KRnet 2004

INDEX

  • Àλ縻
  • Á¶Á÷Á¶Á÷
  • ÇÁ·Î±×·¥ÇÁ·Î±×·¥
  • µî·Ï/Çà»ç¾È³»µî·Ï/Çà»ç¾È³»
  • °Ô½ÃÆÇ °Ô½ÃÆÇ
  • Past KRnet
  • ¼¼ºÎÇÁ·Î±×·¥

    ¼¼ºÎÇÁ·Î±×·¥

     

    [Keynote Speech II] Learning to Make Decisions Optimally
    °ü¸®ÀÚ (krnet) ÀÛ¼ºÀÏ : 2020-05-08 12:21:37 Á¶È¸¼ö : 309
    ÄÚµå¹øÈ£ : 34
    ¹ßÇ¥ÀÚ : Á¤ ¼Û
    ¼Ò¼Ó : KAIST
    ºÎ¼­ : AI ´ëÇпø
    Á÷À§ : ´ëÇпøÀå
    ¼¼¼Ç½Ã°£ : 11:00~11:30
    ¹ßÇ¥ÀÚ¾à·Â : 1. ÇÐ ·Â
    ¡Û 1988.2: ¼­¿ï´ëÇб³ °ø°ú´ëÇÐ Á¦¾î°èÃø°øÇаú Çлç
    ¡Û 1990.2: ¼­¿ï´ëÇб³ °ø°ú´ëÇÐ Á¦¾î°èÃø°øÇаú ¼®»ç
    ¡Û 1995.5: ¹Ì±¹ Univ. of Texas at Austin Àü±â¹×ÄÄÇ»ÅÍ°øÇаú ¹Ú»ç
    2. °æ ·Â
    ¡Û 2019.5 ~ ÇöÀç: KAIST AI´ëÇпø ¿øÀå
    ¡Û 2019.8 ~ ÇöÀç: KAIST AI´ëÇпø ±³¼ö
    ¡Û 2000.3 ~ 2019.8: KAIST Àü±â¹×ÀüÀÚ°øÇкΠ±³¼ö
    ¡Û 1996.9 ~ 2000.2: ¼­°­´ëÇб³ ÀüÀÚ°øÇаú ±³¼ö
    ¡Û 1994.12 ~ 1996.8: ¹Ì±¹ AT&T Bell Labs, Holmdel ¿¬±¸¿ø
    ¡Û 2013.1~ 2015.12: KAIST 5G À̵¿Åë½Å ¿¬±¸¼¾ÅÍ ¼¾ÅÍÀå
    ¡Û 2015.6~ 2018.6: KAIST Àü±â¹×ÀüÀÚ°øÇкΠÄÄÇ»ÅÍ±×·ì ´ëÇ¥±³¼ö
    3. ¼ö »ó
    ¡Û ±¹Á¦Àü±âÀüÀÚ°øÇÐȸ Àª¸®¾ö º£³×Æ® »ó 2ȸ ¼ö»ó (2013³â, 2016³â)
    ¡Û Çѱ¹Åë½ÅÇÐȸ Çص¿Çмú´ë»ó ¼ö»ó (2016³â)
    ¡Û KAIST ±â¼úÇõ½Å´ë»ó ¼ö»ó (2016³â)
    °­¿¬¿ä¾à : One of the grand challenges of networking research today is to build ¡°autonomous¡± or ¡°self-driving¡± networks, where network control decisions are made in real time and in an automated fashion. Yet, building such self-driving networks that are practically deployable has largely remained unrealized due to two major obstacles known as ¡°curse of modeling¡± and ¡°curse of dimensionality¡±. In this talk, I will describe a reinforcement learning approach to overcome these obstacles using an example of wireless scheduling and demonstrate that networks can indeed learn to manage radio resources on their own, i.e., directly from experience interacting with their environment. Furthermore, networks can also learn to predict their future from experience if the network states are recurrent, thereby allowing the design of ¡°foresighted¡± wireless scheduling policies that can solve the long-standing high queueing delay problem in the state-of-the-art max-weight scheduling policy while maintaining throughput optimality.
    ¿Â¶óÀÎÇà»çÀå : https://zoom.us/j/91552578450
    ¿Â¶óÀιßÇ¥Àå :
    ¸ñ·Ïº¸±â

    TOP