We develop technologies to keep people safe on social network platforms and connect people to what matters most to them. This means investing broadly in the areas of Natural Language Processing, Computer Vision, and Machine Learning - specifically multilingual/multimodal understanding, misinformation, tampering, entity detection, and semi-supervised learning.
This paper develops an approach that makes image-recognition systems robust against adversarial images by introducing a novel feature-denoising layer in convolutional networks, and training these networks using adversarial training. The work described in this paper formed the basis for our winning entry in the CAAD Adversarial Image Defense Competition 2018.
Kaiming He, Yuxin Wu, Laurens van der Maaten, Alan Yuille, Cihang Xie
The paper studies approaches that make image-recognition systems more robust against adversarial images by applying pre-processing transformations on the input images.
Chuan Guo, Mayank Rana, Moustapha Cisse, Laurens van der Maaten