Natural Backdoor Attacks on Deep Neural Networks via Raindrops

<div>The effect of different injection rates <svg height="6.1673pt" id="M87" style="vertical-align:-0.2063904pt" version="1.1" viewbox="-0.0498162 -5.96091 6.71534 6.1673" width="6.71534pt" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink"><g transform="matrix(.013,0,0,-0.013,0,0)"><path d="M484 406C484 429 465 448 438 448C400 448 352 405 302 356L156 213L191 373C195 391 197 404 197 416C197 436 191 448 178 448C151 448 83 406 23 350L37 328C67 350 101 369 109 369C114 369 115 362 111 344L29 -5L34 -12C63 -5 86 -1 111 3C124 73 137 138 147 161C165 180 189 202 206 217C215 194 249 111 275 57C304 -1 319 -12 342 -12C368 -12 405 -5 471 92L454 115C427 80 395 55 384 55C374 55 364 58 341 106L262 274C282 296 367 378 414 378C427 378 440 370 444 367C449 363 454 362 462 368C476 379 484 395 484 406Z"></path></g></svg> on our method.</div>

Security and Communication Networks

Natural Backdoor Attacks on Deep Neural Networks via Raindrops

Figure 3