NormDetect is proposed, which anchors a learned per-section head to the normalized zero-shot score, and a blend weight chosen on training folds under a human FPR cap combines the two scores, and one threshold is calibrated on human text only.
Multimodal applications such as educational assistants and medical diagnosis demand high concurrency and low latency, and are increasingly served by edge inference enhanced with Retrieval-Augmented Generation (RAG). However, heterogeneous multimodal models and fragmented edge resources couple three problems that prior...
Zhi Yao, Zhi-Qing Tang, Han-Shuai Cui et al.· IEEE Transactions on Paralle...· 0 citations
FeatFix is introduced, a local exact-feature correction method for cached diffusion inference that replaces the complete draft block output with the exact output computed from the same incoming state, avoiding token- or channel-level partial replacement and full-timestep recomputation.
Han-Shuai Cui, Zhiqing Tang, Z. Yao et al.· arXiv.org· 0 citations
MemTxn is a governance layer outside the answer model that verifies whether an update is supported by its source and restores the application-visible state after a fault, and achieves the highest average F1 across all twelve answer-model configurations.
Han-Shuai Cui, Zhiqing Tang, Z. Yao et al.· arXiv.org· 2 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.