Search papers, labs, and topics across Lattice.
Independent Researcher
1
0
Dimensionality collapse is a key precursor to grokking, and manipulating representation geometry can dramatically speed up generalization in neural networks.
