Yidong Chen

PAMT reveals that aligning translation processes with rewards can significantly enhance multi-domain machine translation performance, especially in complex scenarios.

TARL revolutionizes memory management for long-term agents by enabling nuanced updates that significantly enhance state recovery and reduce corruption.

Achieving near-FP16 quality with 4-bit quantization, this method enhances motion smoothness while maintaining performance on large models.