arXiv subjects

Xiaojun Wu

Publications and source records attributed to Xiaojun Wu.

At least 19 recordsLinked to original sources

arXiv2026

Spin-Density-Driven Hybrid Spin Splitting in Collinear Compensated Magnets

Nonrelativistic spin splitting in collinear compensated magnets is constrained by spin-group symmetry, with distinct alternating and Zeeman-type splitting typically associated with altermagnets and fully compensated ferrimagnets, respectively. From a spin-density perspective, we propose a class of hybrid compensated magnets in which alternating-like and Zeeman-type spin splitting coexist within a single collinear compensated state. Using a minimal two-sector model, we demonstrate that this coexistence arises from competition between directional sublattice dispersion and the sector-dependent energy biases arising from the same structural inequivalence. First-principles calculations provide a proof of concept in a crossed bilayer of unequal-width Janus graphene nanoribbons, where the structurally inequivalent ribbons retain approximately symmetry-related spin-density distributions. The valence and conduction sectors support direction-selective spin transport and energy-selective spin-polarized states, respectively. These results connect the spatial organization of spin density with the electronic response to structural inequivalence, providing a design principle for hybrid spin splitting in compensated magnets.

cond-mat.mtrl-sci↗

arXiv2026

FARO: Function-Space Adaptive Return-to-Risk Optimization via Cross-Block Equal-Risk-Contribution Budget Allocation

Training a large language model is compute-bound, and current recipes divide the work between optimizers by hand: orthogonalized momentum for attention and MLP matrices, Adam-style updates for the rest. This division rests on human priors about which layer type deserves which geometry, and it fixes a single global update budget that every parameter block shares. We introduce FARO, a function-space adaptive optimizer that makes the partition of parameters into blocks the unit of decision. FARO gives each block a geometry by its realized counterfactual return, then freezes the routes and reallocates the frozen per-block budget by cross-block equal risk contribution (ERC), the risk-parity principle from portfolio theory. On a GPT-style model trained on FineWeb, the framework gives two results. First, its FARO-Adaptive arm reaches 3.2780 nats at the full schedule, matching NorMuonAndAdam (3.2794), the best hand-designed recipe: it recovers the human prior without hand-setting per-layer geometry. Second, FARO-ERC improves on that arm: it lowers the clean-data loss at every one of 6 seeds, by 0.0136 nats on average, a modest but robust gain. Its sign test, paired test, and confidence interval all agree, and it exceeds the measured run-to-run noise at no added time or memory. It works by trimming the riskiest blocks rather than expanding the safest, and it holds up under a corrupted stream, still lowering the mean loss and winning the seed majority. The source code is available at https://github.com/wxj630/FARO.

cs.LG↗

arXiv2026

ColaBench: Evaluating Memory-Guided Knowledge Acquisition in Personal Agents

Personal agents need to draw on long-term interaction history to retrieve and apply external knowledge suited to a user's current request. Even an answer grounded in relevant documents can be unsuitable when it overlooks the user's circumstances. This creates two connected challenges: using historical conditions to identify the knowledge needed for a new request, and applying that knowledge to the user's situation. Existing benchmarks advance memory recall, personalized retrieval, and memory-guided action, while the joint evaluation of these two challenges remains underexplored. We introduce ColaBench, a synthetic benchmark in health and daily living, with 349 tasks across 10 Profiles and a knowledge base of 13,033 documents. Its Evidence Graph connects personal conditions to their disclosures across public interaction sessions, records prior document exposure, and links answer requirements to personal and newly acquired document evidence. We also introduce MemKB, which uses state summaries to inform queries and rule application, and memory-document links to retrieve knowledge associated with the user's history. MemKB achieves 86.25% Accuracy with an average of 4.07 interaction rounds per task. Task-level analysis finds larger gains on tasks integrating information across multiple historical Facts; error analysis further identifies unsupported personal assumptions and rule misapplication as obstacles to using retrieved evidence. ColaBench provides traceable tasks and an evaluation framework for studying how memory guides knowledge acquisition and rule application.

cs.AI↗

arXiv2026

Electric-Field-Switchable Altermagnetism via Ligand Rotation in a d0 Metal-Organic Framework

