arXiv subjects

Zherui Yang

Publications and source records attributed to Zherui Yang.

15 recordsLinked to original sources

arXiv2026

DSWorld: A Data Science World Model for Efficient Autonomous Agents

Despite strong capabilities in data understanding and decision-making, autonomous data science agents still heavily rely on trial-and-error workflows that involve expensive computation. This bottleneck motivates models that can anticipate the effects of data science operations before real execution. In this paper, we introduce the concept of Data Science World Model, which model the data science execution environment by predicting environment state transitions conditioned on current workflow states and candidate operations. We further propose DSWorld, a practical framework that combines structured state construction, cost-aware routing, lightweight real execution, and an LLM-based simulator for expensive operations. To support training, we construct an 8K-scale transition trajectory dataset and introduce Reflective World Model Optimization, an error-aware reinforcement learning strategy for improving transition prediction. Experiments show that DSWorld accelerates RL-based agent training by approximately $14\times$ and search-based inference by approximately $3$-$6\times$ while maintaining competitive performance, and outperforms the strongest LLM baseline by 35.6% on transition prediction tasks. The code is available at https://anonymous.4open.science/r/DSWorld.

cs.AI↗

arXiv2026

EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management

Recent progress in Large Language Model (LLM) agents has enabled promising advances in automated data science. However, existing approaches remain fundamentally limited by their static action sets and lack of principled long-horizon context management, hindering their ability to accumulate reusable experience across tasks and operate reliably in multi-stage, iterative data science pipelines. To address these challenges, we introduce EvoDS, a self-evolving autonomous data science agent that learns to expand its skills and adaptively managing long-term context through agentic reinforcement learning. Specifically, EvoDS introduces two key strategies: (1) Autonomous Skill Acquisition (ASA) mechanism, which enables agents to synthesize, validate, and reuse executable skills; and (2) Adaptive Context Compression (ACC) strategy, which treats context management as a learned control problem rather than passive truncation. These strategies are orchestrated within a two-stage multi-agent training scheme, enabling EvoDS to autonomously improve over time. Theoretically, we prove that EvoDS's hierarchical design reduces tool-selection error, and its optimization objective aligns with an information bottleneck principle, ensuring efficient context use. Empirically, EvoDS outperforms state-of-the-art open-source data science agents by an average of 28.9% across four diverse benchmarks while eliminating out-of-token failures. Our code and data are available at https://github.com/usail-hkust/EvoDS.

cs.AI↗

arXiv2026

Learning Laplacian Eigenspace with Mass-Aware Neural Operators on Point Clouds

The eigendecomposition of the Laplace--Beltrami Operator (LBO) is fundamental to geometric analysis, yet computing its low-frequency eigenmodes remains a significant bottleneck due to the high cost of iterative solvers on large-scale data. To amortize this cost, we introduce the Neural Eigenspace Operator (NEO), a feed-forward framework designed to predict the spectrum directly from point clouds. Crucially, NEO circumvents the ill-posed nature of standard eigenvector regression, which suffers from intrinsic sign flips and rotation ambiguities, by learning the stable, invariant low-frequency subspace instead. Specifically, the network predicts a redundant set of basis functions whose span robustly covers the target eigenspace, allowing for the recovery of accurate eigenpairs via a lightweight Rayleigh--Ritz refinement. To handle irregular sampling, we propose a mass-aware neural operator that incorporates per-point area weights into attention-based aggregation, improving robustness to non-uniform densities and enabling zero-shot generalization across resolutions. Our approach achieves near-linear runtime scaling and substantial wall-clock speedups over iterative solvers at comparable accuracy, and exhibits strong zero-shot transfer to high-resolution point clouds. The resulting eigenpairs support standard spectral geometry tasks, while the raw basis functions provide effective point-wise features for downstream learning. Code: https://github.com/Adversarr/NEO.

cs.LG↗

arXiv2026

Simple yet Effective: Low-Rank Spatial Attention for Neural Operators

Neural operators have emerged as data-driven surrogates for solving partial differential equations (PDEs), and their success hinges on efficiently modeling the long-range, global coupling among spatial points induced by the underlying physics. In many PDE regimes, the induced global interaction kernels are empirically compressible, exhibiting rapid spectral decay that admits low-rank approximations. We leverage this observation to unify representative global mixing modules in neural operators under a shared low-rank template: compressing high-dimensional pointwise features into a compact latent space, processing global interactions within it, and reconstructing the global context back to spatial points. Guided by this view, we introduce Low-Rank Spatial Attention (LRSA) as a clean and direct instantiation of this template. Crucially, unlike prior approaches that often rely on non-standard aggregation or normalization modules, LRSA is built purely from standard Transformer primitives, i.e., attention, normalization, and feed-forward networks, yielding a concise block that is straightforward to implement and directly compatible with hardware-optimized kernels. In our experiments, such a simple construction is sufficient to achieve high accuracy, yielding an average error reduction of over 17\% relative to second-best methods, while remaining stable and efficient in mixed-precision training.

