arXiv subjects

Teng Zhang

Publications and source records attributed to Teng Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2026

Metric Properties: From $S$-Divergence to Quantum Jensen Divergence

We extend the trace-logarithmic $S$-divergence from matrices to tracial $C^*$-algebras and finite von Neumann algebras, and show that its square root defines a metric on the invertible positive cone. We also prove an integral representation of the quantum Jensen--Shannon divergence in terms of shifted trace-log distances, implying metricity of its square root on the full positive cone in the same tracial framework. In the matrix case, we answer two questions of Virosztek \cite{Vir21} on Hilbertianity. Finally, we show that symmetric quantum Jensen divergences generated by non-affine operator convex functions yield metrics in the tracial setting via a Nevanlinna--Stieltjes type representation of the derivative, which generalizes a result of Carlen, Lieb and Seiringer.

math.OA↗

arXiv2026

Benchmarking neural surrogates on realistic spatiotemporal multiphysics flows

Predicting multiphysics dynamics is computationally expensive and challenging due to the severe coupling of multi-scale, heterogeneous physical processes. While neural surrogates promise a paradigm shift, the field currently suffers from an "illusion of mastery", as repeatedly emphasized in top-tier commentaries: existing evaluations overly rely on simplified, low-dimensional proxies, which fail to expose the models' inherent fragility in realistic regimes. To bridge this critical gap, we present REALM (REalistic AI Learning for Multiphysics), a rigorous benchmarking framework designed to test neural surrogates on challenging, application-driven reactive flows. REALM features 11 high-fidelity datasets spanning from canonical multiphysics problems to complex propulsion and fire safety scenarios, alongside a standardized end-to-end training and evaluation protocol that incorporates multiphysics-aware preprocessing and a robust rollout strategy. Using this framework, we systematically benchmark over a dozen representative surrogate model families, including spectral operators, convolutional models, Transformers, pointwise operators, and graph/mesh networks, and identify three robust trends: (i) a scaling barrier governed jointly by dimensionality, stiffness, and mesh irregularity, leading to rapidly growing rollout errors; (ii) performance primarily controlled by architectural inductive biases rather than parameter count; and (iii) a persistent gap between nominal accuracy metrics and physically trustworthy behavior, where models with high correlations still miss key transient structures and integral quantities. Taken together, REALM exposes the limits of current neural surrogates on realistic multiphysics flows and offers a rigorous testbed to drive the development of next-generation physics-aware architectures.

cs.LG↗

arXiv2026

Point2RBox-v3: Self-Bootstrapping from Point Annotations via Integrated Pseudo-Label Refinement and Utilization

Driven by the growing need for Oriented Object Detection (OOD), learning from point annotations under a weakly-supervised framework has emerged as a promising alternative to costly and laborious manual labeling. In this paper, we discuss two deficiencies in existing point-supervised methods: inefficient utilization and poor quality of pseudo labels. Therefore, we present Point2RBox-v3. At the core are two principles: 1) Progressive Label Assignment (PLA). It dynamically estimates instance sizes in a coarse yet intelligent manner at different stages of the training process, enabling the use of label assignment methods. 2) Prior-Guided Dynamic Mask Loss (PGDM-Loss). It is an enhancement of the Voronoi Watershed Loss from Point2RBox-v2, which overcomes the shortcomings of Watershed in its poor performance in sparse scenes and SAM's poor performance in dense scenes. To our knowledge, Point2RBox-v3 is the first model to employ dynamic pseudo labels for label assignment, and it creatively complements the advantages of SAM model with the watershed algorithm, which achieves excellent performance in both sparse and dense scenes. Our solution gives competitive performance, especially in scenarios with large variations in object size or sparse object occurrences: 66.09%/56.86%/41.28%/46.40%/19.60%/45.96% on DOTA-v1.0/DOTA-v1.5/DOTA-v2.0/DIOR/STAR/RSAR.