Altermagnetism combines vanishing net magnetization with momentum-dependent spin splitting. While altermagnetic signatures have been identified in metal-organic frameworks (MOFs), non-invasive reversible control remains challenging. Here we report a route to electric-field-switchable altermagnetism in a flexible d0-MOF, in which polar substituents couple the external field to ligand rotation. In the planar ground state, pz orbitals mediate 180-degree ligand-metal-ligand superexchange (J2), stabilizing antiferromagnetic order. Rotating ligands out of plane activates px/py orbitals, enhances 90-degree pathways (J1), and reverses J1/J2 competition. The antiferromagnetic-to-altermagnetic crossover occurs at approx. 34 degrees, with spin splitting of 84 meV at 90 degrees. The 90-degree altermagnetic state is switched on by the electric field and switched off upon field removal, as the constrained-rotation landscape decreases monotonically back toward 0 degrees, enabling barrierless relaxation to the antiferromagnetic state and reversible on/off operation. Ab initio molecular dynamics at 300 K shows thermal fluctuations alone cannot synchronize ligand rotation. Polar substituents provide the required coupling for electric-field control; for fluorinated substituents, the critical field is approx. 0.22 V/Angstrom, comparable to field-effect devices. This work establishes ligand p orbitals, rather than transition-metal d orbitals, as the origin of altermagnetic order and shows that polar substituents enable electric-field control for organic spintronics.

cond-mat.mtrl-sci↗

arXiv2026

Nef but Non-Semi-positive Line Bundles on Hopf Manifolds

We give a complete characterization of the nef, effective, pseudo-effective, and semi-positive cones on Hopf manifolds. In particular, by computing the Ueda classes on non-diagonal Hopf surfaces, we show that the invariant elliptic curve is nef but not semi-positive. Via Bott--Chern cohomology calculations, we construct new examples of nef but non-semi-positive line bundles on Hopf manifolds of arbitrary dimension. We also prove that a minimal compact complex surface $X$ containing an elliptic curve $C$ and a nef line bundle $L$ such that $(X,C,L)$ has finite generalized Ueda type is either a Hopf surface or Serre's example.

math.CV↗

arXiv2026

DynSTEER: Dynamic Stage-wise Trajectory Evaluation and Execution-time Review for Agents

Large language model agents are increasingly deployed for long-horizon task execution, raising a central granularity question for trajectory evaluation: whole-trajectory verification is too coarse to capture concrete failures and their associated evidence in long trajectories, while atomic-step scoring is too fine-grained, noise-sensitive, and computationally expensive. This granularity gap makes a single-reference trajectory paradigm inadequate for assessing the rich space of valid agent execution paths and delays timely feedback and early stopping in long-horizon tasks. To address these issues, we propose DynSTEER, a dynamic stage-wise framework for agent trajectory evaluation. DynSTEER bridges the granularity gap through stage-wise dynamic evaluation that segments rollouts at key execution nodes and adapts its multi-level review strategy based on stage-level results; it compiles a path-tolerant milestone graph from available task inputs to preserve diverse legal paths without reference leakage; and it supports terminating unrecoverable agent executions to curb resource waste. Experimental results show that DynSTEER improves evaluation discriminability by over 85\% compared with whole-trajectory evaluation and saves 17.74\% of execution steps. The code is available at https://github.com/zhichao-stone/DynSTEER

cs.AI↗

arXiv2026

Beauville--Bogomolov--Yau decomposition for Kähler generalized pairs

In this paper, we establish the Beauville--Bogomolov--Yau decomposition for generalized pairs in the Kähler setting, thereby extending this structure theorem to the natural framework of the Kähler generalized Minimal Model Program. More precisely, we prove that, after passing to a finite quasi-étale cover, a Kähler generalized klt pair $(X,Δ,\boldsymbolβ)$ of Calabi--Yau type admits a locally constant fibration $(X,Δ)\to Y$ over a Calabi--Yau variety $Y$ whose fiber $(F,Δ|_{F})$ is rationally connected. Equivalently, after base change to the universal cover of $Y$, the fibration becomes a product, and its global structure is determined by a monodromy action preserving the pair $(F,Δ|_{F})$. As a principal application, when the nef b-part $\boldsymbolβ$ vanishes, we show that the monodromy can be eliminated after a further finite quasi-étale cover, yielding a product decomposition into a rationally connected pair, strict Calabi--Yau varieties, irreducible holomorphic symplectic varieties, and complex tori. The proof introduces new analytic methods involving relative projectivity, localized positivity and flatness of direct image sheaves, generalized pairs in the analytic Minimal Model Program, and foliations on singular Kähler varieties.

math.AG↗

arXiv2026

Clinician-Friendly Foundation Models for Ophthalmic Image Diagnostics without Fine-Tuning or Technical Barriers