cs.LG↗

arXiv2026

Efficient and Robust p-type Transistor based on Ultra-wide-bandgap Semiconductor

The p-type transistor is an indispensable component of semiconductor technology, enabling complementary operation with n-channel transistors for computation, storage, and communication. Achieving both high robustness and high efficiency is highly desirable but challenging for p-type transistors due to limited semiconductors with reliable hole transport and their high activation energies. Here, we achieved a robust yet efficient p-type transistor by heterogeneously integrating an ultra-wide-bandgap semiconductor and a high-k dielectric layer through van der Waals integration. The p-type transistor employs a two-dimensional hole channel on hydrogenated diamond (bandgap 5.6 eV) combined with a high-k (30.5) SrTiO3 perovskite membrane. At room temperature, the transistor exhibits stable operation with a high on-current (~200 mA/mm), low subthreshold swing (70 mV/dec), high hole mobility (566 cm^2/Vs to 572 cm^2/Vs) and high on-off ratio (~10^9). Furthermore, tuning annealing temperature allows operation in either enhancement or depletion mode. The robust p-type transistor with high efficiency holds great potential for future power electronics, UV optoelectronics, and harsh-environment electronic applications.

cond-mat.mtrl-sci↗

arXiv2025

Learning Sparse Approximate Inverse Preconditioners for Conjugate Gradient Solvers on GPUs

The conjugate gradient solver (CG) is a prevalent method for solving symmetric and positive definite linear systems Ax=b, where effective preconditioners are crucial for fast convergence. Traditional preconditioners rely on prescribed algorithms to offer rigorous theoretical guarantees, while limiting their ability to exploit optimization from data. Existing learning-based methods often utilize Graph Neural Networks (GNNs) to improve the performance and speed up the construction. However, their reliance on incomplete factorization leads to significant challenges: the associated triangular solve hinders GPU parallelization in practice, and introduces long-range dependencies which are difficult for GNNs to model. To address these issues, we propose a learning-based method to generate GPU-friendly preconditioners, particularly using GNNs to construct Sparse Approximate Inverse (SPAI) preconditioners, which avoids triangular solves and requires only two matrix-vector products at each CG step. The locality of matrix-vector product is compatible with the local propagation mechanism of GNNs. The flexibility of GNNs also allows our approach to be applied in a wide range of scenarios. Furthermore, we introduce a statistics-based scale-invariant loss function. Its design matches CG's property that the convergence rate depends on the condition number, rather than the absolute scale of A, leading to improved performance of the learned preconditioner. Evaluations on three PDE-derived datasets and one synthetic dataset demonstrate that our method outperforms standard preconditioners (Diagonal, IC, and traditional SPAI) and previous learning-based preconditioners on GPUs. We reduce solution time on GPUs by 40%-53% (68%-113% faster), along with better condition numbers and superior generalization performance. Source code available at https://github.com/Adversarr/LearningSparsePreconditioner4GPU

cs.LG↗

arXiv2025

Superconducting Diode Effect in Gradiently Strained Nb0.5Ti0.5N Films

The superconducting diode effect (SDE), combining superconductivity with diode-like nonreciprocal current flow, recently emerges as an ideal candidate for zero-dissipation electronic circuits. Such technologically advantageous diodes are achieved by intricate material engineering to disrupt inversion symmetry, which leads to the production challenges as well as a limited pool of viable materials. Here we exploit the gradient interfacial strain to experimentally induce the SDE in Nb0.5Ti0.5N (NTN) films grown on MgO substrates. Additionally, the SDE is tunable with an in-plane magnetic field and can be further enhanced by introducing an interfacial anisotropic pinning potential. Our findings establish interfacial strain gradient as a versatile tool for creating and enhancing tunable SDE.

