Tianying Liu

DeltaGate achieves a remarkable boost in recommendation accuracy for returning users after long inactivity, outperforming established models with significantly fewer parameters.

High aliasing rates in token mappings can be misleading; SIDInspector reveals that deterministic controls outperform learned exports in prefix alignment, challenging conventional wisdom in tokenizer evaluation.

Generative recommenders get a major upgrade: HPGR leverages hierarchical pre-training and sparse attention to dramatically improve performance and efficiency by explicitly modeling the structure of user behavior.