Searcharxiv⌕ Search

arXiv subjects

Jie Shi

Publications and source records attributed to Jie Shi.

At least 19 recordsLinked to original sources
arXiv2026

SocialRL: Refining LLMs' Social Intelligence through Multi-turn Reinforcement Learning and Reward Design

Social intelligence enables agents to read social context, infer intent, and adapt over sustained dialogue. As language models become autonomous collaborators, it is central to building effective and trustworthy human-AI interaction. Existing reinforcement learning methods optimize single-turn utterances and sparse outcome rewards, producing short-sighted policies that struggle to manage goal-relationship tensions across multi-turn interactions. We propose SocialRL, a multi-turn reinforcement learning framework addressing both challenges. First, we apply multi-turn reinforcement learning using PPO that propagates delayed outcome rewards back to each turn, enabling long-horizon planning. Second, we design six process reward dimensions capturing the goal-relationship trade-off, including goal advancement, relational attunement, contextual coherence, etc. A reward model dynamically generates fine-grained scoring criteria for each dimension, while a stage-aware weight schedule prioritizes relationship-building in early turns, goal advancement mid-way, and balanced closure late. Across multiple social-dialogue benchmarks, SocialRL improves Goal Achievement by an average of 9.2 percentage points over the corresponding Base models. These results demonstrate the effectiveness of SocialRL across synthetic and real social scenes, as well as standard and challenging social scenarios.

cs.CL↗
arXiv2026

Long-Horizon Agent Trajectory Attribution: A Unified Benchmark and Fine-Grained Annotation Framework

Large language model (LLM) agents increasingly operate through long-horizon trajectories involving user instructions, tool use, external observations, and memory. Existing benchmarks primarily evaluate behavioral outcomes but provide limited support for fine-grained attribution analysis. We introduce trajectory attribution and develop a benchmark and annotation framework for this task. The benchmark organizes heterogeneous trajectories under a unified component schema and provides annotations of the primary attribution component, together with attack and execution chains where applicable. Instantiating the benchmark with trajectories from AgentDojo and the Stage and Canary settings of Agent3Sigma yields more than 1,300 annotated trajectories covering task-aligned actions, unsafe actions, and safety refusals. The benchmark defines two evaluation tasks, primary attribution localization and attribution-chain recovery, and provides reference baselines based on incremental trajectory contribution and component-level leave-one-out perturbation. It captures diverse attribution settings, including local and long-range attribution as well as structured attribution chains. Reference baseline results exhibit substantial performance differences across these settings, providing an initial characterization of the benchmark's attribution challenges. Beyond this initial instantiation, we release a reusable annotation skill that enables trajectories generated by new agent models to be standardized, annotated, and evaluated under the same framework. Project resources and future releases are available at https://github.com/chenjing-2024/agent-trajectory-attribution.

cs.AI↗
arXiv2026

Convexity criterion and radial-profile response for off-shell Kerr geometries: a fuzzy-dark-matter profile as an analytic benchmark

We establish a sufficient one-minimum criterion for the off-shell Kerr family $\Delta(r) = r^2 - 2rm(r) + a^2$ with a positive, nondecreasing mass profile $m(r)$, showing that $1 - 2m'(r) - rm''(r) > 0$ ensures strict convexity and determines root counts for $\Delta$. Using a fuzzy-dark-matter-inspired benchmark satisfying this bound, we derive first-order responses for the outer horizon, extremal branch, photon sphere, and shadow functional under general deformations $m/M_{\text{ADM}} = 1 + \varepsilon h$. We demonstrate that static horizon and photon responses are profile-controlled, spin-odd shadow displacements are completion-dependent, and scale-consistent weak-field limits render local profile-gradient effects negligible ($\ll 10^{-20}$), confirming the strong-field box as a formal radial-profile benchmark rather than a self-consistent rotating scalar-field solution.

gr-qc↗
arXiv2026

EMRI Dephasing from a Torsion-Inspired Near-Zone Kerr Deformation: Motivated by Spin-Polarized Dark Matter

