arXiv subjects

Wei Li

Publications and source records attributed to Wei Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

CheckDistill: Adaptive Checklist Distillation for Efficient Distributional Reward Modeling in Image Editing

Reinforcement learning for instruction-guided image editing requires reward models that provide reliable, accurate, and efficient feedback. Direct scalar prediction compresses multidimensional editing preferences, whereas generative evaluators can expose their judgment through explicit reasoning. However, broad scoring criteria do not specify all instance-dependent editing and preservation constraints, and generating long reasoning trajectories is costly for online optimization. We introduce CheckDistill, a pointwise reward modeling framework that internalizes instruction-adaptive checklists. We decompose each editing condition into atomic, independently verifiable requirements and construct four-dimensional scoring labels together with structured verification evidence. After supervised score initialization, a privileged teacher conditions on checklists, item-level judgments, and dimension-level rationales to supervise the student's own scoring trajectories. A reliability-weighted on-policy self-distillation objective adjusts supervision at score tokens according to agreement between the teacher's expected scores and reference labels, while retaining candidate probability weights in a truncated token support. The deployed student directly predicts four score distributions without checklist inputs or explicit reasoning and aggregates their expectations into a scalar reward. CheckDistill achieves aggregate accuracies of 72.30%, 56.67% and 82.90% on MMRB2, MER-Bench and EditReward-Bench, respectively, exceeding the compared open-source baselines on all three metrics.

cs.CV↗

arXiv2026

Weak type $(1,1)$ estimates for noncommutative rough singular integrals

We establish the weak type $(1,1)$ endpoint estimate for operator-valued rough singular integrals with kernels satisfying the classical $L\log L$ integrability and cancellation conditions. This extends Seeger's endpoint theorem [{J. Amer. Math. Soc.} \textbf{9} (1996)] to the noncommutative setting. A key ingredient is an off-diagonal analysis that controls interactions between different kernel scales and avoids an additional logarithmic loss.

math.FA↗

arXiv2026

DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery

Computational drug discovery requires agents to compose scientific operations and adapt to changing evidence and constraints. Here we introduce DrugHarness, an evolvable agent harness coupling composable scientific execution with conditional skill induction. Its environment integrates over 80 scientific operations, biomedical resources and task-specific code, preserving associations among candidates, results and artifacts. From execution trajectories and outcome feedback, DrugHarness induces action hypotheses with observable applicability conditions, recommended actions and supporting evidence. An independent review agent evaluates candidate skills derived from these hypotheses through source verification and, when needed, targeted execution trials. Admitted skills guide subsequent tasks and are revised through feedback. Evaluation on MolBench and SMDD-Bench demonstrates capabilities in molecular screening, optimization and constrained design. Without evolution, DrugHarness exceeds reported MolClaw-CC results by 10.8 percentage points in affinity-comparison accuracy and 55% in mean top-three active hits on MolBench. Retrospective evaluation on skill-development tasks shows that evolution increases Qwen's affinity-comparison accuracy from 62.2% to 81.1% and optimization success from 79.5% to 87.2%. In a prospective HPK1 study, DrugHarness achieves a more favorable median docking score than Biomni among system-specific candidates. A synthesized compound achieves a cellular IC50 of 14.19 nM, a 2.64-fold potency improvement over the best input reference, with activity competitive with two selected compounds designed by experienced medicinal chemists. These results support conditional skill evolution for scientific agents and demonstrate a workflow connecting computational drug design with experimental validation.

cs.AI↗

arXiv2026

Using Heteroscedasticity for Causal Discovery under Hidden Confounding

The presence of hidden confounding in observational studies poses a fundamental challenge to the identification and estimation of causal structures encoded by directed acyclic graphs (DAGs). While instrumental variables (IVs) have been widely used to address hidden confounding, most existing methods require valid instruments, which may be unavailable in practice. In this paper, we develop a novel framework for causal discovery and inference in DAGs that accommodates both hidden confounding and invalid IVs by exploiting heteroscedasticity. We establish a novel linear characterization of leaf nodes through heteroscedasticity-induced relationships between cross-covariance and self-covariance vectors, enabling recursive recovery of the causal order without requiring any candidate IV to be valid. Given the identified order, we then sequentially construct surrogate IVs to identify the causal effects and directions, thereby recovering the entire DAG. We also propose a data-adaptive procedure for estimating the DAG and causal effects. We show that the proposed method consistently recovers the underlying DAG and yields asymptotically normal estimators of causal effects. Extensive simulation studies demonstrate substantial improvements over existing methods. We further illustrate the practical utility of the proposed method through an application to gene regulatory network analysis.

