Weiwen Liu

SkillGate achieves a 30% boost in trial success for long-horizon agents by fundamentally rethinking how skill selection is rewarded during execution.

SAGE not only matches professional directors in storyboard quality but also slashes authoring time, revolutionizing automated narrative production.

VAD reveals that isolating visual evidence can dramatically improve target reconstruction in multimodal learning, leading to more accurate student outputs.

SkillRise achieves up to 8.5 percentage points better performance than leading methods by effectively reusing transferable skills across related tasks.

LatentSkill achieves a 21.4-point increase in task success while slashing prefill token usage by over 64%, revolutionizing how LLM agents utilize skills.

Stop wasting idle compute: ProAct agents anticipate user needs and proactively gather information, slashing task completion time and hallucinations.

LLMs can be forced to generalize beyond initial constraints by actively searching for adversarial test cases that expose logical divergences in generated code.

LLM agent progress increasingly hinges on better external cognitive infrastructure, not just stronger models.

Synergy's architecture lets agents evolve through experience by proactively recalling rewarded trajectories, hinting at a new way to build agents that learn and adapt in open, collaborative environments.

LMM-based GUI agents stick out like a sore thumb in human-centric mobile environments, but simple techniques can make them blend in without sacrificing utility.

By steering token selection at the logit level, LogitsCoder achieves more efficient and higher-quality reasoning chains for code generation, outperforming existing methods.

GUI agents learn faster and generalize better with a new reward shaping technique that dynamically adapts to successful exploration trajectories, outperforming fixed reward schemes.