Bin Wu

Standard BDMPS-Z jet quenching models miss early off-shell dynamics, but resolving finite parton virtuality reveals that the ratio of medium to jet formation times dictates where medium-induced radiation truly takes over.

Counterfactual Tucker Diffusion (CFTDiff), which integrates the treatment mask and latent Tucker structure into conditional diffusion to recover this distribution given observed control outcomes through efficient nonlinear score learning in a low-dimensional core, and establishes high-probability error bounds.

Grounding exploration trajectories in an external fact-intent DAG prevents catastrophic context drift in autonomous agents, lifting hard CTF exploitation rates from ~25% to 75%.

PrismRec revolutionizes micro-video recommendation by treating video content as a core driver of user preferences, achieving a remarkable 22.65% improvement over existing methods.

Context interference can significantly degrade the performance of search agents, but a novel context refiner shows how to enhance their reliability and efficiency dramatically.

Relying on just the final layer of LLMs can lead to a 6.72% drop in recommendation performance—IMFuse captures the full spectrum of semantic knowledge across layers for superior results.

Fact-graph orchestration enables mathematical reasoning agents to tackle complex proofs more effectively than ever before.

PAIWorld resolves critical multi-view inconsistencies in robotic manipulation, outperforming existing models and unlocking new applications in the process.

Today's LLM agents are surprisingly passive, missing opportunities to proactively learn and remember user preferences that could dramatically improve their long-term usefulness.

Unlock interdisciplinary breakthroughs: SciAtlas provides a panoramic scientific evolution network of 43M papers, enabling AI agents to navigate complex logical connections and discover non-obvious insights.

Stop wasting compute on unreliable rollouts and easy frames: Stream-R1 adaptively focuses video diffusion distillation where it matters most, boosting quality without architectural changes or added inference cost.

Semantic similarity is a poor proxy for agent performance: ranking agents based on execution-aware probing beats description-based retrieval by a wide margin.

LLMs are still wide open to jailbreaks, but this new method cuts attack success rates by nearly 5x by monitoring *intermediate* reasoning steps, not just the final output.