arXiv subjects

Junli Liang

Publications and source records attributed to Junli Liang.

12 recordsLinked to original sources

arXiv2026

RelayS2S: A Dual-Path Speculative Generation for Real-Time Dialogue

Real-time spoken dialogue systems face a fundamental tension between latency and response quality. End-to-end speech-to-speech (S2S) models respond immediately and naturally handle turn-taking, backchanneling, and interruption, but produce semantically weaker outputs. Cascaded pipelines (ASR -> LLM) deliver stronger responses at the cost of latency that grows with model size. We present RelayS2S, a hybrid architecture that runs two paths in parallel upon turn detection. The fast path - a duplex S2S model - speculatively drafts a short response prefix that is streamed immediately to TTS for low-latency response onset, while continuing to monitor live audio events. The slow path - a cascaded ASR -> LLM pipeline - generates a higher-quality continuation conditioned on the committed prefix, producing an uninterrupted utterance. A lightweight learned verifier gates the handoff, committing the prefix when appropriate or falling back gracefully to the cascaded pipeline. With GPT-4.1 as the back-end, RelayS2S substantially reduces response latency while preserving nearly all of the cascaded pipeline's textual quality. On synthetic voice dialogues, it achieves a P90 first-chunk latency of 81 ms, excluding TTS and network latency, compared with 1,006 ms for the cascaded baseline. On real voice dialogues, RelayS2S reduces average first-chunk latency by 479 ms while retaining 99% of the cascaded pipeline's textual quality. These benefits become larger as the slow-path model scales. Because the prefix handoff requires no architectural modification to either component, RelayS2S serves as a lightweight, drop-in addition to existing cascaded pipelines. Our code is publicly available at: https://github.com/mailong25/relays2s

cs.AI↗

arXiv2026

SentGraph: Hierarchical Sentence Graph for Multi-hop Retrieval-Augmented Question Answering

Traditional Retrieval-Augmented Generation (RAG) effectively supports single-hop question answering with large language models but faces significant limitations in multi-hop question answering tasks, which require combining evidence from multiple documents. Existing chunk-based retrieval often provides irrelevant and logically incoherent context, leading to incomplete evidence chains and incorrect reasoning during answer generation. To address these challenges, we propose SentGraph, a sentence-level graph-based RAG framework that explicitly models fine-grained logical relationships between sentences for multi-hop question answering. Specifically, we construct a hierarchical sentence graph offline by first adapting Rhetorical Structure Theory to distinguish nucleus and satellite sentences, and then organizing them into topic-level subgraphs with cross-document entity bridges. During online retrieval, SentGraph performs graph-guided evidence selection and path expansion to retrieve fine-grained sentence-level evidence. Extensive experiments on four multi-hop question answering benchmarks demonstrate the effectiveness of SentGraph, validating the importance of explicitly modeling sentence-level logical dependencies for multi-hop reasoning.

cs.CL↗

arXiv2025

GraphIF: Enhancing Multi-Turn Instruction Following for Large Language Models with Relation Graph Prompt

Multi-turn instruction following is essential for building intelligent conversational systems that can consistently adhere to instructions across dialogue turns. However, existing approaches to enhancing multi-turn instruction following primarily rely on collecting or generating large-scale multi-turn dialogue datasets to fine-tune large language models (LLMs), which treat each response generation as an isolated task and fail to explicitly incorporate multi-turn instruction following into the optimization objectives. As a result, instruction-tuned LLMs often struggle with complex long-distance constraints. In multi-turn dialogues, relational constraints across turns can be naturally modeled as labeled directed edges, making graph structures particularly suitable for modeling multi-turn instruction following. Despite this potential, leveraging graph structures to enhance the multi-turn instruction following capabilities of LLMs remains unexplored. To bridge this gap, we propose GraphIF, a plug-and-play framework that models multi-turn dialogues as directed relation graphs and leverages graph prompts to enhance the instruction following capabilities of LLMs. GraphIF comprises three key components: (1) an agent-based relation extraction module that captures inter-turn semantic relations via action-triggered mechanisms to construct structured graphs; (2) a relation graph prompt generation module that converts structured graph information into natural language prompts; and (3) a response rewriting module that refines initial LLM outputs using the generated graph prompts. Extensive experiments on two long multi-turn dialogue datasets demonstrate that GraphIF can be seamlessly integrated into instruction-tuned LLMs and leads to significant improvements across all four multi-turn instruction-following evaluation metrics.

cs.CL↗

arXiv2025

CrossLinear: Plug-and-Play Cross-Correlation Embedding for Time Series Forecasting with Exogenous Variables

