arXiv subjects

Qiqi Wang

Publications and source records attributed to Qiqi Wang.

At least 19 recordsLinked to original sources

arXiv2026

FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation

Despite the non-autoregressive potential of diffusion language models (dLLMs), existing decoding strategies demonstrate positional bias, failing to fully unlock the potential of arbitrary generation. In this work, we delve into the inherent spectral characteristics of dLLMs and present the first frequency-domain analysis showing that low-frequency components in hidden states primarily encode global structural information and long-range dependencies, while high-frequency components are responsible for characterizing local details. Based on this observation, we propose FourierSampler, which leverages a frequency-domain sliding window mechanism to dynamically guide the model to achieve a "structure-to-detail" generation. FourierSampler outperforms other inference enhancement strategies on LLADA and SDAR, achieving relative improvements of 20.4% on LLaDA1.5-8B and 16.0% on LLaDA-8B-Instruct. It notably surpasses similarly sized autoregressive models like Llama3.1-8B-Instruct.

cs.CL↗

arXiv2026

SA-AI (Spalart-Allmaras with Autogenous Inception) Technical Summary

Natural transition from laminar to turbulent flow can be modeled by using only the Spalart-Allmaras (SA) working variable. The variable serves as its own transition indicator, in a one-equation Reynolds-averaged Navier-Stokes (RANS) closure. Its sub-O(1) range is dynamically inert in the baseline model. That range becomes a Tollmien-Schlichting amplification factor. A blended production term then drives the SA transport equation through laminar instability growth and turbulent eddy-viscosity production. Computations with this formulation on a zero-pressure-gradient flat plate, the NLF(1)-0416 and Eppler 387 airfoils, a two-element section, a circular cylinder through the drag crisis, the Daedalus human-powered-aircraft wing, and a 6:1 prolate spheroid are compared with experimental data and with other transition models.

physics.flu-dyn↗

arXiv2026

Large Models for Battery Prognostics and Health Management: A Review and Future Roadmap

Battery Prognostics and Health Management (BPHM) is critical for ensuring the safe, reliable, and cost-effective operation of batteries across electric vehicles, grid storage, and consumer electronics. Conventional BPHM approaches, including physics-based models and task-centric deep learning methods, face challenges in computational efficiency and parameterization, cross-domain generalization, dependence on extensive labeled run-to-failure data, and model interpretability. Recent Large Models (LMs), built upon Transformer architectures and self-supervised pre-training, offer a transformative new paradigm to overcome these long-standing bottlenecks. This review provides the first comprehensive survey of LM applications in BPHM, systematically examining how these models address challenges in the field. We begin by elucidating the foundational technologies enabling LMs, including Transformer architectures, self-supervised learning, large-scale multimodal datasets, and PEFT techniques. We then categorize recent progress along four critical dimensions: mitigating data scarcity, enhancing generalization and robustness, integrating domain knowledge for interpretability, and enabling system-level automation. Despite promising results, significant challenges remain across data accessibility, intelligence validation, trustworthiness, and deployment feasibility. To guide future research, we propose a roadmap focused on building collaborative data ecosystems, validating intelligence for industrial applications, enhancing trustworthiness with physics-informed designs, and enabling efficient on-device deployment. This review establishes a systematic approach to understand and advance LM-driven BPHM, providing researchers and practitioners with essential insights for developing next-generation battery management systems capable of safe, reliable, and autonomous operation throughout battery lifecycles.

cs.AI↗

arXiv2026

Words at Play: Benchmarking Audio Pun Understanding in Large Audio-Language Models

Puns represent a typical linguistic phenomenon that exploits polysemy and phonetic ambiguity to generate humour, posing unique challenges for natural language understanding. Within pun research, audio plays a central role in human communication except text and images, while datasets and systematic resources for spoken puns remain scarce, leaving this crucial modality largely underexplored. In this paper, we present APUN-Bench, the first benchmark dedicated to evaluating large audio language models (LALMs) on audio pun understanding. Our benchmark contains 4,434 audio samples annotated across three stages: pun recognition, pun word location and pun meaning inference. We conduct a deep analysis of APUN-Bench by systematically evaluating 10 state-of-the-art LALMs, uncovering substantial performance gaps in recognizing, localizing, and interpreting audio puns. This analysis reveals key challenges, such as positional biases in audio pun location and error cases in meaning inference, offering actionable insights for advancing humour-aware audio intelligence.

cs.SD↗

arXiv2026

DiRL: An Efficient Post-Training Framework for Diffusion Language Models

Diffusion Language Models (dLLMs) have emerged as promising alternatives to Auto-Regressive (AR) models. While recent efforts have validated their pre-training potential and accelerated inference speeds, the post-training landscape for dLLMs remains underdeveloped. Existing methods suffer from computational inefficiency and objective mismatches between training and inference, severely limiting performance on complex reasoning tasks such as mathematics. To address this, we introduce DiRL, an efficient post-training framework that tightly integrates FlexAttention-accelerated blockwise training with LMDeploy-optimized inference. This architecture enables a streamlined online model update loop, facilitating efficient two-stage post-training (Supervised Fine-Tuning followed by Reinforcement Learning). Building on this framework, we propose DiPO, the first unbiased Group Relative Policy Optimization (GRPO) implementation tailored for dLLMs. We validate our approach by training DiRL-8B-Instruct on high-quality math data. Our model achieves state-of-the-art math performance among dLLMs and surpasses comparable models in the Qwen2.5 series on several benchmarks.

cs.LG↗

arXiv2025

Thus Spake Long-Context Large Language Model

Long context is an important topic in Natural Language Processing (NLP), running through the development of NLP architectures, and offers immense opportunities for Large Language Models (LLMs), giving LLMs the lifelong learning potential akin to humans. Unfortunately, the pursuit of a long context is accompanied by numerous obstacles. Nevertheless, long context remains a core competitive advantage for LLMs. In the past two years, the context length of LLMs has achieved a breakthrough extension to millions of tokens. Moreover, research on long-context LLMs has expanded beyond length extrapolation to a comprehensive focus on architecture, infrastructure, training, and evaluation technologies. Inspired by the symphonic poem, Thus Spake Zarathustra, we draw an analogy between the journey of extending the context of LLM and the attempts of humans to transcend their mortality. In this survey, we will illustrate how LLM struggles between the tremendous need for a longer context and its equal need to accept the fact that it is ultimately finite. To achieve this, we give a global picture of the lifecycle of long-context LLMs from four perspectives: architecture, infrastructure, training, and evaluation, showcasing the full spectrum of long-context technologies. At the end of this survey, we will present 10 unanswered questions currently faced by long-context LLMs. We hope this survey can serve as a systematic introduction to research on long-context LLMs. Video: https://www.bilibili.com/video/BV11h9AYoEYj. Github: https://github.com/OpenMOSS/Thus-Spake-Long-Context-LLM.

cs.CL↗

arXiv2025

LegalChainReasoner: A Legal Chain-guided Framework for Criminal Judicial Opinion Generation

A criminal judicial opinion represents the judge's disposition of a case, including the decision rationale and sentencing. Automatically generating such opinions can assist in analyzing sentencing consistency and provide judges with references to similar past cases. However, current research typically approaches this task by dividing it into two isolated subtasks: legal reasoning and sentencing prediction. This separation often leads to inconsistency between the reasoning and predictions, failing to meet real-world judicial requirements. Furthermore, prior studies rely on manually curated knowledge to enhance applicability, yet such methods remain limited in practical deployment. To address these limitations and better align with legal practice, we propose a new LegalAI task: Judicial Opinion Generation, which simultaneously produces both legal reasoning and sentencing decisions. To achieve this, we introduce LegalChainReasoner, a framework that applies structured legal chains to guide the model through comprehensive case assessments. By integrating factual premises, composite legal conditions, and sentencing conclusions, our approach ensures flexible knowledge injection and end-to-end opinion generation. Experiments on two real-world and open-source Chinese legal case datasets demonstrate that our method outperforms baseline models.

cs.CL↗

arXiv2025

Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache

Large Language Models struggle with memory demands from the growing Key-Value (KV) cache as context lengths increase. Existing compression methods homogenize head dimensions or rely on attention-guided token pruning, often sacrificing accuracy or introducing computational overhead. We propose FourierAttention, a training-free framework that exploits the heterogeneous roles of transformer head dimensions: lower dimensions prioritize local context, while upper ones capture long-range dependencies. By projecting the long-context-insensitive dimensions onto orthogonal Fourier bases, FourierAttention approximates their temporal evolution with fixed-length spectral coefficients. Evaluations on LLaMA models show that FourierAttention achieves the best long-context accuracy on LongBench and Needle-In-A-Haystack (NIAH). Besides, a custom Triton kernel, FlashFourierAttention, is designed to optimize memory via streamlined read-write operations, enabling efficient deployment without performance compromise.

cs.CL↗

arXiv2025

Development_of_a_novel_high-performance_balanced_homodyne_detector

True random numbers are extracted through measurements of vacuum fluctuations in quantum state components. We propose an improved scheme utilizing an optimization-based simulation methodology to enhance the temporal resolution of quantum state detection and processing efficiency of vacuum fluctuation signals in continuous-variable quantum random number generators (CV-QRNGs), while simultaneously maximizing the entropy content of quantum noise sources. This work presents the first application of optimization simulation methodology to balanced homodyne detector (BHD) circuit design, with particular emphasis on improving high-frequency transmission characteristics. The design framework prioritizes system stability and S-parameter sensitivity to optimize both circuit architecture and critical component parameters. The AC amplifier circuit was implemented through ADS high-frequency simulations using two ABA-52563 RF amplifiers in a cascaded configuration, with circuit modeling performed on Rogers 4350 substrate optimized for high-frequency applications. This approach enabled the development of a switched-configuration BHD featuring: 1) 1.9 GHz bandwidth, 2) 41.5 dB signal-to-noise ratio at 1.75 GHz, 3) 30 dB common-mode rejection ratio at 100 MHz, and 4) frequency response flatness within 1.5 dB across 1.3-1.7 GHz. Additionally, the Husimi function is employed for entropy analysis to reconstruct vacuum state phase-space distributions, validating the detector's quantum measurement fidelity. The implemented system demonstrates a collective generation rate of 20.0504 Gbps across four parallel channels, with all output streams successfully passing NIST SP 800-22 statistical testing requirements.

quant-ph↗

arXiv2024

Parallel and real-time post-processing for quantum random number generators

Quantum systems are particularly suited for generating true randomness due to their inherent unpredictability, which can be justified on physical principles. However, practical implementations of Quantum RNGs (QRNGs) are always subject to noise, or uncontrollable influences, diminishing the quality of raw randomness produced. This necessitates post-processing to convert raw output into genuine randomness. In current QRNG implementations, the critical issue of seed updating is often overlooked, risking security vulnerabilities due to increased security parameters when seeds are reused in post-processing, and frequent seed updates fail to yield net randomness, while reusing seeds relies on the assumption that the original sequence inputs are independent.In this work, we have provided a specific scheme for seed updates that balances practicality and security, exploring the parallel and real-time implementation of multiple seed real-time updating toeplitz hash extractors in an FPGA to achieve parallel QRNGs, focusing on efficient hardware computation resource use. Through logic optimization, we achieved a greater number of parallel channels and a post-processing matrix size three times larger than previous works on the same FPGA platform, utilizing fewer logic resources. This resulted in a higher rate of random number generation and enhanced security. Furthermore, with the use of higher-performance ADCs, we attained a random number production rate exceeding 20Gbps.High-speed random number transfer and seed updating were achieved using the PCIe high-speed interface.This marks a significant step toward chip-based parallel QRNGs, enhancing the practicality of CV QRNGs in trusted, device-independent, and semi-device-independent scenarios.

quant-ph↗

arXiv2022

Approximating linear response of physical chaos

Parametric derivatives of statistics are highly desired quantities in prediction, design optimization and uncertainty quantification. In the presence of chaos, the rigorous computation of these quantities is certainly possible, but mathematically complicated and computationally expensive. Based on Ruelle's formalism, this paper shows that the sophisticated linear response algorithm can be dramatically simplified in higher-dimensional systems featuring a statistical homogeneity in the physical space. We argue that the contribution of the SRB (Sinai-Ruelle-Bowen) measure change, which is an integral part of the full linear response, can be completely neglected if the objective function is appropriately aligned with unstable manifolds. This abstract condition could potentially be satisfied by a vast family of real-world chaotic systems, regardless of the physical meaning and mathematical form of the objective function and perturbed parameter. We demonstrate several numerical examples that support these conclusions and that present the use and performance of a reduced linear response algorithm. In the numerical experiments, we consider physical models described by differential equations, including Lorenz 63, Lorenz 96, and Kuramoto-Sivashinsky.

math.DS↗

arXiv2022

A data driven heuristic for rapid convergence of Scheduled Relaxation Jacobi schemes

The Scheduled Relaxation Jacobi (SRJ) method is a viable candidate as a high performance linear solver for elliptic partial differential equations (PDEs). The method greatly improves the convergence of the standard Jacobi iteration by applying a sequence of $M$ well-chosen overrelaxation and underrelaxation factors in each cycle of the algorithm to effectively attenuate the solution error. In previous work, optimal SRJ schemes (sets of relaxation factors) have been derived to accelerate convergence for specific discretizations of elliptic PDEs. In this work, we develop a family of SRJ schemes which can be applied to solve elliptic PDEs regardless of the specific discretization employed. To achieve favorable convergence, we train an algorithm to select which scheme in this family to apply at each cycle of the linear solve process, based on convergence data collected from applying these schemes to the one-dimensional Poisson equation. The automatic selection heuristic that is developed based on this limited data is found to provide good convergence for a wide range of problems.

math.NA↗

arXiv2022

DRVC: A Framework of Any-to-Any Voice Conversion with Self-Supervised Learning

Any-to-any voice conversion problem aims to convert voices for source and target speakers, which are out of the training data. Previous works wildly utilize the disentangle-based models. The disentangle-based model assumes the speech consists of content and speaker style information and aims to untangle them to change the style information for conversion. Previous works focus on reducing the dimension of speech to get the content information. But the size is hard to determine to lead to the untangle overlapping problem. We propose the Disentangled Representation Voice Conversion (DRVC) model to address the issue. DRVC model is an end-to-end self-supervised model consisting of the content encoder, timbre encoder, and generator. Instead of the previous work for reducing speech size to get content, we propose a cycle for restricting the disentanglement by the Cycle Reconstruct Loss and Same Loss. The experiments show there is an improvement for converted speech on quality and voice similarity.

cs.SD↗

arXiv2022

Assessment of Detached Eddy Simulation and Sliding Mesh Interface in Predicting Tiltrotor Performance in Helicopter and Airplane Modes

This paper presents numerical investigation on performance and flow field of the full-scale XV-15 tiltrotor in both helicopter mode (hovering flight and forward flight) and aeroplane propeller mode using Detached Eddy Simulation, in which the movement of the rotor is achieved using a Sliding Mesh Interface. Comparison of our CFD results against experiment data and other CFD results is performed and presented.

physics.flu-dyn↗

arXiv2022

Efficient computation of linear response of chaotic attractors with one-dimensional unstable manifolds

This paper presents the space-split sensitivity or the S3 algorithm to transform Ruelle's linear response formula into a well-conditioned ergodic-averaging computation. We prove a decomposition of Ruelle's formula that is differentiable on the unstable manifold, which we assume to be one-dimensional. This decomposition of Ruelle's formula ensures that one of the resulting terms, the stable contribution, can be computed using a regularized tangent equation, similar to in a non-chaotic system. The remaining term, known as the unstable contribution, is regularized and converted into an efficiently computable ergodic average. In this process, we develop new algorithms, which may be useful beyond linear response, to compute i) a fundamental statistical quantity we introduce called the density gradient, and ii) the unstable derivatives of the regularized tangent vector field and the unstable direction. We prove that the S3 algorithm, which combines these computational ingredients that enter the stable and unstable contribution, converges like a Monte Carlo approximation of Ruelle's formula. The algorithm presented here is hence a first step toward full-fledged applications of sensitivity analysis in chaotic systems, wherever such applications have been limited due to lack of availability of long-term sensitivities.

math.DS↗

arXiv2021

Scheduled Relaxation Jacobi schemes for non-elliptic partial differential equations

The Scheduled Relaxation Jacobi (SRJ) method is a linear solver algorithm which greatly improves the convergence of the Jacobi iteration through the use of judiciously chosen relaxation factors (an SRJ scheme) which attenuate the solution error. Until now, the method has primarily been used to accelerate the solution of elliptic PDEs (e.g. Laplace, Poisson's equation) as the currently available schemes are restricted to solving this class of problems. The goal of this paper is to present a methodology for constructing SRJ schemes which are suitable for solving non-elliptic PDEs (or equivalent, nonsymmetric linear systems arising from the discretization of these PDEs), thereby extending the applicability of this method to a broader class of problems. These schemes are obtained by numerically solving a constrained minimization problem which guarantees the solution error will not grow as long as the linear system has eigenvalues which lie in certain regions of the complex plane. We demonstrate that these schemes are able to accelerate the convergence of standard Jacobi iteration for the nonsymmetric linear systems arising from discretization of the 1D and 2D steady advection-diffusion equations.

math.NA↗

arXiv2021

High-performance Ba1-xKxFe2As2 superconducting joints for persistent current operation

Superconducting joints are one of the key technologies to make Ba1-xKxFe2As2 (Ba-122) superconducting wires or tapes for high-field applications. Herein, superconducting joints were fabricated by a simple cold-pressing method, and the joint resistance of the iron-based superconducting joint was estimated for the first time. The superconducting properties, microstructures, and elements distribution in the joint regions were investigated. At 4.2 K and 10 T, a transport critical current Ic of 105 A for the joint was obtained, and the critical current ratio (CCR= Ic-joint/Ic-tape) of the joint was 94.6%. On the other hand, the joint show very low joint resistance of 2.7x10^-13 ohm in self-field at 4.2 K. Among iron-based superconductors (IBS), this work is the first to successfully realize a superconducting joint with such high CCR and low joint resistance. This work shows great potential to apply Ba-122 in a range of practical applications, where superconducting joints are essential.

cond-mat.supr-con↗

arXiv2021

Space-split algorithm for sensitivity analysis of discrete chaotic systems with unstable manifolds of arbitrary dimension

Accurate approximations of the change of system's output and its statistics with respect to the input are highly desired in computational dynamics. Ruelle's linear response theory provides breakthrough mathematical machinery for computing the sensitivity of chaotic dynamical systems, which enables a better understanding of chaotic phenomena. In this paper, we propose an algorithm for sensitivity analysis of discrete chaos with an arbitrary number of positive Lyapunov exponents. We combine the concept of perturbation space-splitting regularizing Ruelle's original expression together with measure-based parameterization of the expanding subspace. We use these tools to rigorously derive trajectory-following recursive relations that exponentially converge, and construct a memory-efficient Monte Carlo scheme for derivatives of the output statistics. Thanks to the regularization and lack of simplifying assumptions on the behavior of the system, our method is immune to the common problems of other popular systems such as the exploding tangent solutions and unphysicality of shadowing directions. We provide a ready-to-use algorithm, analyze its complexity, and demonstrate several numerical examples of sensitivity computation of physically-inspired low-dimensional systems.

math.DS↗