stat.ME↗

arXiv2026

Differential equations as difference-differential equations

We analyze the behavior of systems of algebraic differential equations when considered as systems of difference-differential equations, with special emphasis on systems which define minimal sets relative to the theory $\operatorname{DCF}$ of differentially closed fields of characteristic zero with $m$ distinguished commuting derivations. We show that if $p$ is a minimal type relative to $\operatorname{DCF}$ defined by a system of algebraic partial differential equations and inequations and the forking geometry on $p$ is trivial (also called ``disintegrated''), then $p$ remains minimal when regarded as a partial type relative to the theory $\operatorname{DCFA}$ of difference-differentially closed fields of characteristic zero with $m$ commuting derivations in the sense that every completion has $\operatorname{SU}$-rank one. We illustrate this theorem by describing in detail the possible difference-differential equations consistent with differential equations of the form $y' = y^3 + a_2 y^2 + a_1 y + a_0$ where $a_0$, $a_1$, and $a_2$ are differential constants.

math.LO↗

arXiv2026

BluePixel: Toward Native 4K Agentic Image Editing

Instruction-driven image editing has reached native 4K, yet how such editors are trained remains undisclosed, and pretraining one from scratch is beyond most teams. The task itself compounds the difficulty: a single end-to-end editor cannot make precise parametric adjustments, match specialized experts, or interpret the vague instructions users actually write. We argue that 4K editing is a system problem rather than a single-model problem, and present BluePixel, which couples a native-4K editing model with an agent harness; the complete training recipe is disclosed in this paper. The model makes 4K training practical: it turns a 1K-pretrained editor into a native-4K one through progressive VAE adaptation that keeps 4K training as affordable as 1K, absorbs a long tail of vertical editing skills through continual learning, and refines the aesthetic-fidelity trade-off with reinforcement learning. Distillation via source-anchored target projection (SATP) then compresses 50 denoising steps to 4, completing a 4K edit in 8.7 s on a single Kunlunxin P800 XPU while preserving task success rate and background consistency. The harness, BluePixel Agent, makes editing controllable: it decomposes user instructions into specialist and general operations, verifies every result with the same reward model that guided training, and retries within bounded attempts. On VINS-4KEval++ (3,305 native-4K cases), BluePixel ranks second among seven models, behind only GPT Image 2 in non-edited-region consistency.

cs.CV↗

arXiv2026

CommerceGen-Studio: An End-to-End Framework for Commercial-Grade Main-Image Generation

We present CommerceGen-Studio, an end-to-end framework for generating commercial-grade main images for e-commerce Stock Keeping Units (SKUs). Our framework is organized along a three-stage pipeline of information extraction, semantic rewriting, and multi-condition generation: Prompt-Eval-Refine-Loop Agent (PERL-Agent) distills heterogeneous SKU metadata into concise selling-point text lists through an iterative closed loop of VLM and LLM roles scored by a reference-free rubric. Layout Thinker Optimization (LTO) aligns the rewriter with both caption and rendered-image quality via alternating domain-specific policy optimization. SKU-DiT, trained under Reinforce-Distill-Reinforce (RDR) progressive training, conditions jointly on the white-background product image, the trademark crop, and the structured layout to hold subject fidelity, trademark integrity, and character-level text accuracy together. The framework runs in production at a scale of millions of main images per day with a latency of a few seconds per image, which we believe makes it among the first systems of this kind to operate at commercial scale. We also release CommerceGen-Bench, a benchmark of hard cases collected from live traffic across all product categories. On it, our framework achieves the highest overall scores among the evaluated open-source baselines on both editing tracks, with and without prompt enhancement. It also matches much larger closed-source models, and beats several of them at Instruction Following and Product Integration - the two properties that determine whether an image can be published.

cs.CV↗

arXiv2026

I2O2I-Nav: A Benchmark with a Modular Agent for Long-Horizon Indoor-to-Outdoor-to-Indoor Navigation

Service and delivery robots often need to complete long-horizon navigation missions spanning indoor and outdoor environments. Existing embodied navigation benchmarks primarily evaluate isolated stages or partial cross-domain journeys, providing limited support for assessing how heterogeneous navigation capabilities compose under continuous execution. We introduce I2O2I-Nav, a physics-enabled benchmark of 20 real-world composite scenes spanning over 115,000 square meters across multiple buildings and floors, with aligned 3DGS rendering and collision geometry. Its three-level evaluation progresses from independent primitives to two-stage compositions and complete four-stage I2O2I missions. Each stage starts from its predecessor's actual endpoint, while stage-conditional metrics track completion and localize failures along the mission. We further propose the MODULAR-Agent that coordinates frozen navigation policies through completion verification, stage handoffs, and bounded recovery. Evaluation across 1,374 episodes highlights the difficulty of continuous composition. The MODULAR-Agent achieves 40.5-76% success on individual primitives, but full I2O2I success remains low. Compared with a fixed policy chain using the same underlying policies, it more than doubles success on both Indoor-to-Outdoor (I2O) and Outdoor-to-Indoor (O2I), while improving full-mission success from 1.15% to 4.02%. Stage-wise analysis shows that the MODULAR-Agent enables more missions to reach the final object-navigation stage, where conditional success remains low, revealing challenges in both long-horizon stage execution and completion decisions.