Extreme-mass-ratio inspirals (EMRIs) are sensitive probes of weak conservative perturbations in the strong-field region of massive black holes. We study a phenomenological EMRI model motivated by Einstein--Cartan gravity in which a spin-polarized dark-matter spike is described by a Weyssenhoff fluid. After torsion is eliminated algebraically, the local spin contribution contains a repulsive exterior source $U_{tt}^{\rm spin}\propto-\sigma_0^2/r^3$. Solving the corresponding static linearized field equation, however, does not produce a global $1/r^3$ metric perturbation; the response contains a mass renormalization, a logarithmic $r^{-1}$ tail, and an $M/r^2$ term. We therefore introduce $g_{\mu\nu}^{\rm eff}=g_{\mu\nu}^{\rm Kerr}+\alpha h_{\mu\nu}^{\rm eff}$ only as a local near-zone matching ansatz, not as a complete rotating Einstein--Cartan black-hole solution. Within this torsion-inspired deformation we compute circular equatorial inspirals and analytic-kludge waveforms. The fiducial model can produce large phase shifts in an idealized adiabatic calculation, but the forecast is optimistic and does not include a full LISA/Taiji response, Teukolsky/self-force fluxes, eccentricity, inclination, or high-dimensional parameter degeneracies. The results should be read as constraints on an effective near-zone operator rather than as a prediction of minimally coupled Einstein--Cartan dark matter.

gr-qc↗
arXiv2026

Random Fixed Point Theorems for Relaxed Asymptotic Contractions in Random Normed Modules

We introduce the notion of a random relaxed asymptotic contraction in the setting of random normed modules. The contraction condition employs two quasi-metrics that are built directly from the random operator: a lower quasi-metric which adaptively switches between a four-point minimum and the ordinary one-step distance, and an upper quasi-metric which takes the maximum of four fundamental distances. The bounds are allowed to depend on the iteration index and are required to converge locally uniformly almost surely to a Boyd--Wong function. Using the fibre decomposition method based on \(\sigma\)-stability and the local property, we show that any such mapping defined on an essentially bounded, \(\sigma\)-stable and \(L^0\)-closed set admits a unique random fixed point, and all iterates converge in the \((\epsilon,\lambda)\)-topology. Our result strictly generalizes the random analogue of Kirk's asymptotic contraction theorem and unifies several deterministic and random fixed point theorems under a single flexible framework.

math.FA↗
arXiv2026

VoxSafeBench: Not Just What Is Said, but Who, How, and Where

As speech language models (SLMs) transition from personal devices into shared, multi-user environments, their responses must account for far more than the words alone. Who is speaking, how they sound, and where the conversation takes place can each turn an otherwise benign request into one that is unsafe, unfair, or privacy-violating. Existing benchmarks, however, largely focus on basic audio comprehension, study individual risks in isolation, or conflate content that is inherently harmful with content that only becomes problematic due to its acoustic context. We introduce VoxSafeBench, among the first benchmarks to jointly evaluate social alignment in SLMs across three dimensions: safety, fairness, and privacy. VoxSafeBench adopts a Two-Tier design: Tier1 evaluates content-centric risks using matched text and audio inputs, while Tier2 targets audio-conditioned risks in which the transcript is benign but the appropriate response hinges on the speaker, paralinguistic cues, or the surrounding environment. To validate Tier2, we include intermediate perception probes and confirm that frontier SLMs can successfully detect these acoustic cues yet still fail to act on them appropriately. Across 22 tasks with bilingual coverage, we find that safeguards appearing robust on text often degrade in speech: safety awareness drops for speaker- and scene-conditioned risks, fairness erodes when demographic differences are conveyed vocally, and privacy protections falter when contextual cues arrive acoustically. Together, these results expose a pervasive speech grounding gap: current SLMs frequently recognize the relevant social norm in text but fail to apply it when the decisive cue must be grounded in speech. Code and data are publicly available at: https://amphionteam.github.io/VoxSafeBench_demopage/

cs.SD↗
arXiv2026

Fixed Points of Asymptotic Pointwise Contractions under Local Uniform Convergence

We introduce a weak asymptotic version of nonlinear contraction, termed \emph{asymptotic pointwise contraction}. For a mapping on a metric space, this notion requires the existence of a sequence of functions that dominate the distances between the $n$-th iterates of any two points. The sequence is assumed to converge pointwise to a limit function, and the convergence is required to be uniform on every bounded set (i.e., locally uniform). The limit function is then controlled by a Boyd--Wong type condition: there exists a nondecreasing, right upper semicontinuous function strictly below the identity on positive numbers, and the limit function is bounded above by this function evaluated at a maximum term that involves not only the distance between the two points but also distances from each point to its image and mutual distances between each point and the image of the other. By standard analytic arguments we prove that if the mapping is continuous on a complete metric space and possesses a bounded orbit, then its iterates converge to a unique fixed point. This result extends Kirk's asymptotic contraction theorem by replacing global uniform convergence on $[0,\infty)$ with the weaker condition of local uniform convergence.