Artificial intelligence (AI) shows remarkable potential in medical imaging diagnostics, yet most current models require retraining when applied across different clinical settings, limiting their scalability. We developed GlobeReady, a deployment-oriented platform powered by the RetiGlobe foun- dation model and local feature augmentation. RetiGlobe was pretrained in two stages: 1) self-supervised learning using DINOv2 on 38 million synthetic ophthalmic images, and 2) contrastive learning using CLIP on 475,845 real image-text pairs spanning diverse ethnicities, imaging devices, and geographic regions worldwide. We evaluate GlobeReady on 488,448 ophthalmic images, including color fundus photographs (CFPs) and optical coherence tomography scans, from multi-centres in China, Singapore, Vietnam and the UK. Prospective testing included usability assessment with 31 ophthalmologists. Exploratory analyses evaluated domain generalisability, Bayesian uncertainty quantification, out-of-distribution (OOD) detection, and feature-based case retrieval.

cs.CV↗

arXiv2026

DataFoundry: Evolving Data Preparators via Recursive Self-Improvement

Domain adaptation of large language models increasingly depends on constructing high-quality training data, yet existing data-preparation pipelines typically address quality only after generation through post-hoc filtering. This creates a fundamental mismatch: data-quality issues often originate from the construction process itself, while quality control is applied only to its outputs. We introduce \textsc{DataFoundry}, a framework for \textbf{evolving data preparators through recursive self-improvement} before large-scale data production. \textsc{DataFoundry} represents a data preparator as an evolvable runtime specification and instantiates its evolution with a \textsc{Skills-as-Modules} architecture, in which a central \textsc{Controller} orchestrates modular skills to compile executable runtimes, diagnose deficiencies on small pilot sets using domain-appropriate criteria, and translate diagnostic feedback into adapters that revise individual preparation components while preserving stable interfaces. We evaluate \textsc{DataFoundry} on DataPrep-Bench across mathematics, finance, law, and medicine, and find that recursively evolved preparators produce training data with higher downstream utility than baselines. Experiments across different backbones further demonstrate that these improvements are not tied to a particular model, while analyses and case studies further reveal the framework's optimization dynamics and illustrate how its evolution unfolds in practice.

cs.CL↗

arXiv2026

Refined Chern characteristic classes of blow-ups

We prove a blow-up formula for refined Chern characteristic classes of compact complex manifolds. To this end, we establish a version of Riemann--Roch without denominators for the refined Chern characteristic classes. In particular, as an application, we study the behaviour of the refined Chern characteristic classes of the Iwasawa manifold under a blow-up transformation.

math.AG↗

arXiv2026

Harmonic Hierarchy of Altermagnetic Spin Splitting from Symmetry-Adapted Wavefunctions

Altermagnets combine magnetic compensation with spin-momentum-locked splitting in the absence of spin-orbit coupling, yet existing descriptions, formulated primarily in terms of spin symmetry and lattice geometry, provide limited insight into the electric-structure perspective of its angular harmonic form. Here, we identify a wavefunction-level framework for altermagnetism in two-dimensional square lattices. Using symmetry-adapted polynomial wavefunctions, we show that the harmonic structure of momentum-space spin splitting is inherited from the geometry of the electronic wavefunctions which can be selected by crystal fields. Identical orbital sectors preserve conventional antiferromagnetic degeneracy, whereas intertwined linear and quadratic wavefunctions generate d-wave and g-wave altermagnetic anisotropies, respectively. Tight-binding analysis connects this hierarchy to inequivalent same-spin hopping channels. First-principles calculations on the g-wave mcm-type reticular material platforms confirm high-symmetry-linear degeneracy together with finite generic-k splitting. Our results establish a hierarchy linking wavefunction geometry, orbital realization, microscopic hopping anisotropy, and altermagnetic electronic structure.

cond-mat.mtrl-sci↗

arXiv2026

Envs-FORGE: Frontier-Optimized Reward-Grounded Environment Synthesis for Agent RL

