Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
Self-distillation in CTR prediction simplifies to a logit-averaged loss plus a disagreement penalty, enabling sample-free training and boosting single-tower accuracy.