Time series forecasting with exogenous variables is a critical emerging paradigm that presents unique challenges in modeling dependencies between variables. Traditional models often struggle to differentiate between endogenous and exogenous variables, leading to inefficiencies and overfitting. In this paper, we introduce CrossLinear, a novel Linear-based forecasting model that addresses these challenges by incorporating a plug-and-play cross-correlation embedding module. This lightweight module captures the dependencies between variables with minimal computational cost and seamlessly integrates into existing neural networks. Specifically, it captures time-invariant and direct variable dependencies while disregarding time-varying or indirect dependencies, thereby mitigating the risk of overfitting in dependency modeling and contributing to consistent performance improvements. Furthermore, CrossLinear employs patch-wise processing and a global linear head to effectively capture both short-term and long-term temporal dependencies, further improving its forecasting precision. Extensive experiments on 12 real-world datasets demonstrate that CrossLinear achieves superior performance in both short-term and long-term forecasting tasks. The ablation study underscores the effectiveness of the cross-correlation embedding module. Additionally, the generalizability of this module makes it a valuable plug-in for various forecasting tasks across different domains. Codes are available at https://github.com/mumiao2000/CrossLinear.

cs.LG↗

arXiv2025

TeleEval-OS: Performance evaluations of large language models for operations scheduling

The rapid advancement of large language models (LLMs) has significantly propelled progress in artificial intelligence, demonstrating substantial application potential across multiple specialized domains. Telecommunications operation scheduling (OS) is a critical aspect of the telecommunications industry, involving the coordinated management of networks, services, risks, and human resources to optimize production scheduling and ensure unified service control. However, the inherent complexity and domain-specific nature of OS tasks, coupled with the absence of comprehensive evaluation benchmarks, have hindered thorough exploration of LLMs' application potential in this critical field. To address this research gap, we propose the first Telecommunications Operation Scheduling Evaluation Benchmark (TeleEval-OS). Specifically, this benchmark comprises 15 datasets across 13 subtasks, comprehensively simulating four key operational stages: intelligent ticket creation, intelligent ticket handling, intelligent ticket closure, and intelligent evaluation. To systematically assess the performance of LLMs on tasks of varying complexity, we categorize their capabilities in telecommunications operation scheduling into four hierarchical levels, arranged in ascending order of difficulty: basic NLP, knowledge Q&A, report generation, and report analysis. On TeleEval-OS, we leverage zero-shot and few-shot evaluation methods to comprehensively assess 10 open-source LLMs (e.g., DeepSeek-V3) and 4 closed-source LLMs (e.g., GPT-4o) across diverse scenarios. Experimental results demonstrate that open-source LLMs can outperform closed-source LLMs in specific scenarios, highlighting their significant potential and value in the field of telecommunications operation scheduling.

cs.CL↗

arXiv2023

ClST: A Convolutional Transformer Framework for Automatic Modulation Recognition by Knowledge Distillation

With the rapid development of deep learning (DL) in recent years, automatic modulation recognition (AMR) with DL has achieved high accuracy. However, insufficient training signal data in complicated channel environments and large-scale DL models are critical factors that make DL methods difficult to deploy in practice. Aiming to these problems, we propose a novel neural network named convolution-linked signal transformer (ClST) and a novel knowledge distillation method named signal knowledge distillation (SKD). The ClST is accomplished through three primary modifications: a hierarchy of transformer containing convolution, a novel attention mechanism named parallel spatial-channel attention (PSCA) mechanism and a novel convolutional transformer block named convolution-transformer projection (CTP) to leverage a convolutional projection. The SKD is a knowledge distillation method to effectively reduce the parameters and complexity of neural networks. We train two lightweight neural networks using the SKD algorithm, KD-CNN and KD-MobileNet, to meet the demand that neural networks can be used on miniaturized devices. The simulation results demonstrate that the ClST outperforms advanced neural networks on all datasets. Moreover, both KD-CNN and KD-MobileNet obtain higher recognition accuracy with less network complexity, which is very beneficial for the deployment of AMR on miniaturized communication devices.

cs.LG↗

arXiv2021

Neurodynamic TDOA localization with NLOS mitigation via maximum correntropy criterion

In this paper, we exploit the maximum correntropy criterion (MCC) to robustify the traditional time-difference-of-arrival (TDOA) location estimator in the presence of non-line-of-sight (NLOS) propagation conditions. For the sake of statistical efficiency, the correntropy-based robust loss is imposed on the underlying time-of-arrival composition via joint estimation of the source position and onset time, instead of the TDOA counterpart generated in the postprocessing of sensor-collected timestamps. We then employ a neurodynamic optimization approach to tackle the highly nonconvex MCC formulation. Furthermore, we examine the local stability of equilibrium for the corresponding projection-type neural network model. Simulation investigations in representative NLOS propagation scenarios demonstrate that our neurodynamic robust TDOA localization solution is capable of outperforming several existing schemes in terms of positioning accuracy.

eess.SP↗

arXiv2020

TDOA-based localization with NLOS mitigation via robust model transformation and neurodynamic optimization

