Search papers, labs, and topics across Lattice.
1
0
3
Provably undetectable backdoors can be injected into pre-trained image classifiers, even with white-box access, by exploiting sparse perturbations and Gaussian dithering.
