Search papers, labs, and topics across Lattice.
Address:
1
0
3
Explicitly constraining class geometry with differentiable prototypes significantly boosts tensor neural network performance under extreme label scarcity, particularly through training-time regularization.
