FYI: You can trick image-recog AI into, say, detecting cats as dogs – by abusing scaling code to poison training data

FYI: You can trick image-recog AI into, say, detecting cats as dogs – by abusing scaling code to poison training data

You may want to check what you’re actually teaching your neural networks in future Boffins in Germany have devised a technique to subvert neural network frameworks so they misidentify images without any telltale signs of tampering.…

Read More…

Enter your Email Address

Leave a Reply

Your email address will not be published. Required fields are marked *