cond-mat.supr-con↗

arXiv2025

MM-Agent: LLM as Agents for Real-world Mathematical Modeling Problem

Mathematical modeling is a cornerstone of scientific discovery and engineering practice, enabling the translation of real-world problems into formal systems across domains such as physics, biology, and economics. Unlike mathematical reasoning, which assumes a predefined formulation, modeling requires open-ended problem analysis, abstraction, and principled formalization. While Large Language Models (LLMs) have shown strong reasoning capabilities, they fall short in rigorous model construction, limiting their utility in real-world problem-solving. To this end, we formalize the task of LLM-powered real-world mathematical modeling, where agents must analyze problems, construct domain-appropriate formulations, and generate complete end-to-end solutions. We introduce MM-Bench, a curated benchmark of 111 problems from the Mathematical Contest in Modeling (MCM/ICM), spanning the years 2000 to 2025 and across ten diverse domains such as physics, biology, and economics. To tackle this task, we propose MM-Agent, an expert-inspired framework that decomposes mathematical modeling into four stages: open-ended problem analysis, structured model formulation, computational problem solving, and report generation. Experiments on MM-Bench show that MM-Agent significantly outperforms baseline agents, achieving an 11.88\% improvement over human expert solutions while requiring only 15 minutes and \$0.88 per task using GPT-4o. Furthermore, under official MCM/ICM protocols, MM-Agent assisted two undergraduate teams in winning the Finalist Award (\textbf{top 2.0\% among 27,456 teams}) in MCM/ICM 2025, demonstrating its practical effectiveness as a modeling copilot. Our code is available at https://github.com/usail-hkust/LLM-MM-Agent

cs.AI↗

arXiv2025

SUPRA: Subspace Parameterized Attention for Neural Operator on General Domains

Neural operators are efficient surrogate models for solving partial differential equations (PDEs), but their key components face challenges: (1) in order to improve accuracy, attention mechanisms suffer from computational inefficiency on large-scale meshes, and (2) spectral convolutions rely on the Fast Fourier Transform (FFT) on regular grids and assume a flat geometry, which causes accuracy degradation on irregular domains. To tackle these problems, we regard the matrix-vector operations in the standard attention mechanism on vectors in Euclidean space as bilinear forms and linear operators in vector spaces and generalize the attention mechanism to function spaces. This new attention mechanism is fully equivalent to the standard attention but impossible to compute due to the infinite dimensionality of function spaces. To address this, inspired by model reduction techniques, we propose a Subspace Parameterized Attention (SUPRA) neural operator, which approximates the attention mechanism within a finite-dimensional subspace. To construct a subspace on irregular domains for SUPRA, we propose using the Laplacian eigenfunctions, which naturally adapt to domains' geometry and guarantee the optimal approximation for smooth functions. Experiments show that the SUPRA neural operator reduces error rates by up to 33% on various PDE datasets while maintaining state-of-the-art computational efficiency.

cs.LG↗

arXiv2024

Field-free superconducting diode effect and magnetochiral anisotropy in FeTe0.7Se0.3 junctions with the inherent asymmetric barrier

Nonreciprocal electrical transport, characterized by an asymmetric relationship between current and voltage, plays a crucial role in modern electronic industries. Recent studies have extended this phenomenon to superconductors, introducing the concept of the superconducting diode effect (SDE). The SDE is characterized by unequal critical supercurrents along opposite directions. Due to the requirement on broken inversion symmetry, the SDE is commonly accompanied by electrical magnetochiral anisotropy (eMCA) in the resistive state. Achieving a magnetic field-free SDE with field tunability is pivotal for advancements in superconductor devices. Conventionally, the field-free SDE has been achieved in Josephson junctions by intentionally intercalating an asymmetric barrier layer. Alternatively, internal magnetism was employed. Both approaches pose challenges in the selection of superconductors and fabrication processes, thereby impeding the development of SDE. Here, we present a field-free SDE in FeTe0.7Se0.3 (FTS) junction with eMCA, a phenomenon absent in FTS single nanosheets. The field-free property is associated with the presence of a gradient oxide layer on the upper surface of each FTS nanosheet, while the eMCA is linked to spin-splitting arising from the absence of inversion symmetry. Both the SDE and eMCA respond to magnetic fields with distinct temperature dependencies. This work presents a versatile and straightforward strategy for advancing superconducting electronics.

cond-mat.supr-con↗

arXiv2024

Electrical Control Grain Dimensionality with Multilevel Magnetic Anisotropy

In alignment with the increasing demand for larger storage capacity and longer data retention, electrical control of magnetic anisotropy has been a research focus in the realm of spintronics. Typically, magnetic anisotropy is determined by grain dimensionality, which is set during the fabrication of magnetic thin films. Despite the intrinsic correlation between magnetic anisotropy and grain dimensionality, there is a lack of experimental evidence for electrically controlling grain dimensionality, thereby impeding the efficiency of magnetic anisotropy modulation. Here, we demonstrate an electric field control of grain dimensionality and prove it as the active mechanism for tuning interfacial magnetism. The reduction in grain dimensionality is associated with a transition from ferromagnetic to superparamagnetic behavior. We achieve a non-volatile and reversible modulation of the coercivity in both the ferromagnetic and superparamagnetic regimes. Subsequent electrical and elemental analysis confirms the variation in grain dimensionality upon the application of gate voltages, revealing a transition from a multidomain to a single-domain state accompanied by a reduction in grain dimensionality. Furthermore, we exploit the influence of grain dimensionality on domain wall motion, extending its applicability to multilevel magnetic memory and synaptic devices. Our results provide a strategy for tuning interfacial magnetism through grain size engineering for advancements in high-performance spintronics.