Reinforcement learning (RL) for terminal agents needs executable training environments with reliable rewards and useful difficulty. Fixed recipes such as few-shot, Self-Instruct, and Evol-Instruct apply the same prompting policy to every seed, even when the current policy would benefit from a harder, easier, or simply different task. We present Envs-FORGE, a prompting policy that converts verifier rewards into per-seed environment-synthesis actions. Envs-FORGE estimates seed pass rates, scores six projection--direction actions around a target learning frontier, and solves a per-seed mixed-integer linear program (MILP) to choose the action that conditions generation. The selected action drives synchronized rewriting of the instruction, fixtures, oracle solution, tests, and Docker environment; only gold-verified bundles enter RL training. The indexed MILP form also supports optional soft skill coverage for portfolio planning. On Qwen 3.5 35B, Envs-FORGE improves Pass@1 over Base by 9.2 percentage points on tb-core (40.0% to 49.2%) and 6.4 points on tb-2.0 (23.0% to 29.4%), exceeding the strongest fixed-recipe baseline by 2.4 and 2.1 points. It reaches 77.1% on SWE-bench Verified versus 73.4% for Base, and improves tb-core by 6.8--9.2 points across the evaluated 4B--35B models. All synthesis methods export 100 verified environments and use 2.27M--2.88M synthesis tokens, placing the comparison at the same downstream training-set size and the same operational scale. The source code is available at https://github.com/DataArcTech/DataArc-SynData-Toolkit/.

cs.CL↗

arXiv2026

LazyTrain: Limited-resource Allocation toward Zero-waste Yield Optimization in Large Language Model Training

Training large language models on limited hardware is increasingly a scheduling problem across GPU compute, host memory, PCIe transfer, and storage bandwidth. Existing offloading systems reduce GPU residency, and MegaTrain shows that a CPU-master layer-streaming executor can train large models on a single GPU, but fixed checkpointing and placement heuristics still leave communication exposed on the critical path. We propose LazyTrain, an optimization layer over a layer-streaming executor. LazyTrain formulates checkpoint selection, activation placement, recomputation, and CPU-GPU-NVMe communication overlap as a mixed-integer scheduling problem, then executes the solved policy during training. It further couples 8-bit optimizer states with fast gradient clipping as a single Hybrid 8-bit operator: state compression reduces optimizer-state memory, while fast clipping counteracts the additional CPU-side update overhead. Across H800 experiments from Qwen2.5-3B to Qwen3.6-27B, LazyTrain improves sustained TFLOPS over matched baselines runs by approximately 1.24$\times$; RTX 3090 experiments likewise increase the maximum feasible batch size by one at each model scale. In the primary Qwen3.6-27B H800 MetaMathQA run, LazyTrain reaches 219.95 TFLOPS and 1361 tokens/s at batch size 72, peaks at 68.84\,GB of GPU memory, and obtains 95.42\% exact-match accuracy on the full evaluation split. The source code is available at https://github.com/DataArcTech/LazyTrain.

cs.CL↗

arXiv2026

Bayesian-Agent: Posterior-Guided Skill Evolution Across LLM Agent Harnesses

LLM agents increasingly rely on prompts, tools, memory, SOPs, skills, and harness feedback, yet current self-evolution pipelines often update these assets through heuristic reflection or raw success counts. Such updates are brittle when trajectories are sparse, expensive, and context-dependent. We introduce Bayesian-Agent, a native and cross-harness framework that treats reusable agent skills as Bayesian evidence objects. Bayesian-Agent records verified trajectories, maintains posterior beliefs over skill reliability and failure modes, and turns those beliefs into auditable skill actions and model-facing guardrails. This posterior view provides a finite-sample alternative to raw empirical-rate skill updates and frames prompt, context, and harness engineering as inference over the external decision environment. On RealFin-Bench, Bayesian skill evolution matches or improves the raw empirical-rate control and yields large gains on native runs. In the incremental mode, incremental repair improves SOP-Bench from 80\% to 95\%, Lifelong AgentBench from 90\% to 100\%, and RealFin-Bench from 45\% to 65\%. Backend and model-scaling ablations further show that posterior-guided repair can operate across BA native, MiniSWEAgent, and Claude Code, provided the harness produces verifiable task artifacts. The source code is available at https://github.com/DataArcTech/Bayesian-Agent.

cs.CL↗

arXiv2026

Geometry-Driven Magnetoelectric Coupling in Two-Dimensional Compensated Ferrimagnets

The magnetoelectric coupling in compensated magnets enables stray-field-free manipulation of spin-splitting, holding great promise for spintronics, but inherently hindered by the symmetry mismatch between spatial-inversion-broken ferroelectricity and time-reversal-broken spin states. Here, based on a symmetry-decoupled analysis of magnetoelectric coupling in compensated magnets, we establish a geometry-driven spin-ferroelectric coupling mechanism in bilayer breathing kagome lattices. Within this geometric framework interlocking the out-of-plane electric polarization with cooperative intralayer structural distortions, we demonstrate that polarization switching drives a deterministic reversal of the global spin splitting. First-principles calculations on a prototype bilayer Nb3Cl8 successfully validate this mechanism, demonstrating the switching of spin-splitting states through an energetically feasible, asynchronous layer-by-layer transition pathway. Our proposed coupling originates from lattice geometry and structural symmetry, establishing a unique route toward switchable spin splitting in compensated ferrimagnets.