cs.CV↗

arXiv2026

Bourin-type inequalities for $τ$-measurable operators in fully symmetric spaces

Let $\mathcal{M}\subset B(\mathcal{H})$ be a semifinite von Neumann algebra, where $B(\mathcal{H})$ denotes the algebra of all bounded linear operators on a Hilbert space $\mathcal{H}$, and let $τ$ be a fixed faithful normal semifinite trace on $\mathcal{M}$.Let $E_τ$ be the fully symmetric space associated with a fully symmetric Banach function space $E$ on $[0,\infty)$.Using a complex interpolation argument based on the three-lines theorem on a strip, we show that for positive operators $a,b\in E_τ$ and $t\in[0,1]$, $$ \|a^t b^{1-t}+b^t a^{1-t}\|_{E_τ}\le 2^{\max\{2|t-1/2|-1/2,\;0\}}\;\|a+b\|_{E_τ}. $$ In particular, we obtain the sharp constant $1$ for $t\in[1/4,3/4]$: $$ \|a^t b^{1-t}+b^t a^{1-t}\|_{E_τ}\le \|a+b\|_{E_τ}. $$ This extends the work of Kittaneh--Ricard in \emph{Linear Algebra Appl.} \textbf{710} (2025), 356--362 and covers the results of Liu--He--Zhao in \emph{Acta Math. Sci. Ser. B (Engl. Ed.)} \textbf{46} (2026), 62--68

math.FA↗

arXiv2026

Angle duality and a gap principle for convex combinations of incomplete polynomials on the unit circle

In this paper, we establish an angle duality and a gap principle for convex combinations of incomplete polynomials, extending two results of Ge and Gonek in [IMRN, 2024].Our approach is geometric: we introduce an ``angle gain'' mechanism for points inside the lune region and quantify how moving away from the unit circle forces a definite increase in the relevant angle functional.This yields a robust lower bound that is uniform under convex mixing and leads to the desired separation phenomenon.The main difficulty is that incomplete polynomials and their convex combinations may have highly nonuniform root distributions on the unit circle, so classical convex-hull type constraints are too coarse; one must instead control the local geometry of chords and boundary arcs and relate it to critical-point behavior through sharp trigonometric inequalities.

math.CV↗

arXiv2026

Johnson's determinantal identity for contiguous minors of Toeplitz matrices, with an accretive extension

Let $A$ be an $n\times n$ real Toeplitz matrix satisfying $A+A^{\top}=2\mathbb J_n$, where $\mathbb J_n$ is the all-ones matrix.If $A_r(i,j)$ denotes the $r\times r$ contiguous submatrix of $A$ consisting of rows $i,i+1,\dots,i+r-1$ and columns $j,j+1,\dots,j+r-1$, then for every $n\ge 2$ one has $$\det A_{n-1}(1,2)+\det A_{n-1}(2,1)=2\det A_{n-1}(1,1).$$ This confirms a conjecture of Charles R.~Johnson (2003). The proof combines a rank-one determinant expansion with Dodgson's condensation formula, and then invokes a polynomial-identity argument in the Toeplitz parameters: after obtaining an equality of squares in the integral domain $\mathbb{Z}[b_1,\dots,b_{n-1}]$, we factor it to deduce an identity up to sign and determine the sign by a suitable specialization.We also give an extension of the Bayat--Teimoori arithmetic--geometric mean identity: for every real accretive matrix $A$, one has the sharp inequality $$ \sqrt{\det A_{n-1}(1,1)\ \det A_{n-1}(2,2)} \ \ge\ \left|\frac{\det A_{n-1}(1,2)+\det A_{n-1}(2,1)}{2}\right|,$$ with equality whenever the symmetric part has rank one, i.e.\ $A+A^{\top}=α\,ww^{\top}$ for some $α\in\mathbb R$ and $w\in\mathbb R^n\setminus\{0\}$,recovering the Bayat--Teimoori equality as a special case.

math.FA↗

arXiv2026

From Reasoning LLMs to BERT: A Two-Stage Distillation Framework for Search Relevance

Query-service relevance prediction in e-commerce search systems faces strict latency requirements that prevent the direct application of Large Language Models (LLMs). To bridge this gap, we propose a two-stage reasoning distillation framework to transfer reasoning capabilities from a powerful teacher LLM to a lightweight, deployment-friendly student model. In the first stage, we address the limitations of general-purpose LLMs by constructing a domain-adapted teacher model. This is achieved through a three-step process: domain-adaptive pre-training to inject platform knowledge, supervised fine-tuning to elicit reasoning skills, and preference optimization with a multi-dimensional reward model to ensure the generation of reliable and preference-aligned reasoning paths. This teacher can then automatically annotate massive query-service pairs from search logs with both relevance labels and reasoning chains. In the second stage, to address the challenges of architectural heterogeneity in standard distillation, we introduce Contrastive Reasoning Self-Distillation (CRSD). By modeling the behavior of the same student model under ``standard'' and ``reasoning-augmented'' inputs as a teacher-student relationship, CRSD enables the lightweight model to internalize the teacher's complex decision-making mechanisms without needing the explicit reasoning path at inference. Offline evaluations and online A/B testing in the Meituan search advertising system demonstrate that our framework achieves significant improvements across multiple metrics, validating its effectiveness and practical value.