math.FA↗
arXiv2026

A Fixed Point Theorem for Random Asymptotically Pointwise Contractions

This paper combines the decomposition technique ($\sigma$-stability) in random functional analysis with the deterministic theory of asymptotically pointwise contractions to provide a complete self-contained derivation of a fixed point theorem for random asymptotically pointwise contractions. We assume the contraction function is linear $\psi(t)=\lambda t$ ($\lambda<1$) and focus on the linear case under the assumption that $G$ is bounded. By choosing $p$ sufficiently large so that $5^{1/p}\lambda<1$, we apply the deterministic theorem in $L^p(E)$. The paper gives detailed explanations of concepts such as random normed modules, the $(\epsilon,\lambda)$-topology, and $\sigma$-stability, and reviews the historical development of fixed point theory in the introduction.

math.FA↗
arXiv2026

A Diffusion-Contrastive Graph Neural Network with Virtual Nodes for Wind Nowcasting in Unobserved Regions

Accurate weather nowcasting remains one of the central challenges in atmospheric science, with critical implications for climate resilience, energy security, and disaster preparedness. Since it is not feasible to deploy observation stations everywhere, some regions lack dense observational networks, resulting in unreliable short-term wind predictions across those unobserved areas. Here we present a deep graph self-supervised framework that extends nowcasting capability into such unobserved regions without requiring new sensors. Our approach introduces "virtual nodes" into a diffusion and contrastive-based graph neural network, enabling the model to learn wind condition (i.e., speed, direction and gusts) in places with no direct measurements. Using high-temporal resolution weather station data across the Netherlands, we demonstrate that this approach reduces nowcast mean absolute error (MAE) of wind speed, gusts, and direction in unobserved regions by more than 30% - 46% compared with interpolation and regression methods. By enabling localized nowcasts where no measurements exist, this method opens new pathways for renewable energy integration, agricultural planning, and early-warning systems in data-sparse regions.

cs.LG↗
arXiv2026

Detectability and Systematic Bias from First-Order Phase-Transition Dephasing in Kerr EMRIs

We study gravitational-wave dephasing induced by an effective first-order phase transition in a Kerr extreme mass-ratio inspiral (EMRI). The transition is modeled phenomenologically as a finite-width restructuring of the dissipative flux sector, and its observational consequences are quantified with standard LISA matched-filter diagnostics. For a representative system with $M=2\times10^{5}M_\odot$, $\mu=1.4M_\odot$, and $\hat a=0.90$, we obtain $\rho_{\rm B}=5.064$, $\rho_{\rm T}=4.073$, $\rho_{\rm R}=1.051$, and a mismatch $\mathcal M=2.986\times10^{-3}$ after maximization over extrinsic time and phase shifts. Although the normalized mismatch remains small, the accumulated phase difference grows to $\Delta\Phi_{22}^{\rm SF}\sim 5\times10^{3}\,\mathrm{rad}$, indicating that a narrow transition window can generate a large coherent deformation of the inspiral clock while leaving the waveform globally close to the baseline branch in detector-weighted norm. The resulting signal therefore lies in a bias-sensitive regime, characterized by small mismatch, order-unity residual norm, and large cumulative dephasing. Our results suggest that the dominant consequence of the transition sector is not loss of detectability, but loss of faithfulness for precision inference. This motivates future LISA EMRI waveform models that incorporate parameterized transition sectors directly into the waveform manifold.

gr-qc↗
arXiv2026

Analytical derivation of long-term dephasing caused by phase transitions in the context of Kerr black holes

Extreme Mass Ratio Inspirals (EMRIs) constitute a prime target for future space-based gravitational-wave observatories such as LISA. In this paper, we analytically investigate the long-term phase shift (dephasing) in the gravitational wave signal induced by a first-order quantum chromodynamics (QCD) phase transition within a neutron star orbiting a supermassive Kerr black hole. By modeling the transition from a hadronic phase to a quark core phase, we quantify the sudden change in the tidal deformability ($\Lambda$) of the secondary object. Utilizing the Teukolsky formalism and Post-Newtonian expansions, we derive a strict analytical scaling law for the accumulated dephasing. We demonstrate that the Kerr spin parameter $a$ and the critical phase transition orbital velocity $v_c$ significantly amplify the dephasing effect. Our analytical framework provides a robust tool for probing the non-perturbative QCD equation of state at high baryon densities using gravitational wave astronomy.

