Natural language processing and 2D vision models have attained remarkable proficiency on many tasks primarily by escalating the scale of training data.

Neural Information Processing Systems http://nips.cc/

Segment Anything Model (SAM) has recently gained much attention for its outstanding generalization to unseen data and tasks. Despite its promising prospect, the vulnerabilities of SAM, especially to universal adversarial perturbation (UAP) have not been thoroughly investigated yet. In this paper, we propose Dark-SAM, the first prompt-free universal attack framework against SAM, including a semantic decoupling-based spatial attack and a texture distortion-based frequency attack. We first divide the output of SAM into foreground and background. Then, we design a shadow target strategy to obtain the semantic blueprint of the image as the attack target.

The user selects a workload matrixArepresenting a desired first-order optimization method.

OpenAI tapped for voice control tech in U.S. drone swarm trial OpenAI has partnered with two defense technology companies that the Pentagon has selected to compete to develop voice-controlled, drone swarming software for the U.S. military, according to multiple people familiar with the matter. OpenAI has partnered with two defense technology companies that the Pentagon has selected to compete to develop voice-controlled, drone swarming software for the U.S. military, according to multiple people familiar with the matter. OpenAI's technology would only be used to translate voice commands from battlefield commanders to digital instructions for the drones, according to two of the people. It wouldn't be used for the operation of the drone swarm, weapons integration or targeting authority, the two people said. All of the people asked not to be named to discuss sensitive matters that aren't public.