Search papers, labs, and topics across Lattice.
1
0
2
15
ReLU networks trained with gradient descent surprisingly converge to near minimum-l2-norm solutions in high dimensions, even without orthogonal data.