This paper revisits the problem of locating a signal-emitting source from time-difference-of-arrival (TDOA) measurements under non-line-of-sight (NLOS) propagation. Many currently fashionable methods for NLOS mitigation in TDOA-based localization tend to solve their optimization problems by means of convex relaxation and, thus, are computationally inefficient. Besides, previous studies show that manipulating directly on the TDOA metric usually gives rise to intricate estimators. Aiming at bypassing these challenges, we turn to retrieve the underlying time-of-arrival framework by treating the unknown source onset time as an optimization variable and imposing certain inequality constraints on it, mitigate the NLOS errors through the $\ell_1$-norm robustification, and finally apply a hardware realizable neurodynamic model based on the redefined augmented Lagrangian and projection theorem to solve the resultant nonconvex optimization problem with inequality constraints. It is validated through extensive simulations that the proposed scheme can strike a nice balance between localization accuracy, computational complexity, and prior knowledge requirement.

eess.SP↗

arXiv2019

Accelerated Symmetric ADMM and Its Applications in Signal Processing

The alternating direction method of multipliers (ADMM) were extensively investigated in the past decades for solving separable convex optimization problems. Fewer researchers focused on exploring its convergence properties for the nonconvex case although it performed surprisingly efficient. In this paper, we propose a symmetric ADMM based on different acceleration techniques for a family of potentially nonsmooth nonconvex programing problems with equality constraints, where the dual variables are updated twice with different stepsizes. Under proper assumptions instead of using the so-called Kurdyka-Lojasiewicz inequality, convergence of the proposed algorithm as well as its pointwise iteration-complexity are analyzed in terms of the corresponding augmented Lagrangian function and the primal-dual residuals, respectively. Performance of our algorithm is verified by some preliminary numerical examples on applications in sparse nonconvex/convex regularized minimization signal processing problems.

math.NA↗

arXiv2018

Robust MIMO Radar Target Localization based on Lagrange Programming Neural Network

This paper focuses on target localization in a widely distributed multiple-input-multiple-output (MIMO) radar system. In this system, range measurements, which include the sum of distances between transmitter and target and the distances from the target to receivers, are used. We can obtain an accurate estimated position of the target by minimizing the measurement errors. In order to make our model come closer to reality, we introduce two kinds of noises, namely, Gaussian noise and outliers. When we evaluate a target localization algorithm, its localization accuracy and computational complexity are two main criteria. To improve the positioning accuracy, the original problem is formulated as solving a non-smooth constrained optimization problem in which the objective function is either l1-norm or l0-norm term. To achieve a real-time solution, the Lagrange programming neural network (LPNN) is utilized to solve this problem. However, it is well known that LPNN requires twice-differentiable objective function and constraints. Obviously, the l1-norm or l0-norm term in the objective function does not satisfy this requirement. To address this non-smooth optimization problem, this paper proposes two modifications based on the LPNN framework. In the first method, a differentiable proximate l1-norm function is introduced. While in the second method, locally competitive algorithm is utilized. Simulation and experimental results demonstrate that the performance of the proposed algorithms outperforms several existing schemes.

eess.SP↗

arXiv2018

Robust Real-time Ellipse Fitting Based on Lagrange Programming Neural Network and Locally Competitive Algorithm

Given a set of 2-dimensional (2-D) scattering points, which are usually obtained from the edge detection process, the aim of ellipse fitting is to construct an elliptic equation that best fits the collected observations. However, some of the scattering points may contain outliers due to imperfect edge detection. To address this issue, we devise a robust real-time ellipse fitting approach based on two kinds of analog neural network, Lagrange programming neural network (LPNN) and locally competitive algorithm (LCA). First, to alleviate the influence of these outliers, the fitting task is formulated as a nonsmooth constrained optimization problem in which the objective function is either an l1-norm or l0-norm term. It is because compared with the l2-norm in some traditional ellipse fitting models, the lp-norm with p<2 is less sensitive to outliers. Then, to calculate a real-time solution of this optimization problem, LPNN is applied. As the LPNN model cannot handle the non-differentiable term in its objective, the concept of LCA is introduced and combined with the LPNN framework. Simulation and experimental results show that the proposed ellipse fitting approach is superior to several state-of-the-art algorithms.

eess.IV↗

arXiv2014

Kernel Least Mean Square with Adaptive Kernel Size

Kernel adaptive filters (KAF) are a class of powerful nonlinear filters developed in Reproducing Kernel Hilbert Space (RKHS). The Gaussian kernel is usually the default kernel in KAF algorithms, but selecting the proper kernel size (bandwidth) is still an open important issue especially for learning with small sample sizes. In previous research, the kernel size was set manually or estimated in advance by Silvermans rule based on the sample distribution. This study aims to develop an online technique for optimizing the kernel size of the kernel least mean square (KLMS) algorithm. A sequential optimization strategy is proposed, and a new algorithm is developed, in which the filter weights and the kernel size are both sequentially updated by stochastic gradient algorithms that minimize the mean square error (MSE). Theoretical results on convergence are also presented. The excellent performance of the new algorithm is confirmed by simulations on static function estimation and short term chaotic time series prediction.

stat.ML↗