Yuxuan Liu

Multilingual moral reasoning can be dramatically improved by grounding decisions in culturally specific contexts and theoretical frameworks, leading to significant performance gains.

Learnable observation front ends can dramatically improve low-rate speech reconstruction by capturing richer semantic information than traditional sampling methods.

MBD-LMs achieve a 2x increase in decoding throughput while maintaining high accuracy, revolutionizing the efficiency of diffusion-based text generation.

Qwen-AgentWorld achieves unprecedented simulation fidelity, outperforming existing models and enabling scalable agentic reinforcement learning across diverse real-world environments.

Task success rates for agentic phone use soar from 36.67% to 45.33% through a novel combination of real and mock environments in training.

CCPL boosts few-shot CLIP adaptation performance by anchoring prompts to frozen concept prototypes, revealing that concept constraints can enhance transferability to unseen classes.

ScaffoldAgent's utility-guided approach ensures that outlines evolve dynamically, leading to superior long-form report generation and improved factual accuracy.

Coding agents struggle to create complete and engaging games, with top performers barely reaching 41.46% success in end-to-end game generation.

Filtering out high-frequency tokens from LLM embeddings can significantly enhance their semantic quality and downstream performance.

Forget hand-crafting mobile benchmarks – PhoneWorld lets you automatically generate them from real-world GUI trajectories, leading to massive performance gains for phone-use agents.