252 papers
The paper introduces FireRedASR2S, a state-of-the-art industrial-grade all-in-one automatic speech recognition system that unifies high-performance modules for speech transcription, voice activity detection, language identification, and punctuation prediction, achieving superior results across Mandarin, Chinese dialects, and English benchmarks compared to existing solutions.
This paper proposes a robust, efficient three-point movable antenna placement strategy for near-field wireless sensing that minimizes the worst-case squared position error bound by leveraging centro-symmetric distribution and moment-based analysis, outperforming conventional fixed arrays while matching exhaustive search benchmarks with negligible computational complexity.
This paper systematically analyzes widely used speech tokenizers and reveals that they primarily encode phonetic rather than lexical-semantic information, highlighting a critical mismatch with text-derived semantics that necessitates new design approaches for effective multimodal LLM integration.
This paper presents a comprehensive analysis of UAV-based 3D spectrum sensing and Radio Environment Map (REM) reconstruction, demonstrating that robust algorithms like simple Kriging and Gaussian Process Regression, combined with altitude-aware trajectory planning, increased bandwidth, and airframe-induced antenna pattern calibration, significantly enhance mapping accuracy even under sparse sampling and complex shadowing conditions.
This paper proposes and compares two novel paradigms for multi-modal intelligent channel modeling in 6G systems—fine-tuned Large Language Models (LLM4CM) and a pre-trained Wireless Channel Foundation Model (WiCo)—both grounded in the Synesthesia of Machines concept to enable precise, scalable, and physically interpretable channel prediction across complex communication environments.
This paper introduces Simulation-in-the-Reasoning (SiR), a conceptual framework that enhances Large Language Models for autonomous transportation by embedding domain-specific simulators directly into the reasoning loop to transform hypothetical narratives into falsifiable, empirically grounded workflows.
This paper proposes a data-driven output feedback controller for nonlinear systems that utilizes kernel-interpolated inverse models and a data-driven reference selection framework to achieve verifiable practical output regulation using noise-free input/output measurements.
This paper proposes a control-theoretic framework for designing recommendation systems as state-feedback optimal control problems that balance engagement and stability, demonstrating that while appropriate weighting of objectives ensures system stability, excessive focus on engagement can lead to pathological polarization and destabilization.
This paper introduces a high-fidelity, labeled digital twin dataset generated from a MATLAB/Simulink electromagnetic transient model of a ten-inverter microgrid, capturing synchronized three-phase waveforms and power metrics across eleven distinct operating and disturbance scenarios to serve as a robust benchmark for training surrogate models and analyzing cyber-physical resilience.
This paper proposes a formation control framework for heterogeneous autonomous agents that leverages the superposition property of wireless multiple access channels to compute convex combinations of neighbor signals, thereby guaranteeing convergence to a prescribed formation under time-varying graphs and unknown channel coefficients while significantly reducing communication overhead compared to traditional interference-avoiding protocols.
This paper presents a scalable, non-intrusive in-situ timing diagnosis architecture for SRAM-based FPGAs that utilizes distributed phase-swept delay monitoring to probabilistically characterize and spatially differentiate between global power-distribution-network degradation and localized configuration-induced routing perturbations during normal operation.
This paper introduces ARCHE, an end-to-end learned image compression framework that achieves state-of-the-art rate-distortion efficiency by unifying hierarchical, spatial, and channel-based priors with adaptive feature recalibration, all while maintaining computational efficiency without relying on recurrent or transformer-based components.
This paper introduces a calibration-reasoning framework that fine-tunes foundational Audio Large Language Models through a calibration stage and Group Relative Policy Optimization-based reinforcement learning to achieve state-of-the-art performance in multidimensional speech quality assessment, artifact localization, and MOS prediction.
This paper proposes a data-driven successive linearization method for optimal voltage control in power distribution systems that overcomes the infeasibility of fixed linear approximations under nonlinear power flow constraints by dynamically updating linearizations around the most recent operating point, thereby ensuring fast convergence and adaptability to load variations.
This paper introduces Deep Unfolded Full Waveform Inversion (DUFWI), a physics-informed deep learning method that achieves real-time, high-quality speed-of-sound reconstruction for edema detection by overcoming the computational intensity and local minima issues of traditional inversion techniques.
This paper analyzes the fundamental performance limits of a secure MIMO integrated sensing and communications channel with multiple-antenna eavesdroppers and sensing receivers, deriving a quasi-optimal precoder structure and proposing a practical two-stage iterative algorithm to maximize the weighted sum rate.
This paper introduces ParaS2S, a reinforcement learning framework and corresponding benchmark (ParaS2SBench) that utilizes a novel PolyTone-trained automatic judge to effectively align speech-to-speech models with paralinguistic cues, achieving superior performance in response content and speaking style compared to supervised fine-tuning while requiring fewer paired demonstrations.
This paper introduces SAAIPAA, a physics-based framework that optimizes the placement of corner reflectors to execute aspect-angle-invariant physical adversarial attacks against SAR target recognition models, achieving high fooling rates even when the attacker lacks knowledge of the SAR platform's viewing angles.
This paper presents a real-time, on-device method for quantifying sensor output uncertainty by propagating calibration parameter quantization errors, demonstrating significant computational speedups over Monte Carlo methods and improved edge detection accuracy on low-power embedded hardware.
This paper proposes Circular ECG Super-Resolution (CECGSR), a closed-loop framework that leverages negative feedback and a Plug-and-Play strategy to outperform existing open-loop methods in reconstructing high-resolution ECG signals from low-resolution, noisy inputs.