻠﻔﻰ ﺍﺯ ﺍﻳﻦ ﺗﻮﺍﺑﻊ ﺩﺭ ANNﻫﺎ ﻣﻮﺭﺩ ﺍﺳــﺘﻔﺎﺩﻩ ﻗﺮﺍﺭ ﻣﻰﮔﻴﺮﺩ ،ﻭﻟﻰ ﭘﺮ ﻛﺎﺭﺑﺮﺩﺗﺮﻳﻦ ﺁﻧﻬﺎ ،ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ ﺳﻴﮕﻤﻮﺋﻴﺪ )ﻣﺎﻧﻨﺪ (sﺍﺳﺖ ﻛﻪ ﺑﻪ ﺻﻮﺭﺕ ﺯﻳﺮ ﺗﻌﺮﻳﻒ ﻣﻰﺷﻮﺩ:
ﺷﻜﻞ (7ﻧﺤﻮﻩ ﻋﻤﻠﻜﺮﺩ ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ
ﻧﺘﻴﺠﻪ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﺩﺭ ﺗﻼﺷــﻨﺪ ﺗــﺎ ﺧﺼﻮﺻﻴﺎﺕ ﺍﺳﺎﺳــﻰ ﺳــﻠﻮﻝﻫﺎﻯ ﻋﺼﺒﻰ ﻭ ﺍﺗﺼــﺎﻻﺕ ﺁﻧﻬــﺎ ﺑﺎ ﻳﻜﺪﻳﮕﺮ ﺭﺍ ﺷﻨﺎﺳــﺎﻳﻰ ﻛﻨﻨﺪ .ﺳــﭙﺲ ﺑﻪ ﻃــﻮﺭ ﻣﻌﻤﻮﻝ ﻳﻚ ﻛﺎﻣﭙﻴﻮﺗﺮ ﺭﺍ ﺑﺮﺍﻯ ﺷﺒﻴﻪﺳــﺎﺯﻯ ﺍﻳﻦ ﺧﺼﻮﺻﻴــﺎﺕ ﺑﺮﻧﺎﻣﻪﺭﻳﺰﻯ ﻣﻰﻛﻨﻨﺪ .ﺍﮔﺮ ﭼﻪ ﺑﻪ ﺩﻟﻴﻞ ﺍﻳﻨﻜﻪ ﺩﺍﻧﺶ ﻣﺎ ﺍﺯ ﺳﻠﻮﻝﻫﺎﻯ ﻋﺼﺒﻰ ﻧﺎﻗﺺ ﺍﺳﺖ ﻭ ﻗﺪﺭﺕ ﻣﺤﺎﺳﺒﺎﺕ ﻣﺎ ﻣﺤﺪﻭﺩ ﺍﺳــﺖ ،ﻣﺪﻝﻫﺎﻯ ﻣﺎ ﻟﺰﻭﻣﺎً ﺁﺭﻣﺎﻥﻫﺎﻯ ﺧﺎﻡ ﻭ ﻧﺎﻗﺼﻰ ﺍﺯ ﺷﺒﻜﻪﻫﺎﻯ ﻭﺍﻗﻌﻰ ﺳﻠﻮﻝﻫﺎﻯ ﻋﺼﺒﻰ ﺍﺳﺖ .ﺑﺎ ﺍﻳﻦ ﻫﻤﻪ ﺷﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻣﺼﻨﻮﻋﻰ ﺩﺭ ﺣﻞ ﻣﺴﺎﺋﻞ ﭘﻴﭽﻴﺪﻩ ﻛﻪ ﺩﻳﮕﺮ ﻣﺘﺪﻫﺎ ﻭ ﺭﻭﺵﻫﺎ ﻗﺎﺩﺭ ﺑﻪ ﺍﻧﺠﺎﻡ ﺁﻥ ﻧﻴﺴﺘﻨﺪ ،ﺑﺴﻴﺎﺭ ﻣﻔﻴﺪ ﻭﺍﻗﻊ ﺷﺪﻩﺍﻧﺪ ﻭ ﻛﺎﺭﺑﺮﺩ ﺁﻥ ﻧﻴﺰ ﺭﻭﺯ ﺑﻪ ﺭﻭﺯ ﺩﺭ ﺣﺎﻝ ﺍﻓﺰﺍﻳﺶ ﺍﺳﺖ .ﺑﻨﺎﺑﺮﺍﻳﻦ ﻣﺎ ﻣﻰﺗﻮﺍﻧﻴﻢ ﺑﺎ ﺍﺳــﺘﻔﺎﺩﻩ ﺍﺯ ﺷــﺒﻜﻪﻫﺎﻯ ﻋﺼﺒﻰ ﻫﺮ ﭼﻪ ﺑﻴﺸــﺘﺮ ﺑﻪ ﺷﺒﻴﻪﺳﺎﺯﻯ ﺍﻧﺴﺎﻥ ﺗﻮﺳﻂ ﻛﺎﻣﭙﻴﻮﺗﺮﻫــﺎ ﻧﺰﺩﻳﻚ ﺷــﻮﻳﻢ ،ﺑﻪ ﻣﻨﻈﻮﺭ ﻭﺍﮔﺬﺍﺭﻯ ﻛﺎﺭﻫﺎﻯ ﺗﻜــﺮﺍﺭﻯ ،ﻭﻗﺖﮔﻴﺮ ﻭ ﻣﺴﺎﺋﻠﻰ ﻛﻪ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﭘﻴﺸﺮﻓﺖ ﺑﺸﺮﻯ ﺩﻳﮕﺮ ﺩﺭﺧﻮﺭ ﺑﺸﺮ ﻧﻴﺴﺖ.
ﺷﻜﻞ :6ﺗﺎﺑﻊ ﺗﺒﺪﻳﻞ
21 ﺳﺎﻝ ﺑﻴﺴﺘﻢ ﺷﻤﺎﺭﻩ 127