Real world uses of deep learning require predictable model behavior unde...
Recent advances in training vision-language models have demonstrated
unp...
Vision language models like CLIP are robust to natural distribution shif...
State-of-the-art image classifiers trained on massive datasets (such as
...