Search papers, labs, and topics across Lattice.
1
0
2
Ditching learned positional embeddings for a gated, Manhattan-distance decay boosts vision transformer accuracy by 6.5% on Mini-ImageNet.
