Institution profile

Zhejiang Gongshang University

Academic institutionasia · cn
Official website
Research library138linked papers
Opportunities0open roles
Selected work

Representative Papers

Anytime Safe PAC Efficient Reasoning

Jan 30, 2026

This work addresses the high computational cost and latency of large reasoning models, as well as the challenge of controlling performance loss in existing selective inference methods under online non-stationary environments. To this end, the authors propose B-PAC, a novel selective inference framework that introduces Probably Approximately Correct (PAC) safety guarantees valid at any time. By constructing a test supermartingale based on inverse propensity score estimators, B-PAC dynamically adjusts routing thresholds using accumulated statistical evidence, enabling safe and efficient online inference under partial feedback and non-stationary data distributions. Experimental results demonstrate that the method reduces model invocations by up to 81.01% while consistently ensuring that performance loss remains below a user-specified threshold.

1 citationsRead paper

Multi-Pair Temporal Sentence Grounding via Multi-Thread Knowledge Transfer Network

Dec 20, 2024arXiv.org

Existing temporal sentence grounding (TSG) methods train on untrimmed video–sentence query pairs independently, neglecting inter-pair correlations—leading to knowledge redundancy, inefficient training, and limited generalization. This paper proposes a novel multi-pair joint TSG paradigm, enabling a single model to collaboratively optimize multiple video–query pairs simultaneously. To this end, we design a multi-threaded knowledge transfer network featuring: (i) cross-modal contrastive learning to strengthen fine-grained alignment; (ii) a dual-granularity prototype matching mechanism—operating at both object/phrase level (spatial) and action/sentence level (temporal); and (iii) adaptive threshold-based hard negative mining coupled with self-supervised representation learning. Extensive experiments on multiple benchmarks demonstrate substantial improvements in both grounding accuracy and inference efficiency, achieving new state-of-the-art performance. Ablation studies confirm the effectiveness of inter-pair knowledge transfer and the model’s strong generalization capability across diverse queries and videos.

1 citationsRead paper

The First and Second Order Asymptotics of Covert Communication over AWGN Channels

May 29, 2023arXiv.org

This work investigates the asymptotic covert capacity over an additive white Gaussian noise (AWGN) channel under a KL-divergence covertness constraint δ. For n channel uses, it establishes— for the first time—the exact first- and second-order asymptotics: the first-order term is √(nδ ln e), and the second-order term is (nδ)^(1/4)(ln e)^(3/4)√2·Q⁻¹(ε). Methodologically, it introduces a novel information-geometric quasi-ε-neighborhood construction, extending the one-dimensional Gaussian minimum-KL result to n dimensions; this is combined with truncated Gaussian coding, refined KL-divergence analysis, and second-moment-constrained power optimization to achieve the optimal power scaling law. Crucially, the theoretical analysis derives, for the first time, an explicit link between the average power upper bound and the covertness parameter δ. The results establish the fundamental second-order asymptotic covert capacity benchmark for AWGN channels.

1 citationsRead paper
Recent publications

Latest Papers