arXiv subjects

Zhiyang Su

Publications and source records attributed to Zhiyang Su.

6 recordsLinked to original sources

arXiv2026

Federated Heterogeneous Language Model Optimization for Hybrid Automatic Speech Recognition

Training automatic speech recognition (ASR) models increasingly relies on decentralized federated learning to ensure data privacy and accessibility, producing multiple local models that require effective merging. In hybrid ASR systems, while acoustic models can be merged using established methods, the language model (LM) for rescoring the N-best speech recognition list faces challenges due to the heterogeneity of non-neural n-gram models and neural network models. This paper proposes a heterogeneous LM optimization task and introduces a match-and-merge paradigm with two algorithms: the Genetic Match-and-Merge Algorithm (GMMA), using genetic operations to evolve and pair LMs, and the Reinforced Match-and-Merge Algorithm (RMMA), leveraging reinforcement learning for efficient convergence. Experiments on seven OpenSLR datasets show RMMA achieves the lowest average Character Error Rate and better generalization than baselines, converging up to seven times faster than GMMA, highlighting the paradigm's potential for scalable, privacy-preserving ASR systems.

cs.CL↗

arXiv2026

CiteLLM: An Agentic Platform for Trustworthy Scientific Reference Discovery

Large language models (LLMs) have created new opportunities to enhance the efficiency of scholarly activities; however, challenges persist in the ethical deployment of AI assistance, including (1) the trustworthiness of AI-generated content, (2) preservation of academic integrity and intellectual property, and (3) protection of information privacy. In this work, we present CiteLLM, a specialized agentic platform designed to enable trustworthy reference discovery for grounding author-drafted claims and statements. The system introduces a novel interaction paradigm by embedding LLM utilities directly within the LaTeX editor environment, ensuring a seamless user experience and no data transmission outside the local system. To guarantee hallucination-free references, we employ dynamic discipline-aware routing to retrieve candidates exclusively from trusted web-based academic repositories, while leveraging LLMs solely for generating context-aware search queries, ranking candidates by relevance, and validating and explaining support through paragraph-level semantic matching and an integrated chatbot. Evaluation results demonstrate the superior performance of the proposed system in returning valid and highly usable references.

cs.CL↗

arXiv2026

RAL2M: Retrieval Augmented Learning-To-Match Against Hallucination in Compliance-Guaranteed Service Systems

Hallucination is a major concern in LLM-driven service systems, necessitating explicit knowledge grounding for compliance-guaranteed responses. In this paper, we introduce Retrieval-Augmented Learning-to-Match (RAL2M), a novel framework that eliminates generation hallucination by repositioning LLMs as query-response matching judges within a retrieval-based system, providing a robust alternative to purely generative approaches. To further mitigate judgment hallucination, we propose a query-adaptive latent ensemble strategy that explicitly models heterogeneous model competence and interdependencies among LLMs, deriving a calibrated consensus decision. Extensive experiments on large-scale benchmarks demonstrate that the proposed method effectively leverages the "wisdom of the crowd" and significantly outperforms strong baselines. Finally, we discuss best practices and promising directions for further exploiting latent representations in future work.

cs.CL↗

arXiv2025

Contextualized Token Discrimination for Speech Search Query Correction

Query spelling correction is an important function of modern search engines since it effectively helps users express their intentions clearly. With the growing popularity of speech search driven by Automated Speech Recognition (ASR) systems, this paper introduces a novel method named Contextualized Token Discrimination (CTD) to conduct effective speech query correction. In CTD, we first employ BERT to generate token-level contextualized representations and then construct a composition layer to enhance semantic information. Finally, we produce the correct query according to the aggregated token representation, correcting the incorrect tokens by comparing the original token representations and the contextualized representations. Extensive experiments demonstrate the superior performance of our proposed method across all metrics, and we further present a new benchmark dataset with erroneous ASR transcriptions to offer comprehensive evaluations for audio query correction.

cs.SD↗

arXiv2017

CeMon: A Cost-effective Flow Monitoring System in Software Defined Networks

Network monitoring and measurement are crucial in network management to facilitate quality of service routing and performance evaluation. Software Defined Networking (SDN) makes network management easier by separating the control plane and data plane. Network monitoring in SDN is relatively light-weight since operators only need to install a monitoring module into the controller. Active monitoring techniques usually introduce extra overhead into the network. The state-of-the-art approaches utilize sampling, aggregation and passive measurement techniques to reduce measurement overhead. However, little work has focused on reducing the communication cost of network monitoring. Moreover, most of the existing approaches select polling switch nodes by sub-optimal local heuristics.Inspired by the visibility and central control of SDN, we propose CeMon, a generic low-cost high-accuracy monitoring system that supports various network management tasks. We first propose a Maximum Coverage Polling Scheme (MCPS) to optimize the polling cost for all active flows. The problem is formulated as a weighted set cover problem which is proved to be NP-hard. Heuristics are presented to obtain the polling scheme efficiently and handle traffic dynamics practically. In order to balance the cost and flexibility, an Adaptive Fine-grained Polling Scheme (AFPS) is proposed as a complementary method to implement flow level measurement tasks. Three sampling algorithms are further proposed to eliminate measurement overhead while maintain high accuracy. Both emulation and simulation results show that MCPS reduces more than 50% of the communication cost with negligible loss of accuracy for different topologies and traffics. We also use real traces to demonstrate that AFPS reduces the cost by up to 20% with only 1.5% loss in accuracy.

cs.NI↗

arXiv2017

FlowCover: Low-cost Flow Monitoring Scheme in Software Defined Networks

Network monitoring and measurement are crucial in network management to facilitate quality of service routing and performance evaluation. Software Defined Networking (SDN) makes network management easier by separating the control plane and data plane. Network monitoring in SDN is lightweight as operators only need to install a monitoring module into the controller. Active monitoring techniques usually introduce too many overheads into the network. The state-of-the-art approaches utilize sampling method, aggregation flow statistics and passive measurement techniques to reduce overheads. However, little work in literature has focus on reducing the communication cost of network monitoring. Moreover, most of the existing approaches select the polling switch nodes by sub-optimal local heuristics. Inspired by the visibility and central control of SDN, we propose FlowCover, a low-cost high-accuracy monitoring scheme to support various network management tasks. We leverage the global view of the network topology and active flows to minimize the communication cost by formulating the problem as a weighted set cover, which is proved to be NP-hard. Heuristics are presented to obtain the polling scheme efficiently and handle flow changes practically. We build a simulator to evaluate the performance of FlowCover. Extensive experiment results show that FlowCover reduces roughly 50% communication cost without loss of accuracy in most cases.

cs.NI↗