Search papers, labs, and topics across Lattice.
1
0
3
5
Two-layer neural nets solve modular addition via a "flawed indicator function" built from noisy single-frequency Fourier features, where phase symmetry enables a majority-voting noise cancellation scheme.
