Search papers, labs, and topics across Lattice.
1
0
3
2
Forget full attention: Rec2PM distills long user histories into compact, interpretable "Preference Memory" tokens, slashing latency and memory while boosting accuracy in generative recommendation.