gr-qc↗
arXiv2026

Discern Truth from Falsehood: Reducing Over-Refusal via Contrastive Refinement

Large language models (LLMs) aligned for safety often suffer from over-refusal, the tendency to reject seemingly toxic or benign prompts by misclassifying them as toxic. This behavior undermines models' helpfulness and restricts usability in sensitive or nuanced contexts. While prior work has proposed mitigation strategies such as data augmentation and activation steering, these approaches often face a trade-off: reducing over-refusal typically degrades the model's ability to reject genuinely harmful content. We argue that this issue arises from the ambiguous influence of toxic and seemingly toxic prompts on the model's learning dynamics. To address it, we introduce a preceding alignment stage, DCR: Discernment via Contrastive Refinement. Both theoretically and empirically, we demonstrate that contrastive refinement improves an LLM's capacity to distinguish truly toxic prompts from superficially toxic ones. Evaluation across diverse benchmarks shows that our method effectively reduces over-refusal while preserving the safety benefits of alignment. Importantly, it achieves this with minimal degradation of general capabilities, offering a more principled and robust direction for safety alignment.

cs.CL↗
arXiv2026

From Thermodynamic Criticality to Geometric Criticality: A Linear Kernel Map from Matter Susceptibilities to Black-Hole Shadows

We construct an explicit linear map from compact, conserved thermodynamic/effective-medium perturbations of the stress-energy tensor to the metric response in static, spherically symmetric spacetimes, and from there to geometric observables of direct relevance to horizon-scale imaging: the shadow radius and photon-sphere frequency. The response is expressed through $L^{1}$-bounded kernels written in a piecewise "local $+$ tail" form, which makes transparent the separation between near-photon-sphere sensitivity and far-zone contributions (including AdS tails). Under mild assumptions on the matter susceptibilities near a critical point, dominated convergence transfers the thermodynamic exponent to the geometric susceptibility, $\gamma_{\rm sh}=\gamma_{\rm th}$, with controlled analytic corrections. We further provide AdS far-zone bounds with explicit outside-support constants depending only on background geometric data at the photon sphere and shell geometry. A reproducible numerical pipeline with convergence diagnostics is presented and benchmarked.

gr-qc↗
arXiv2026

Z(3) Metastable Bubbles and Chiral Dynamics Across a Dark-QCD Deconfinement Transition

We present a self-contained theoretical analysis of a dark-QCD chiral transition in which the Polyakov-loop sector retains an explicit $Z(3)$ structure and couples consistently to the chiral order parameter.Working within a coupled chiral--Polyakov effective theory, we map the homogeneous vacuum landscape and identify a metastability window bounded by spinodal loss of stability.We then construct $Z(3)$ domain-wall solutions including chiral backreaction, extracting temperature-dependent wall profiles and surface tension.Finally, we connect homogeneous metastability and wall microphysics to thermal bubble nucleation by evaluating the critical radius $R_c(T)$ and the nucleation exponent $S_3(T)/T$ in the thin-wall regime, providing a compact set of reproducible diagnostics for the decay of the metastable phase.Our results establish a coherent pipeline from vacuum structure to nonperturbative interfaces and nucleation barriers, suitable for systematic extensions to full multi-field bounce calculations and dark-sector cosmological applications.

physics.gen-ph↗
arXiv2026

Metastable Transitions and $\Gamma$-Convergent Eyring-Kramers Asymptotics in Landau-QCD Gradient Systems

We develop a rigorous analytical framework for metastable stochastic transitions in Landau-type gradient systems inspired by QCD phenomenology. The functional $F(\sigma;u)=\int_\Omega [\frac{\kappa}{2}|\nabla\sigma|^2+V(\sigma;u)]\,dx$, depending smoothly on a control parameter $u\in\mathcal U$, is analyzed through the Euler-Lagrange map $\mathcal{E}(\sigma;u)=-\kappa\Delta\sigma+V'(\sigma;u)$ and its Hessian $\mathcal{L}_{\sigma,u}=-\kappa\Delta+V''(\sigma;u)$. By combining variational methods, $\Gamma$- and Mosco convergence, and spectral perturbation theory, we establish the persistence and stability of local minima and index-one saddles under parameter deformations and variational discretizations. The associated mountain-pass solutions form Cerf-continuous branches away from the discriminant set $\mathcal D=\{u:\det\mathcal L_{\sigma,u}=0\}$, whose crossings produce only fold or cusp catastrophes in generic one- and two-parameter slices. The $\Gamma$-limit is taken with respect to the $L^2(\Omega)$ topology, ensuring compactness, convergence of gradient flows, and spectral continuity of $\mathcal L_{\sigma,u}$. As a consequence, the Eyring-Kramers formula for the mean transition time between metastable wells retains quantitative validity under both parameter deformations and discretization refinement, with convergent free-energy barriers, unstable eigenvalues, and zeta-regularized determinant ratios. This construction unifies the classical intuition of Eyring, Kramers, and Langer with modern variational and spectral analysis, providing a mathematically consistent and physically transparent foundation for metastable decay and phase conversion in Landau-QCD-type systems.

