KCC 2021
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
DeeperSearch: ´º·² ³×Æ®¿öÅ© Àû´ëÀû °ø°ÝÀÇ Çâ»ó°ú È®Àå |
¿µ¹®Á¦¸ñ(English Title) |
DeeperSearch: Deep, Deeper, and Wider |
ÀúÀÚ(Author) |
°¹Î¼ö
¸¶ÀÌ Åü Áî¾û
¾ÆÀÌÅçŲ ¹ÙÀ̱¸Å¸³ë¹Ù
»êÀÚ¸£º¤ ¶óÅ©¸ð³ëÇÁ
Min-Soo Kang
Tung Duong Mai
Aitolkyn Baigutanova
Sanjarbek Rakhmonov
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 48 NO. 01 PP. 2025 ~ 2027 (2021. 06) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Blackbox adversarial attack poses huge security risks to AI-based software. However, current researches usually focus on computer vision - a well-explored domain and often assumes ideal, somewhat impractical setting. In this work, we attempted to generate adversarial reaction using a simple search-based method called DeepSearch. We then test the method on three more challenging but realistic settings and expand the method into audio domain. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|