Zhen Wang

CrossUAV is introduced, a UAV benchmark for joint object detection and instance segmentation that provides a unified evaluation platform for cross-granularity task collaboration and Cross-Granularity Socialized Collaboration is proposed, a progressive and adaptive framework that regulates when, where, and how tasks exchange information across network hierarchies.

LISA accelerates training and enhances output quality in visual-condition generation by aligning side network features with likelihood scores, all without extra inference costs.

Memory recovery in LLM agents is not just a byproduct of task success; it's a distinct capability that remains underexplored, with current models showing only moderate performance in reconstructing user states.

Task-conditioned collaboration outperforms traditional fusion methods, yielding significant performance gains in air-ground perception tasks.

"Lost at the End" reveals a striking primacy bias in multimodal KB-VQA, where early context dramatically outperforms late context, challenging conventional evaluation metrics.

Training-time data augmentation can slash validation loss in language model pretraining, making it feasible to train effectively on limited datasets for hundreds of epochs.

LLMs can boost K-12 writing skills, but over-reliance on AI feedback hits a ceiling, revealing the sweet spot for human-AI collaboration in education.

Achieve high-fidelity image editing without sacrificing source fidelity by straightening the latent trajectory and adaptively blending source and target velocities.