cs.IR↗

arXiv2026

Vertical Semi-Federated Learning for Efficient Online Advertising

Traditional vertical federated learning schema suffers from two main issues: 1) restricted applicable scope to overlapped samples and 2) high system challenge of real-time federated serving, which limits its application to advertising systems. To this end, we advocate a new practical learning setting, Semi-VFL (Vertical Semi-Federated Learning), for real-world industrial applications, where the learned model retains sufficient advantages of federated learning while supporting independent local serving. To achieve this goal, we propose the carefully designed Joint Privileged Learning framework (JPL) to i) alleviate the absence of the passive party's feature with federated equivalence imitation and ii) adapt to the heterogeneous full sample space with cross-branch rank alignment. Extensive experiments conducted on real-world advertising datasets validate the effectiveness of our method over baseline methods.

cs.LG↗

arXiv2026

Schur--Horn type inequalities for hyperbolic polynomials

We establish a Schur--Horn type inequality for symmetric hyperbolic polynomials. As an immediate consequence, we resolve a conjecture of Nam Q. Le on Hadamard-type inequalities for hyperbolic polynomials. Our argument is based on the Schur--Horn theorem, the Birkhoff theorem, and Gårding's concavity theorem for hyperbolicity cones. Beyond the eigenvalue level, we develop a symmetrization principle on hyperbolicity cones: if a hyperbolic polynomial is invariant under a finite group action, then its value increases under the associated Reynolds operator (group averaging). Applied to the sign-flip symmetries of linear principal minor polynomials introduced by Blekherman et al., this yields a short proof of the hyperbolic Fischer--Hadamard inequalities for PSD-stable lpm polynomials.

math.FA↗

arXiv2026

Weak majorization inequalities for the cubic and quartic coefficients of $e^{(A+B)t}$ versus $e^{At}e^{Bt}$

Let $A,B\in\mathbb{H}_n$ and set $H=A+B$. For each integer $k\ge 1$ define $$ Q_k:=\sum_{p=0}^k \binom{k}{p} A^pB^{k-p}, R_k:=\Re\,Q_k=\frac{Q_k+Q_k^*}{2}. $$ Then $H^k=\left.\frac{d^k}{dt^k}e^{Ht}\right|_{t=0}$ and $Q_k=\left.\frac{d^k}{dt^k}(e^{At}e^{Bt})\right|_{t=0}$. We prove that, for $k=3,4,$ $$ λ(H^k)\prec_w σ(Q_k). $$ Equivalently, the eigenvalues of the cubic and quartic Taylor coefficients of $e^{(A+B)t}$ are weakly majorized by the singular values of the corresponding coefficients of the Golden--Thompson product $e^{At}e^{Bt}$. Our argument combines Ky Fan variational principles with explicit commutator identitiesfor $R_k-H^k$ at orders $k=3,4$, reducing the problem to the positivity of certain double-commutator trace forms tested against Ky Fan maximizing projections. We also record a general sufficient condition for higher orders based on commutator decompositions.

math.FA↗

arXiv2025

WorldRFT: Latent World Model Planning with Reinforcement Fine-Tuning for Autonomous Driving

Latent World Models enhance scene representation through temporal self-supervised learning, presenting a perception annotation-free paradigm for end-to-end autonomous driving. However, the reconstruction-oriented representation learning tangles perception with planning tasks, leading to suboptimal optimization for planning. To address this challenge, we propose WorldRFT, a planning-oriented latent world model framework that aligns scene representation learning with planning via a hierarchical planning decomposition and local-aware interactive refinement mechanism, augmented by reinforcement learning fine-tuning (RFT) to enhance safety-critical policy performance. Specifically, WorldRFT integrates a vision-geometry foundation model to improve 3D spatial awareness, employs hierarchical planning task decomposition to guide representation optimization, and utilizes local-aware iterative refinement to derive a planning-oriented driving policy. Furthermore, we introduce Group Relative Policy Optimization (GRPO), which applies trajectory Gaussianization and collision-aware rewards to fine-tune the driving policy, yielding systematic improvements in safety. WorldRFT achieves state-of-the-art (SOTA) performance on both open-loop nuScenes and closed-loop NavSim benchmarks. On nuScenes, it reduces collision rates by 83% (0.30% -> 0.05%). On NavSim, using camera-only sensors input, it attains competitive performance with the LiDAR-based SOTA method DiffusionDrive (87.8 vs. 88.1 PDMS).

