Search papers, labs, and topics across Lattice.
1
0
2
5
Feature decorrelation during training not only sharpens saliency maps, but also *improves* model accuracy, challenging the conventional wisdom that interpretability comes at the cost of performance.