cond-mat.mtrl-sci↗

arXiv2024

Giant third-order nonlinear Hall effect in misfit layer compound (SnS)${1.17}$(NbS$_2$)$_3$

Nonlinear Hall effect (NLHE) holds immense significance in recognizing the band geometry and its potential applications in current rectification. Recent discoveries have expanded the study from second-order to third-order nonlinear Hall effect (THE), which is governed by an intrinsic band geometric quantity called the Berry Connection Polarizability (BCP) tensor. Here we demonstrate a giant THE in a misfit layer compound, (SnS)${1.17}$(NbS$_2$)$_3$. While the THE is prohibited in individual NbS$_2$ and SnS due to the constraints imposed by the crystal symmetry and their band structures, a remarkable THE emerges when a superlattice is formed by introducing a monolayer of SnS. The angular-dependent THE and its scaling relationship indicate that the phenomenon could be correlated to the band geometry modulation, concurrently with the symmetry breaking. The resulting strength of THE is orders of magnitude higher compared to recent studies. Our work illuminates the modulation of structural and electronic geometries for novel quantum phenomena through interface engineering.

cond-mat.mes-hall↗

arXiv2022

Phase diagram of superconducting vortex ratchet motion in a superlattice with noncentrosymmetry

Ratchet motion of superconducting vortices, which is a directional flow of vortices in superconductors, is highly useful for exploring quantum phenomena and developing superconducting devices, such as superconducting diode and microwave antenna. However, because of the challenges in the quantitative characterization of the dynamic motion of vortices, a phase diagram of the vortex ratchet motion is still missing, especially in the superconductors with low dimensional structures. Here we establish a quantitative phase diagram of the vortex ratchet motion in a highly anisotropic superlattice superconductor, (SnS)1.17NbS2, using nonreciprocal magnetotransport. The (SnS)1.17NbS2, which possesses a layered atomic structure and noncentrosymmetry, exhibits nonreciprocal magnetotransport in a magnetic field perpendicular and parallel to the plane, which is considered a manifest of ratchet motion of superconducting vortices. We demonstrated that the ratchet motion is responsive to current excitation, magnetic field and thermal perturbation. Furthermore, we extrapolated a giant nonreciprocal coefficient ({\gamma}), which quantitatively describes the magnitude of the vortex ratchet motion, and eventually established phase diagrams of the ratchet motion of the vortices with a quantitative description. Last, we propose that the ratchet motion originates from the coexistence of pancake vortices (PVs) and Josephson vortices (JVs). The phase diagrams are desirable for controlling the vortex motion in superlattice superconductors and developing next-generation energy-efficient superconducting devices.

cond-mat.supr-con↗

arXiv2017

AstroCloud: A Distributed Cloud Computing and Application Platform for Astronomy

Virtual Observatory (VO) is a data-intensively online astronomical research and education environment, which takes advantages of advanced information technologies to achieve seamless and global access to astronomical information. AstroCloud is a cyber-infrastructure for astronomy research initiated by Chinese Virtual Observatory (China-VO) project, and also a kind of physical distributed platform which integrates lots of tasks such as telescope access proposal management, data archiving, data quality control, data release and open access, cloud based data processing and analysis. It consists of five application channels, i.e. observation, data, tools, cloud and public and is acting as a full lifecycle management system and gateway for astronomical data and telescopes. Physically, the platform is hosted in six cities currently, i.e. Beijing, Nanjing, Shanghai, Kunming, Lijiang and Urumqi, and serving more than 17 thousand users. Achievements from international Virtual Observatories and Cloud Computing are adopted heavily. In the paper, backgrounds of the project, architecture, Cloud Computing environment, key features of the system, current status and future plans are introduced.

astro-ph.IM↗

arXiv2016

Investigating the Impacts of Recommendation Agents on Impulsive Purchase Behaviour

The usage of recommendation agents (RAs) in the online marketplace can help consumers to locate their desired products. RAs can help consumers effectively obtain comprehensive product information and compare their candidate target products. As a result, RAs have affected consumers shopping behaviour. In this study, we investigate the usage and the influence of RAs in the online marketplace. Based on the Stimulus-Organism-Response (SOR) model, we propose that the stimulus of using RAs (informativeness, product search effectiveness and the lack of sociality stress) can affect consumers attitude (perceived control and satisfaction), which further affects their behavioural outcomes like impulsive purchase. We validate this research model with survey data from 157 users of RAs. The data largely support the proposed model and indicate that the RAs can significantly contribute to impulsive purchase behaviour in online marketplaces. Theoretical and practical contributions are discussed.

cs.CY↗