85e42ea64f4d6fdebb88e5360025de2ec4e34ea0,art/attacks/fast_gradient.py,FastGradientMethod,generate,#FastGradientMethod#Any#,111

Before Change

            x_adv = self._compute(x, y, self.eps, self.eps, self.random_init)

        adv_preds = np.argmax(self.classifier.predict(x_adv), axis=1)
        if self.targeted:
            rate = np.sum(adv_preds == np.argmax(y, axis=1)) / x_adv.shape[0]
        else:
            rate = np.sum(adv_preds != np.argmax(y, axis=1)) / x_adv.shape[0]
        logger.info("Success rate of FGM attack: %.2f%%", rate)

        return x_adv

After Change

                if rate > rate_best or adv_x_best is None:
                    rate_best = rate
                    adv_x_best = adv_x.copy()

        logger.info("Success rate of FGM attack: %.2f%%", rate_best)

        return adv_x_best

Instances