Inference-time scaling is often treated as a token-budget problem, but a failed coding agent can seek qualitatively different resources: computation, information, or feedback. We introduce Banzhaf Resource Attribution (BRA), a controlled factorial framework that measures each resource's marginal contribution to repair...
Saiyue Lyu, Jerrold Huang, Yun-Bei Zhang et al.· 0 citations
Semantics-Preserving Quantization (SPQ), which gives more calibration weight to concept-sensitive parameters, preserves residual features, and corrects the mean error in video memory, is proposed.
Xi Xiao, Yun-Bei Zhang, Lin Zhao et al.· 0 citations
This work studies the challenges of preventing unsafe behavior from persisting and enabling recovery when failures occur through a controlled testbed of stateful authorization tasks, where fixed LLM editors optimize executable agent components and independent traces record their effects.
Yun-Bei Zhang, Janet Wang, Saiyue Lyu et al.· 1 citation
Three input conditions are used: the original image, question-blind structure extracted by a vision-language model, or gold structure derived from the diagram source, or gold structure derived from the diagram source, using the same solver model and generation settings.
Yun-Bei Zhang, Janet Wang, J. Hamm et al.· arXiv.org· 0 citations
Multi-agent systems derive their capabilities from sharing evidence, delegating tasks, and combining information across agents. The same process creates a safety problem: contributions that are admissible in isolation can jointly enable a prohibited use. Blocking every sensitive action avoids disclosure but defeats the...
Yun-Bei Zhang, Saiyue Lyu, Janet Wang et al.· 0 citations
DeepJEPA is introduced, a weight-tied joint-embedding predictive world model that treats transition depth as an inner test-time scaling axis and learns when another recurrent update is worth computing for each candidate and rollout step.
Zi-Jian Jin, Yun-Bei Zhang, Yuan-Zhe Liu et al.· 1 citation
These results support a state-based view of robot memory: effective memory for control is not simply retained visual history, but compact task-relevant state derived from the interaction history.
Yu-You Zhang, Yun-Bei Zhang, Miao Li et al.· arXiv.org· 1 citation
A mechanistic analysis of the VLM model's internal attention patterns shows that a simple describe-then-decide prompting strategy increases vision attention by 30-40% during generation, and task-specific fine-tuning improves dermatology classification but reduces cross-domain medical question-answering performance in t...
Janet Wang, Yun-Bei Zhang, Xiao Wang et al.· arXiv.org· 0 citations
Robots must often continue a task after a target moves, the viewpoint shifts, or an obstacle appears, even though their earlier observations and committed actions reflect the previous scene. Many simulation robustness benchmarks fix external conditions at reset, leaving this temporal challenge underexamined. We introdu...
Yun-Bei Zhang, Zi-Jian Jin, Yuan-Zhe Liu et al.· arXiv.org· 3 citations· ⚡1
ForGE is proposed, a unified framework for adapting frozen models through per-query routing over a joint action space that spans both support form and thinking depth and improves accuracy at 42-45% lower token cost on both main hosts.
Xi Xiao, Yun-Bei Zhang, Chen Liu et al.· 0 citations
Inspired by the Information Bottleneck principle, Prompted Information Bottlenecks (PIB) is introduced, a framework that regularizes layer-wise compression-sufficiency trade-offs and promotes a more coherent cross-layer information path.
Yuqi Li, Xi Xiao, Yun-Bei Zhang et al.· arXiv.org· 10 citations
This survey presents feedback-driven Test-Time Intelligence (TTI) as a unified perspective for understanding deployment-time improvement and relates test-time adaptation, test-time learning, and test-time scaling, highlighting both their distinctions and their growing overlap in hybrid systems.