Zheng Zhang

Adversarial training can backfire, but InsCAT ensures that detectors focus on true object features rather than misleading textures, achieving a remarkable FPR of just 1.8% in physical tests.

Achieving a 2.93x improvement in training efficiency for trillion-parameter models could redefine the scalability of large-scale AI systems.

Spatial reasoning in vision-language models can be drastically improved with a structured knowledge graph, cutting distance estimation errors by over 31%.

FlowPET not only recovers low-contrast lesions better than current methods but does so by leveraging Hamiltonian dynamics to preserve weak signals in high-noise environments.

Achieving a 100% network port opening rate, FirmCure revolutionizes the rehosting of Linux firmware by eliminating expert dependency and enhancing efficiency.

Kairos achieves top-tier performance in Physical AI while ensuring efficient state management over extended time horizons, setting a new standard for operational world models.

A unified evaluation framework for portrait composition could revolutionize how AI interprets and generates artistic images.

ALMANAC reveals that agents can significantly improve their collaborative competence by learning from detailed human mental model annotations.

Concept erasure in text-to-video diffusion models is far more effective when targeted at specific layers where concepts are naturally more separable from other information.