cs.RO↗

arXiv2025

Peeling-Induced Rolling and Heterogeneous Adhesion in Blistered Films

Blisters, delaminated regions that form in multilayered structures under compressive stresses, are observed across a wide range of length scales, from two-dimensional materials to protective coatings and laminated composites. Far from being passive defects, such interfacial features have emerged as functional motifs for three-dimensional architectures and reconfigurable surfaces. Here we reveal an unusual peel response of a blistered thin film on a soft substrate. When peeled from one end, the advancing peel front triggers reattachment at the blister edge once a critical separation is reached, initiating spontaneous rolling of the film on the substrate. This peel-to-roll transition produces a sharp drop in the measured adhesion force, which then remains constant throughout the rolling phase. Using experiments, scaling analysis, and molecular dynamics simulations, we resolve the contact morphology at the transition and identify the contact length at which rolling initiates. We show that this length arises from interactions between the two contact edges and is independent of the work of adhesion. Once rolling begins, a dynamically imposed dwell time - defined by the rolling length and peel speed - translates contact history into spatial variations in adhesion force, thereby governing the magnitude of the force drop. Together, these results point to a new pathway for generating spatially tunable, heterogeneous adhesion from otherwise homogeneous interfaces.

cond-mat.soft↗

arXiv2025

RPM-MCTS: Knowledge-Retrieval as Process Reward Model with Monte Carlo Tree Search for Code Generation

Tree search-based methods have made significant progress in enhancing the code generation capabilities of large language models. However, due to the difficulty in effectively evaluating intermediate algorithmic steps and the inability to locate and timely correct erroneous steps, these methods often generate incorrect code and incur increased computational costs. To tackle these problems, we propose RPM-MCTS, an effective method that utilizes Knowledge-Retrieval as Process Reward Model based on Monte Carlo Tree Search to evaluate intermediate algorithmic steps. By utilizing knowledge base retrieval, RPM-MCTS avoids the complex training of process reward models. During the expansion phase, similarity filtering is employed to remove redundant nodes, ensuring diversity in reasoning paths. Furthermore, our method utilizes sandbox execution feedback to locate erroneous algorithmic steps during generation, enabling timely and targeted corrections. Extensive experiments on four public code generation benchmarks demonstrate that RPM-MCTS outperforms current state-of-the-art methods while achieving an approximately 15% reduction in token consumption. Furthermore, full fine-tuning of the base model using the data constructed by RPM-MCTS significantly enhances its code capabilities.

cs.AI↗

arXiv2025

High efficiency superconducting diode effect in a gate-tunable double-loop SQUID

In superconducting quantum interference devices (SQUIDs), the superconducting diode effect may be generated by interference of multiple harmonic components in the current-phase relationships (CPRs) of different branches forming SQUID loops. Through the inclusion of two gate-tunable Josephson junctions in series in each interference branch of a double-loop SQUID, we demonstrate independent control over both the harmonic content and the amplitude of three interfering CPRs, facilitating significant improvement in the maximum diode efficiency. Through optimized gate-controlled tuning of individual Josephson energies, diode efficiency exceeding 50% is demonstrated. Flux-dependent oscillations show quantitative agreement with a simple model of SQUID operation.

cond-mat.supr-con↗

arXiv2025

Unleashing the Intrinsic Visual Representation Capability of Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) have demonstrated remarkable proficiency in multimodal tasks. Despite their impressive performance, MLLMs suffer from the modality imbalance issue, where visual information is often underutilized compared to textual representations in deeper layers, leading to degraded visual performance or hallucinations. This issue stems from the predominant reliance on next-text-token-prediction during training, which fails to provide direct visual supervisory signals, resulting in progressive homogenization of visual representations throughout the layers. To this end, we propose Latent Visual Reconstruction (LaVer), a novel training framework that facilitates MLLMs in learning more discriminative visual representations via masked image modeling in the joint latent semantic space of LLM. Our method offers direct visual activation to MLLMs, which exhibit increased visual attention allocation, indicating enhanced utilization of visual information. Extensive experiments across diverse benchmarks prove the superiority of our approach in various scenarios, especially those requiring dense visual capabilities. Code of LaVer is available at https://github.com/Fir-lat/LaVer.

