• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ÇÐȸÁö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ÇÐȸÁö > µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) ´ÙÁß ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ À¯ÇüÀ» È°¿ëÇÑ µö·¯´× ½Å°æ¸Á ¼³°è¿Í Àû¿ë
¿µ¹®Á¦¸ñ(English Title) A Deep Neural Network Design Method using Multiple Sensor Stream Data Types and its Application
ÀúÀÚ(Author) ¹Ú»ó¾Æ   ¿À¼Ò¿¬   À̹̰栠 À̹μö   Sanga Park   Soyeon Oh   Mikyeong Lee   Minsoo Lee  
¿ø¹®¼ö·Ïó(Citation) VOL 38 NO. 01 PP. 0016 ~ 0037 (2022. 04)
Çѱ۳»¿ë
(Korean Abstract)
ÃÖ±Ù »ç¹°ÀÎÅͳÝÀÇ ¹ß´Þ·Î, ¹æ´ëÇÑ ¼¾¼­ µ¥ÀÌÅ͵é·ÎºÎÅÍ »õ·Î¿î °¡Ä¡¸¦ ã¾Æ³»´Â ½ºÆ®¸² µ¥ÀÌÅÍÀÇ ºÐ¼®ÀÌ Áß¿äÇØÁö°í ÀÖ´Ù. ¶ÇÇÑ, ´ë¿ë·® µ¥ÀÌÅÍÀÇ ÇнÀÀÌ ¿ëÀÌÇØÁü¿¡ µû¶ó, µö·¯´× ±â¹ýÀ» Àû¿ëÇÏ¿© ½ºÆ®¸² µ¥ÀÌÅ͸¦ ºÐ¼®ÇÏ´Â ¿¬±¸µéÀÌ ÁøÇàµÇ°í ÀÖ´Ù. ±âÁ¸ÀÇ ¿¬±¸¿¡¼­ ºÐ·ù ¶Ç´Â ȸ±Í¿Í °°Àº ºÐ¼® À¯ÇüÀ» °í·ÁÇÏ¿© µö·¯´× ½Å°æ¸ÁÀ» ¼³°èÇÏ°í ÀÖÁö¸¸, µ¿ÀÏÇÑ À¯Çü¿¡ ´ëÇؼ­µµ ´Ù¾çÇÑ ±¸Á¶ÀÇ ½Å°æ¸ÁÀÌ Á¸ÀçÇÑ´Ù. ¶ÇÇÑ, ¿©·¯ ½ºÆ®¸² µ¥ÀÌÅ͸¦ ÇÔ²² ºÐ¼®ÇÏ´Â °æ¿ì °¢ µ¥ÀÌÅ͸¶´Ù ÁÖ¸ñÇØ¾ß ÇÒ Æ¯Â¡ÀÌ ´Ù¸¦ ¼ö ÀÖ´Ù. ±×·¯³ª, ÀÌ·¯ÇÑ »óȲ¿¡¼­ Àû¿ëÇÒ ¼ö ÀÖ´Â ½Å°æ¸Á ¼³°è ±â¹ý¿¡ ´ëÇÑ ÀÏ°üµÈ ¹æ¹ý·ÐÀÌ Á¸ÀçÇÏÁö ¾Ê´Â´Ù. µû¶ó¼­, º» ³í¹®¿¡¼­´Â ´ÙÁß ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ¿¡ ´ëÇØ, °¢ ½ºÆ®¸² µ¥ÀÌÅÍ°¡ °¡Áø Ư¼º¿¡ ±â¹ÝÇÏ¿© À¯ÇüÀ» ±¸ºÐÇÒ ¼ö ÀÖ´Â ±ÔÄ¢À» Á¦½ÃÇÑ´Ù. ¶ÇÇÑ, ÀÌ·¯ÇÑ À¯ÇüÈ­ ±ÔÄ¢¿¡ ±â¹ÝÇÑ ÀÏ°üµÈ µö·¯´× ½Å°æ¸Á ¼³°è ±â¹ýÀ» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÏ´Â ±â¹ýÀ» ÇコÄÉ¾î ºÐ¾ß¿¡ »ç¿ëµÇ°í ÀÖ´Â ´Ù¾çÇÑ ¹ÙÀÌ¿À¼¾¼­ µ¥ÀÌÅͼ¿¡ ´ëÇØ ½ÇÇèÇÏ¿©, ±âº» ¸ðµ¨ ´ëºñ Æò±Õ Á¤È®µµ´Â 80.93%¿¡¼­ 92.76%·Î Çâ»óµÇ¸ç, ÇнÀ½Ã°£ÀÇ °æ¿ì 1081.27ÃÊ¿¡¼­ 200.32ÃÊ·Î ´ÜÃàµÊÀ» º¸¿´´Ù. À̸¦ ÅëÇØ Á¦¾ÈÇÏ´Â ±â¹ýÀÌ ´ÙÁß ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅ͸¦ ºÐ¼®Çϱâ À§ÇÑ µö·¯´× ½Å°æ¸ÁÀ» ¼³°èÇÏ´Â µ¥ È¿°úÀûÀÓÀ» È®ÀÎÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
With the development of the Internet of Things(IoT), it is becoming important to analyze stream data to find new values from massive sensor data. As the learning of large-capacity data becomes easier, various studies for analyzing stream data by applying a deep learning technique are being conducted. Although neural networks are designed for analyzing stream data for the types of analysis such as classification or regression, various structures of neural networks can exist for the same type. Also, when multiple stream data is analyzed together, meaningful characteristics may vary. However, no applicable consistent methodology for neural network design exists. Therefore, we present a rule to classify types based on the characteristics of each stream data. In addition, we propose a consistent deep learning neural network design technique based on these rules. We conducted experiments applying the proposed method to the various biosensor datasets used in healthcare research. The experiments show improvement from 80.93% to 92.76% in average accuracy, and a decrease from 1081.27 seconds to 200.32 seconds in training time compared to the basic model. Finally, it was confirmed that our method is effective in designing a deep learning neural network to analyze multi-sensor stream data.
Å°¿öµå(Keyword) µö·¯´×   ´ÙÁß ¼¾¼­ ½ºÆ®¸² µ¥ÀÌÅÍ   Io   Deep Learning   Multi Sensor Stream Data  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå