arXiv subjects

Xingyu Ma

Publications and source records attributed to Xingyu Ma.

13 recordsLinked to original sources

arXiv2026

xDailyBench: Benchmarking LLMs on Professional Consultation for Real-Life Problems

Large language models (LLMs) are increasingly used for everyday assistance, yet existing benchmarks only partially reflect the requests users naturally make in practice. Real-world requests are often open-ended, casually specified, and context-dependent, requiring models not only to follow explicit instructions but also to infer unstated needs from user background and situational context. We introduce xDailyBench, a benchmark of 248 carefully curated tasks spanning 51 scenarios across personal life, white-collar work, learning and research, and cross-domain activities. The tasks are grounded in requests that users have actually completed or genuinely intended to accomplish with AI, and are evaluated with fine-grained binary rubrics covering both explicit and implicit requirements. We evaluate 11 frontier models under standardized agentic settings. The best models achieve a task-level score of 75.6\%, while all models perform substantially worse on implicit than explicit requirements, with gaps no less than 9 percentage points. These results reveal implicit requirement inference as a persistent bottleneck for reliably satisfying real-world everyday user needs.

cs.AI↗

arXiv2026

Speech Meets ELF: Audio Conditional Continuous-Target Diffusion for Speech Recognition and Translation

Speech-to-text (S2T) systems for recognition (ASR) and translation (S2TT) typically generate discrete text tokens. In contrast, continuous-target language modelling performs generation in a continuous space, yet its potential for S2T remains unexplored. To bridge this gap, we propose ELF-S2T, an audio-conditioned continuous-target generative model for S2T. Built upon the pre-trained Embedded Language Flows (ELF) backbone, ELF-S2T processes speech via a frozen Whisper encoder and a single linear projector, prepending the resulting audio condition to the noisy text latent for in-context, flow-matching denoising. To prevent the model from over-relying on its pre-trained text context, we introduce audio forcing during training, and further amplify the audio condition via classifier-free guidance at inference. Experiments on LibriSpeech and CoVoST2 show that ELF-S2T achieves competitive ASR and S2TT performance. Crucially, our error analysis reveals that, although ASR and S2TT errors look very different on the surface, both stem from the same underlying cause, a close distance confusion in the continuous latent space. This finding naturally aligns with the continuous representation generation paradigm, indicating a common semantic mapping process beneath recognition and translation. Our code and pretrained models are publicly available at https://github.com/Sslnon/ELF-S2T.

cs.SD↗

arXiv2026

Modeling Music as a Time-Frequency Image: A 2D Tokenizer for Music Generation

Autoregressive music generation depends strongly on the audio tokenizer. Existing high-fidelity codecs often use residual multi-codebook quantization, which preserves reconstruction quality but complicates language modeling after sequence flattening, as the residual hierarchy imposes strong sequential dependencies and can amplify error accumulation. We propose BandTok, a generation-oriented 2D Mel-spectrogram tokenizer that represents each frame with Mel-frequency band tokens from a single shared codebook. This design yields a physically interpretable time-frequency token grid with a more independent token structure, making it better suited for autoregressive modeling. BandTok improves reconstruction with a multi-scale PatchGAN objective and EMA codebook updates. We further introduce an autoregressive language model with 2D Rotary Position Embedding (2D RoPE) to preserve temporal and frequency-band structure during generation. Experiments show that BandTok improves over residual-codebook tokenizers and achieves strong results in a data-limited setting. The source code and generation demos for this work are publicly available.

cs.SD↗

arXiv2026

Impurity-Scattering Assisted Umklapp Scattering as the Origin of Low-Temperature Resistivity in the Normal-State of Cuprate Superconductors

The transport experiments reveal that the low-temperature resistivity in the normal-state of cuprate superconductors is quadratic in temperature (T-quadratic) in the underdoped pseudogap phase, while it is linear in temperature (T-linear) in the overdoped strange-metal phase, however, the full understanding of these different behaviours is still a challenging issue. Here starting from the microscopic electronic structure of cuprate superconductors, the low-temperature resistivity in the normal-state is investigated from the underdoped pseudogap phase to the overdoped strange-metal phase. It is shown that the mechanism requires both the impurity scattering and the umklapp scattering: the impurity scattering is needed to restrict the modification of the distribution function to at around the antinodal region,while the impurity-scattering assisted umklapp scattering from a spin excitation is at the heart of the behaviour in the low-temperature resistivity, where the doping dependence of the temperature scale exists, and presents a similar behavior of the antinodal spin pseudogap crossover temperature. In the low-temperature region above the temperature scale in the overdoped strange-metal phase, the resistivity is T-linear, however, in the low-temperature region below the temperature scale in the underdoped pseudogap phase, the opening of the spin pseudogap lowers the spin excitation density of states at around the antinodal region, which reduces the strength of the electron umklapp scattering from a spin excitation associated with the antinode, and thus leads to a T-quadratic behaviour of the resistivity.

cond-mat.supr-con↗

arXiv2025

GLM-TTS Technical Report

This work proposes GLM-TTS, a production-level TTS system designed for efficiency, controllability, and high-fidelity speech generation. GLM-TTS follows a two-stage architecture, consisting of a text-to-token autoregressive model and a token-to-waveform diffusion model. With only 100k hours of training data, GLM-TTS achieves state-of-the-art performance on multiple open-source benchmarks. To meet production requirements, GLM-TTS improves speech quality through an optimized speech tokenizer with fundamental frequency constraints and a GRPO-based multi-reward reinforcement learning framework that jointly optimizes pronunciation, speaker similarity, and expressive prosody. In parallel, the system enables efficient and controllable deployment via parameter-efficient LoRA-based voice customization and a hybrid phoneme-text input scheme that provides precise pronunciation control. Our code is available at https://github.com/zai-org/GLM-TTS. Real-time speech synthesis demos are provided via Z.ai (audio.z.ai), the Zhipu Qingyan app/web (chatglm.cn).

cs.SD↗

arXiv2024

Enhancing Text-to-SQL Capabilities of Large Language Models via Domain Database Knowledge Injection

Text-to-SQL is a subtask in semantic parsing that has seen rapid progress with the evolution of Large Language Models (LLMs). However, LLMs face challenges due to hallucination issues and a lack of domain-specific database knowledge(such as table schema and cell values). As a result, they can make errors in generating table names, columns, and matching values to the correct columns in SQL statements. This paper introduces a method of knowledge injection to enhance LLMs' ability to understand schema contents by incorporating prior knowledge. This approach improves their performance in Text-to-SQL tasks. Experimental results show that pre-training LLMs on domain-specific database knowledge and fine-tuning them on downstream Text-to-SQL tasks significantly improves the Execution Match (EX) and Exact Match (EM) metrics across various models. This effectively reduces errors in generating column names and matching values to the columns. Furthermore, the knowledge-injected models can be applied to many downstream Text-to-SQL tasks, demonstrating the generalizability of the approach presented in this paper.

cs.CL↗

arXiv2024

Low-temperature T$^{2}$ resistivity in the underdoped pseudogap phase versus T-linear resistivity in the overdoped strange-metal phase of cuprate superconductors

The transport experiments demonstrate a dramatic switch from the low-temperature linear in temperature (T-linear) resistivity in the overdoped strange-metal phase of cuprate superconductors to the low-temperature quadratic in temperature (T-quadratic) resistivity in the underdoped pseudogap phase, however, a consensus on the origin of this unusual switch is still lacking. Here the resistivity in the underdoped pseudogap phase of cuprate superconductors is investigated using the Boltzmann transport equation. The resistivity originates from the electron umklapp scattering mediated by the spin excitation, however, the dominant contribution mainly comes from the antinodal umklapp scattering. In particular, a low temperature $T_{\rm scale}$ scales with $\Delta^{2}_{p}$ in the underdoped regime due to the opening of a momentum dependent spin pseudogap, where $\Delta_{p}$ is the minimal umklapp vector at the antinode. Moreover, this $T_{\rm scale}$ decreases with the increase of doping in the underdoped regime, and then is reduced to a very low temperature in the overdoped regime. In the underdoped regime, the resistivity is T-quadratic at the low temperatures below $T_{\rm scale}$, where the strength of the T-quadratic resistivity weakens as the doping is raised. However, in the overdoped regime, the resistivity is T-linear at the low temperatures above $T_{\rm scale}$. The results in this paper together with the recent study on the resistivity in the overdoped regime therefore show that the electron umklapp scattering from a spin excitation responsible for the low-temperature T-linear resistivity in the overdoped regime naturally produces the low-temperature T-quadratic resistivity in the underdoped regime resulting from the opening of a momentum dependent spin pseudogap.

cond-mat.supr-con↗

arXiv2023

Correlation between the strength of low-temperature T-linear normal-state resistivity and $T_{\rm c}$ in overdoped electron-doped cuprate superconductors

The recently observed an intimate link between the nature of the strange metallic normal-state and superconductivity in the overdoped electron-doped cuprate superconductors is calling for an explanation. Here the intrinsic correlation between the strength of the low-temperature linear-in-temperature normal-state resistivity and superconducting transition temperature $T_{\rm c}$ in the overdoped electron-doped cuprate superconductors is studied within the framework of the kinetic-energy-driven superconductivity. On the one hand, the main ingredient is identified into a electron pairing mechanism involving the spin excitation, and then $T_{\rm c}$ has a dome-like shape doping dependence with the maximal $T_{\rm c}$ that occurs at around the optimal electron doping. On the other hand, in the normal-state above $T_{\rm c}$, the low-temperature linear-in-temperature normal-state resistivity in the overdoped regime arises from the momentum relaxation due to the electron umklapp scattering mediated by the same spin excitation. This same spin excitation that governs both the electron umklapp scattering responsible for the low-temperature linear-in-temperature normal-state resistivity and electron pairing responsible for superconductivity naturally generates a correlation between the strength of the low-temperature linear-in-temperature normal-state resistivity and $T_{\rm c}$ in the overdoped regime.

cond-mat.supr-con↗

arXiv2023

Low-temperature T-linear resistivity in the strange metal phase of overdoped cuprate superconductors due to umklapp scattering from a spin excitation

The strange-metal phase of overdoped cuprate superconductors exhibits a linear in temperature resistivity in the low temperature, however, the origin of this remarkable anomaly is still not well understood. Here the linear temperature dependence of the resistivity in the strange-metal phase of overdoped cuprate superconductors is investigated. The momentum dependence of the transport scattering rate arising from the umklapp scattering between electrons by the exchange of the spin excitations is derived and employed to calculate the resistivity by making use of the Boltzmann equation. It is shown that the resistivity is mainly dominated by the antinodal and nodal umklapp scattering. In particular, a very low temperature $T_{\rm scale}$ scales with $Δ^{2}_{p}$, where $Δ_{p}$ is the minimum umklapp vector at the antinode. In the low temperature above $T_{\rm scale}$, the resistivity is linear in temperature with the temperature linear coefficient that decreases with the increase of doping, however, in the far lower temperature below $T_{\rm scale}$, the resistivity is instead quadratic in temperature. The theory also shows that the same spin excitation that acts like a bosonic glue to hold the electron pairs together also mediates scattering of electrons in the strange-metal phase responsible for the linear in temperature resistivity in the low temperature.

cond-mat.supr-con↗

arXiv2023

Electromagnetic response of cuprate superconductors with coexisting electronic nematicity

The electronically nematic order has emerged as a key feature of cuprate superconductors,however, its correlation with the fundamental properties such as the electromagnetic response remains unclear. Here the nematic-order state strength dependence of the electromagnetic response in cuprate superconductors is investigated within the framework of the kinetic-energy-driven superconductivity. It is shown that a significant anisotropy of the electromagnetic response is caused by the electronic nematicity. In particular, in addition to the pure d-wave component of the superconducting gap, the pure s-wave component of the superconducting gap is generated by the electronic nematicity, therefore there is a coexistence and competition of the d-wave component and the s-wave component. This coexistence and competition leads to that the maximal condensation energy appears at around the optimal strength of the electronic nematicity, and then decreases in both the weak and strong strength regions, which in turn induces the enhancement of superconductivity, and gives rise to the dome-like shape of the nematic-order state strength dependence of the superfluid density.