physics.gen-ph↗
arXiv2026

R$^2$BD: A Reconstruction-Based Method for Generalizable and Efficient Detection of Fake Images

Recently, reconstruction-based methods have gained attention for AIGC image detection. These methods leverage pre-trained diffusion models to reconstruct inputs and measure residuals for distinguishing real from fake images. Their key advantage lies in reducing reliance on dataset-specific artifacts and improving generalization under distribution shifts. However, they are limited by significant inefficiency due to multi-step inversion and reconstruction, and their reliance on diffusion backbones further limits generalization to other generative paradigms such as GANs. In this paper, we propose a novel fake image detection framework, called R$^2$BD, built upon two key designs: (1) G-LDM, a unified reconstruction model that simulates the generation behaviors of VAEs, GANs, and diffusion models, thereby broadening the detection scope beyond prior diffusion-only approaches; and (2) a residual bias calculation module that distinguishes real and fake images in a single inference step, which is a significant efficiency improvement over existing methods that typically require 20$+$ steps. Extensive experiments on the benchmark from 10 public datasets demonstrate that R$^2$BD is over 22$\times$ faster than existing reconstruction-based methods while achieving superior detection accuracy. In cross-dataset evaluations, it outperforms state-of-the-art methods by an average of 13.87\%, showing strong efficiency and generalization across diverse generative methods. The code and dataset used for evaluation are available at https://github.com/QingyuLiu/RRBD.

cs.CV↗
arXiv2025

DFALLM: Achieving Generalizable Multitask Deepfake Detection by Optimizing Audio LLM Components

Audio deepfake detection has recently garnered public concern due to its implications for security and reliability. Traditional deep learning methods have been widely applied to this task but often lack generalisability when confronted with newly emerging spoofing techniques and more tasks such as spoof attribution recognition rather than simple binary classification. In principle, Large Language Models (LLMs) are considered to possess the needed generalisation capabilities. However, previous research on Audio LLMs (ALLMs) indicates a generalization bottleneck in audio deepfake detection performance, even when sufficient data is available. Consequently, this study investigates the model architecture and examines the effects of the primary components of ALLMs, namely the audio encoder and the text-based LLM. Our experiments demonstrate that the careful selection and combination of audio encoders and text-based LLMs are crucial for unlocking the deepfake detection potential of ALLMs. We further propose an ALLM structure capable of generalizing deepfake detection abilities to out-of-domain spoofing tests and other deepfake tasks, such as spoof positioning and spoof attribution recognition. Our proposed model architecture achieves state-of-the-art (SOTA) performance across multiple datasets, including ASVSpoof2019, InTheWild, and Demopage, with accuracy reaching up to 95.76% on average, and exhibits competitive capabilities in other deepfake detection tasks such as attribution, and localisation compared to SOTA audio understanding models. Data and codes are provided in supplementary materials.

cs.SD↗
arXiv2025

Skeletons Matter: Dynamic Data Augmentation for Text-to-Query

The task of translating natural language questions into query languages has long been a central focus in semantic parsing. Recent advancements in Large Language Models (LLMs) have significantly accelerated progress in this field. However, existing studies typically focus on a single query language, resulting in methods with limited generalizability across different languages. In this paper, we formally define the Text-to-Query task paradigm, unifying semantic parsing tasks across various query languages. We identify query skeletons as a shared optimization target of Text-to-Query tasks, and propose a general dynamic data augmentation framework that explicitly diagnoses model-specific weaknesses in handling these skeletons to synthesize targeted training data. Experiments on four Text-to-Query benchmarks demonstrate that our method achieves state-of-the-art performance using only a small amount of synthesized data, highlighting the efficiency and generality of our approach and laying a solid foundation for unified research on Text-to-Query tasks. We release our code at https://github.com/jjjycaptain/Skeletron.

cs.CL↗