cond-mat.mtrl-sci↗

arXiv2026

UniTacVLA: Unified Tactile Understanding and Prediction in Vision Language Action Models

Vision-language-action (VLA) models have achieved strong performance in many robotic manipulation tasks, yet remain limited in contact-rich dexterous manipulation. To overcome this limitation, recent vision-tactile-language-action (VTLA) methods incorporate tactile sensing into VLA models to provide direct contact information. However, they typically treat tactile signals as passive auxiliary inputs, making it difficult to model tactile semantics and future physical interactions. To this end, we propose a unified tactile learning framework for contact-rich manipulation that models tactile signals as dynamic interaction cues for both contact understanding and prediction. Specifically, we construct a unified tactile latent space and jointly model current tactile states and future contact changes through tactile chain-of-thought reasoning and coarse-to-fine future tactile prediction, thereby forming a state-aware and dynamics-aware tactile prior. Based on this prior, we introduce a tactile-action mixed controller that combines real-time and predicted tactile feedback to refine low-frequency action chunks with high-frequency corrections. Real-world experiments on four categories of contact-rich tasks, including adjustment, insertion, wiping, and assembly, under both clean and externally perturbed settings, show that our method improves success rate, manipulation accuracy, and contact robustness over existing methods, demonstrating its effectiveness in dexterous physical interaction.

cs.RO↗

arXiv2026

Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning

Large language models (LLMs) demonstrate remarkable reasoning capabilities in tasks such as algorithmic coding and mathematical problem-solving. Recent methods have improved reasoning through expanded corpus and multistage training combining reinforcement learning and supervised fine-tuning. Although some methods suggest that small but targeted dataset can incentivize reasoning via only distillation, a reasoning scaling laws is still taking shape, increasing computational costs. To address this, we propose a data-efficient distillation framework (DED) that optimizes the Pareto frontier of reasoning distillation. Inspired by the on-policy learning and diverse roll-out strategies of reinforcement learning, the key idea of our approach is threefold: (1) We identify that benchmark scores alone do not determine an effective teacher model. Through comprehensive comparisons of leading reasoning LLMs, we develop a method to select an optimal teacher model. (2) While scaling distillation can enhance reasoning, it often degrades out-of-domain performance. A carefully curated, smaller corpus achieves a balanced trade-off between in-domain and out-of-domain capabilities. (3) Diverse reasoning trajectories encourage the student model to develop robust reasoning skills. We validate our method through evaluations on mathematical reasoning (AIME 2024/2025, MATH-500) and code generation (LiveCodeBench), achieving state-of-the-art results with only 0.8k carefully curated examples, bypassing the need for extensive scaling. Our systematic analysis demonstrates that DED outperforms existing methods by considering factors beyond superficial hardness, token length, or teacher model capability. This work offers a practical and efficient pathway to advanced reasoning while preserving general capabilities.

cs.LG↗

arXiv2026

Towards Robust EEG Decoding Based on Riemannian Self-Attention

Brain-Computer Interface (BCI) based on electroencephalography (EEG) enables direct interaction between the brain and external environments and has significant applications in assistive technologies, medical rehabilitation, and entertainment. Recently, EEG decoding methods based on Symmetric Positive Definite (SPD) learning have demonstrated superior performance. However, these methods typically employ basic network architectures and do not explicitly capture local relationships between EEG signals. This limitation is problematic for EEG signals due to their inherently low Signal-to-Noise Ratio (SNR). Moreover, most existing Riemannian manifold-based methods are restricted to specific metrics. The most widely used is the Affine-Invariant Metric (AIM). However, it has a quadratic dependency on the SPD matrices and cannot handle ill-conditioned SPD matrices, which hinders the effectiveness of networks. In contrast, the Bures-Wasserstein Metric (BWM) exhibits linear dependence on SPD matrices and demonstrates superior performance for ill conditioning. To overcome these challenges, we propose a Riemannian self-attention network based on the BWM. Additionally, the recently introduced power-deformed generalized Bures-Wasserstein metric reveals a nonlinear relationship between SPD matrices and matrix power deformation. This metric provides a more nuanced representation of the geometric structure of the SPD manifold. Consequently, we extend our model to a learnable version. For simplicity, we refer to it as GBWAtt. Experimental results on three EEG benchmarking datasets validate the robustness and effectiveness of our proposed method. The code is available at https://github.com/jissc/GBWAtt.

cs.LG↗