cs.CV↗

arXiv2025

Sharp Schoenberg type inequalities and the de Bruin--Sharma problem

In this paper, we confirm two conjectures proposed by Georgiev, Gómez-Serrano, Tao, and Wagner~\cite{GGTW25} on Schoenberg type inequalities of order $4$, thereby providing a complete solution to the de Bruin--Sharma problem. We further develop a new interpolation framework to study Schoenberg type inequalities and, in particular, give a new proof of Pereira's result. Motivated by Sendov's conjecture, we then derive sharp Schoenberg type inequalities of orders $-1$ and $-2m$ (with $m \in \mathbb{N}$), as well as non-sharp inequalities valid for all negative orders $p \le -1$. Finally, we discuss a dual counterpart of the Schoenberg type inequalities of order $-2$.

math.CV↗

arXiv2025

Error Correction in Radiology Reports: A Knowledge Distillation-Based Multi-Stage Framework

The increasing complexity and workload of clinical radiology leads to inevitable oversights and mistakes in their use as diagnostic tools, causing delayed treatments and sometimes life-threatening harm to patients. While large language models (LLMs) have shown remarkable progress in many tasks, their utilities in detecting and correcting errors in radiology reporting are limited. This paper proposes a novel dual-knowledge infusion framework that enhances LLMs' capability for radiology report proofreading through systematic integration of medical expertise. Specifically, the knowledge infusion combines medical knowledge graph distillation (MKGD) with external knowledge retrieval (EXKR), enabling an effective automated approach in tackling mistakes in radiology reporting. By decomposing the complex proofreading task into three specialized stages of detection, localization, and correction, our method mirrors the systematic review process employed by expert radiologists, ensuring both precision and clinical interpretability. To perform a robust, clinically relevant evaluation, a comprehensive benchmark is also proposed using real-world radiology reports with real-world error patterns, including speech recognition confusions, terminology ambiguities, and template-related inconsistencies. Extensive evaluations across multiple LLM architectures demonstrate substantial improvements of our approach: up to 31.56% increase in error detection accuracy and 37.4% reduction in processing time. Human evaluation by radiologists confirms superior clinical relevance and factual consistency compared to existing approaches.

cs.CL↗

arXiv2025

A GPU-boosted high-performance multi-working condition joint analysis framework for predicting dynamics of textured axial piston pump

Accurate simulation to dynamics of axial piston pump (APP) is essential for its design, manufacture and maintenance. However, limited by computation capacity of CPU device and traditional solvers, conventional iteration methods are inefficient in complicated case with textured surface requiring refined mesh, and could not handle simulation during multiple periods. To accelerate Picard iteration for predicting dynamics of APP, a GPU-boosted high-performance Multi-working condition joint Analysis Framework (GMAF) is designed, which adopts Preconditioned Conjugate Gradient method (PCG) using Approximate Symmetric Successive Over-Relaxation preconditioner (ASSOR). GMAF abundantly utilizes GPU device via elevating computational intensity and expanding scale of massive parallel computation. Therefore, it possesses novel performance in analyzing dynamics of both smooth and textured APPs during multiple periods, as the establishment and solution to joint algebraic system for pressure field are accelerated magnificently, as well as numerical integral for force and moment due to oil flow. Compared with asynchronized convergence strategy pursuing local convergence, synchronized convergence strategy targeting global convergence is adopted in PCG solver for the joint system. Revealed by corresponding results, oil force in axial direction and moment in circumferential directly respond to input pressure, while other components evolve in sinusoidal patterns. Specifically, force and moment due to normal pressure instantly reach their steady state initially, while ones due to viscous shear stress evolve during periods. After simulating dynamics of APP and pressure distribution via GMAF, the promotion of pressure capacity and torsion resistance due to textured surface is revealed numerically, as several 'steps' exist in the pressure field corresponding to textures.

cs.DC↗