Abstract: Neural networks assume training and test data share the same distribution and label space, but real-world violations degrade performance when domain and category shifts occur simultaneously.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results