Zi-Liang Wang

A metric interaction framework that models object-level and scene-level interactions in physical Cartesian space at a shared metric scale and improves diverse VLA and WAM baselines with a small number of additional parameters and training steps is introduced.

Inspired by iterative error feedback in structured prediction, FAPR represents the current segmentation mask as a dynamic failure state and models each repair operation as a state-transition operator, enabling later operations to adapt to preceding changes.

Despite achieving 84.8% accuracy, multimodal models struggle with long-horizon reasoning in electrical circuits, revealing critical gaps in their understanding of physical conventions.

E^3RL not only overcomes the autoregressive curse but also enhances LLMs' reasoning capabilities, achieving up to 6.514% better performance than previous state-of-the-art models.

Semantic ID quality hinges on a delicate balance of robustness and fidelity, with the new DRQ method offering a fresh lens on tokenizer performance.

LLMs can significantly boost their emotional intelligence simply by role-playing conversations with themselves, iteratively refining their ability to both recognize and express emotions.

Generative recommendation models can match the expressiveness of discriminative models by explicitly incorporating item attribute information during sequence decoding, leading to substantial performance gains.