Defense against Artificial Intelligence Hacking

AI and deep learning is influential and pervasive in our daily lives with the potential to revolutionize various facts of society. We cannot manipulate or trick this advanced technology easily. Indeed, some crafty individuals utilize adversarial examples as a means to fool AI systems. These instances present an image specifically engineered to confuse our models while sidestepping noticeable alteration from a human perspective. Essentially very minor changes can be made which may go unnoticed by humans but lead the DL model into falsely identifying objects at high confidence level levels imagine if self-driving vehicles began interpreting the stop signs. In this project we used AlexNet as a convolutional neural network. We can load a pretrained version of the network trained on more than a million images from the our database. It showcases exceptional skill at identifying 1000 distinct categories including common place objects such as mice, pencils and keyboards alongside a variety of animals thus exhibiting remarkable prowess in recognizing images. Our objective involves in developing resilient models capable for enough counterattacks efficiently from those manipulating images adversely geared toward attacking it.