µö¸¶Àεå´Â ³×ÀÌÃÄ¿¡ ¾ËÆÄ°íÀÇ »õ·Î¿î ¹öÀüÀÎ ¡°¾ËÆÄ°í Á¦·Î¡±¿¡ ´ëÇÑ ³í¹®À» °ÔÀçÇß½À´Ï´Ù. ÀÌ ¹öÀüÀº À̼¼µ¹ ¹× Ä¿Á¦¿Í ´ë±¹ÇÑ ¹öÀü°ú ´Ù¸¨´Ï´Ù.
¾ËÆÄ°í Á¦·Î´Â ÀÌÀü ¹öÀü°ú ´ÙÀ½ ºÎºÐ¿¡¼ Â÷ÀÌ°¡ ÀÖ½À´Ï´Ù.
- ¾î¶°ÇÑ ±âÁ¸ (Àΰ£ ¹× ¾ËÆÄ°íÀÇ) ´ë±¹À» ·¹ÆÛ·±½º·Î »ïÁö ¾Ê½À´Ï´Ù. ÀԷ°ªÀº ¹ÙµÏÆÇ¿Í ¹éµ¹, Èæµ¹ »ÓÀÔ´Ï´Ù.
- ±âÁ¸¿¡ ¡°Á¤Ã¥°ú °¡Ä¡¡±¸¦ °¢°¢ °è»êÇÏ´ø ½Å°æ¸ÁÀ» Çϳª·Î ÅëÇÕÇÏ¿© È¿À²¼ºÀ» ³ô¿´½À´Ï´Ù.
- ±âÁ¸ ¹öÀü°ú ´Þ¸®, ·£´ýÇÏ°Ô °æ¿ìÀÇ ¼ö¸¦ µûÁ®¼ ´ÙÀ½ ¼ö¸¦ ãÁö ¾Ê½À´Ï´Ù(rollout). ¿¬»êÀº ¿À·ÎÁö ½Å°æ¸ÁÀ» ÀÌ¿ëÇØ ÀÌ·ç¾îÁý´Ï´Ù.
¾ËÆÄ°í Á¦·Î´Â ¾Æ¹«·± »çÀüÁ¤º¸ ¾øÀÌ ±¸µ¿À» ½ÃÀÛÇß½À´Ï´Ù. ±¸µ¿ ÈÄ 3Àϸ¸¿¡ À̼¼µ¹ ¹öÀüÀ»(AlphaGo Lee), 21Àϸ¸¿¡ Ä¿Á¦ ¹öÀüÀ»(AlphaGo Master) ¶Ù¾î³Ñ¾ú½À´Ï´Ù. 40ÀÏ ÈÄ, ¾ËÆÄ°í´Â ÇöÁ¸ÇÏ´Â ¸ðµç Àΰ£ ¹× ÀΰøÁö´ÉÀÇ ELO ·¹ÀÌÆÃÀ» µ¹ÆÄÇß½À´Ï´Ù.
* Ãâó:
https://deepmind.com/blog/alphago-zero-learning-scratch/