Ass-Network: Adversarial Sample Synthesis Network (A Big Announcement of ASS-Net) Hello from Tokyo & NY. We are highly honored to announce that we have successfully completed the research paper on the revolutionary generative neural network model. Surprisingly, this DL architecture works without electricity. This is a totally innovative green technology, too. A-Net will save millions of people's lives in the long term We will present this at Fake-CVRP and Fake-NIPS this year. Collaboration offers are super welcome. Please comment or message me! Title: ASS-Network: Adversarial Sample Synthesis Network Abstract We present a new Generative Adversarial Network Architecture for perspective augmented image samples which can effectively discriminate over physical attacks and recover contextual information through our dual-stage GAN architecture. Using a novel method known as Adversarial Sample Synthesis, our sampling strategy trains on pairs of images where we utilize an adversarial sub-sampling approach to effectively learn the divergence between the fake and real image distributions. Our methods have shown to effectively discriminating between fakes and real images while also doubling as a generative model which can perform an inferential recovery process to predict the entire context of the image. Our model requires very little hyper-parameter tuning and converges relatively fast, thus yielding a low-cost for training and saving energy. Although our results show great promise, we will not release the underlying model to consider ethics and surrounding consequences that may emerge from the misuse of this model. Full contents URL: https://www.team-ai.com/a-net Best, Daisuke & Yuya & Ian |