cond-mat.supr-con↗

arXiv2021

Characteristic energy of the nematic-order state and its connection to enhancement of superconductivity in cuprate superconductors

The new development in sublattice-phase-resolved imaging of electronic structure now allow for the visualisation of the nematic-order state characteristic energy of cuprate superconductors in a wide doping regime. However, it is still unclear how this characteristic energy of the nematic-order state is correlated with the enhancement of superconductivity. Here the doping dependence of the nematic-order state characteristic energy in cuprate superconductors and of its possible connection to the enhancement of superconductivity is investigated within the framework of the kinetic-energy-driven superconductivity. It is shown that the characteristic energy of the nematic-order state is found to be particularly large in the underdoped regime, then it smoothly decreases upon the increase of doping, in full agreement with the corresponding experimental observations. Moreover, the characteristic energy of the nematic-order state as a function of the nematic-order state strength in the underdoped regime presents a similar behavior of the superconducting transition temperature. This suggests a possible connection between the nematic-order state characteristic energy and the enhancement of the superconductivity.

cond-mat.supr-con↗

arXiv2020

Nonlinear hydrodynamic instability and turbulence in pulsatile flow

Pulsating flows through tubular geometries are laminar provided that velocities are moderate. This in particular is also believed to apply to cardiovascular flows where inertial forces are typically too low to sustain turbulence. On the other hand flow instabilities and fluctuating shear stresses are held responsible for a variety of cardiovascular diseases. Here we report a nonlinear instability mechanism for pulsating pipe flow that gives rise to bursts of turbulence at low flow rates. Geometrical distortions of small, yet finite amplitude are found to excite a state consisting of helical vortices during flow deceleration. The resulting flow pattern grows rapidly in magnitude, breaks down into turbulence, and eventually returns to laminar when the flow accelerates. This scenario causes shear stress fluctuations and flow reversal during each pulsation cycle. Such unsteady conditions can adversely affect blood vessels and have been shown to promote inflammation and dysfunction of the shear stress sensitive endothelial cell layer.

physics.flu-dyn↗

arXiv2017

Transition to turbulence in pulsating pipe flow

Fluid flows in nature and applications are frequently subject to periodic velocity modulations. Surprisingly, even for the generic case of flow through a straight pipe, there is little consensus regarding the influence of pulsation on the transition threshold to turbulence: while most studies predict a monotonically increasing threshold with pulsation frequency (i.e. Womersley number, $α$), others observe a decreasing threshold for identical parameters and only observe an increasing threshold at low $α$. In the present study we apply recent advances in the understanding of transition in steady shear flows to pulsating pipe flow. For moderate pulsation amplitudes we find that the first instability encountered is subcritical (i.e. requiring finite amplitude disturbances) and gives rise to localized patches of turbulence ("puffs") analogous to steady pipe flow. By monitoring the impact of pulsation on the lifetime of turbulence we map the onset of turbulence in parameter space. Transition in pulsatile flow can be separated into three regimes. At small Womersley numbers the dynamics are dominated by the decay turbulence suffers during the slower part of the cycle and hence transition is delayed significantly. As shown in this regime thresholds closely agree with estimates based on a quasi steady flow assumption only taking puff decay rates into account. The transition point predicted in the zero $α$ limit equals to the critical point for steady pipe flow offset by the oscillation Reynolds number. In the high frequency limit puff lifetimes are identical to those in steady pipe flow and hence the transition threshold appears to be unaffected by flow pulsation. In the intermediate frequency regime the transition threshold sharply drops (with increasing $α$) from the decay dominated (quasi steady) threshold to the steady pipe flow level.

physics.flu-dyn↗