Liang Wang

Current action-conditioned world models are limited by their reliance on visual patterns, failing to generalize physical dynamics across different robot embodiments.

Memory-augmented manipulation models can now achieve state-of-the-art performance while remaining data-efficient and generalizable across diverse tasks and environments.

SyncPlan achieves unprecedented coordination efficiency in LLM-based multi-agent systems, cutting wall-clock runtime to less than 0.05% of existing methods while maintaining high task success rates.

GRACE accelerates real-time ad retrieval by improving eligibility matching rates and drastically reducing latency, making it feasible to generate thousands of ads on demand.

EviSD achieves state-of-the-art performance in question-answering tasks by leveraging privileged evidence, outperforming existing methods while maintaining efficiency in response generation.

A simple logistic regression can match the performance of advanced models in evaluating emotion descriptions, raising questions about the validity of current multimodal benchmarks.

FlowWAM achieves a remarkable 92.94% success rate in manipulation tasks by harnessing optical flow as a video-native action representation.

Climb-settle cadence can eliminate overshoot errors in quadruped stair navigation, outperforming traditional methods even at lower loop rates.

DrivingDepth achieves state-of-the-art depth estimation by leveraging sparse LiDAR to fine-tune pixel-wise scale without sacrificing geometric coherence.

Transforming image quality assessment from a single score to a nuanced diagnosis of multiple quality issues could revolutionize smartphone ISP tuning.

Arko-T achieves superior performance in text-to-structured 3D generation while being ten times more cost-effective than leading models.

Structured supervision can boost VLA model performance by over 50% in complex robotic tasks, transforming how we approach fine-tuning in manipulation.

E-TTS achieves up to a 33.14% performance boost in robotic manipulation by leveraging historical context and iterative refinement, redefining how we approach test-time scaling.

Achieving comparable performance to full-precision models, BITEMBED slashes storage costs and enhances embedding efficiency with extreme low-bit quantization.

Robots can now navigate complex environments without continuous goal updates, relying solely on their internal spatial memory.

CRANE achieves a remarkable 96.9% Grounded Success in knowledge editing for reasoning MLLMs, overcoming traditional failure modes that plague existing methods.

Personalizing LLMs through a sociologically grounded framework reveals the hierarchical nature of user behavior, leading to significant performance gains across tasks.

EAPO enables agents to learn when to forgo tool use, achieving a remarkable 10.45% performance boost while slashing tool calls by over 18%.

Achieve state-of-the-art results in agentic knowledge base question answering by distilling gold-action policies into on-policy student rollouts, bridging the gap between sparse rewards and weakly supervised intermediate actions.

Ditch the multi-camera setup: VERM leverages foundation models to synthesize a single, task-optimized "virtual eye" view for robots, slashing training time by 1.89x.