Haoyu Wang

By treating slide design as an inverse planning problem, SPIRE reveals latent design intents that traditional methods miss, leading to superior personalization outcomes.

CLIR uncovers 8x more unique bugs than existing fuzzing tools, revolutionizing compiler testing efficiency and effectiveness.

d-OPSD enables dLLMs to learn from their own future outputs, drastically improving sample efficiency and performance in reasoning tasks.

Geometry-aware reconstruction reduces hallucinations in object poses, leading to more reliable robot planning.

KVEraser achieves a 3-4x speedup over full recomputation while maintaining high performance in long-context tasks, revolutionizing how we handle context updates in LLMs.

Acoda can induce LLMs to misinterpret code analysis with a 70% success rate, transforming code security in the age of AI.

LC-QAT achieves superior performance in 2-bit quantization with just a fraction of the training data, setting a new standard for data-efficient model optimization.

UniSVQ achieves state-of-the-art performance in 2-bit quantization, outperforming traditional methods while enhancing inference speed.

Raw context outperforms compact memory designs, revealing that memory structure is crucial for effective video generation in action-conditioned models.

Ultra Flash achieves real-time high-resolution video generation at unprecedented frame rates, pushing the boundaries of what’s possible in streaming video AI.

GMBFormer achieves state-of-the-art urban green-space extraction by leveraging a global memory bank that enhances semantic reuse without compromising on visual fidelity.

Stop struggling with subjective LLM evaluation: RUBRIC-ARROW aligns models better by alternating between generating evaluation rubrics and judging against them, using only pairwise preferences.