cs.RO↗

arXiv2026

HyBRA-EEG: Hyperbolic Global-Local Brain-Region Aggregation for Self-Supervised EEG Representation Learning

Electroencephalography (EEG) representation learning is challenged by noisy signals, inter-subject variability, and heterogeneous electrode montages in different datasets and recording settings. Recent self-supervised EEG pretraining methods improve transferability by converting heterogeneous recordings into unified token or sequence representations, but channel-aware designs may still encode montage- or dataset-specific regularities. We propose HyBRA, a self-supervised framework based on hyperbolic global-local brain-region aggregation. HyBRA maps heterogeneous channel observations into a fixed regional token space by combining electrode coordinates, region priors, and feature-dependent hyperbolic distances. A channel-evidence injection branch preserves fine-grained discriminative cues during aggregation. The resulting tokens are processed by a Region-Time Transformer and pretrained with masked reconstruction of latent features and region-level signals. HyBRA outperforms the compared methods in average balanced accuracy on six benchmark datasets. Under a controlled six-fold evaluation on OpenBMI-MI, HyBRA matches CSBrain in balanced accuracy, improves AUROC by 1.84 percentage points and AUPRC by 5.23 percentage points, and uses 96.8\% fewer parameters. Matched comparisons on TUAB and TUEV show higher balanced accuracy for hyperbolic matching than for Euclidean distance or cosine similarity.

cs.LG↗

arXiv2026

PrivacyGuard: Source-Linked Sanitization and Locally Resolvable PII References for Cloud GUI Agents

Cloud graphical user interface (GUI) agents send instructions and screenshots to remote planners, exposing personally identifiable information (PII). Existing placeholder defenses mask detected spans and reuse exact strings. Fragmented or changed forms can expose PII or break entity continuity across turns. PrivacyGuard addresses this private reference problem with source-linked sanitization (SAN) and a structured reference lifecycle (REF). SAN reconstructs fragmented or overlapping evidence and rewrites the complete source extent. REF links changed appearances to session entities and supports locally resolvable references without returning raw values to the upload path. In five-turn sessions scored against common gold, PrivacyGuard reduces audited source-value leakage by 33.55\%--82.60\% relative to the closest prior full-pipeline system in all four reader--language combinations. REF raises strict cross-form continuity from 0.04 under exact-value reuse to 0.81. In reference-required Android tasks, verified completion rises from 8/30 and 7/30 to 17/30 and 20/30. On AndroidLab, PrivacyGuard achieves the highest or tied-highest task-success point estimate among protection pipelines for all four planners. On PrivScreen, it reduces fixed-reader field exposure by 62.72\%--62.84\% relative to the same baseline. PrivacyGuard lowers upload exposure while preserving a locally resolvable task path.

cs.CR↗

arXiv2026

Dynamical chiral symmetry breaking in rotating QED$_3$: critical flavor number and the occupation-reweighting channel

We study dynamical chiral symmetry breaking in $(2{+}1)$-dimensional QED on a rigidly rotating disc, $ΩR<1$, within the rainbow Schwinger--Dyson framework. At zero temperature the rotation response of the condensate vanishes identically, so $N_c^{\rm SD}(Ω,T{=}0)=N_c^{\rm SD}(0)=2.60\pm0.01$ (using three operational definitions). At finite temperature, rotation suppresses the condensate through a dominant occupation-reweighting channel, $\hatκ_{\rm occ}=-1.55$, accounting for $\ge 99.995\%$ of the linearized total response at the reference parameters, with self-consistent feedback amplifying it by $26.5$. The screening channel is reported at fixed regularization, $\hatκ_{\rm scr}(\varepsilon = 0.01)=-0.97$; its $\varepsilon \to 0$ behavior is logarithmically divergent. The crossover flavor $N^{*}(Ω)$ falls monotonically at fixed temperature (a corollary). The critical flavor number is unchanged by rotation at $T=0$ (theorem). Together these results set the finite-temperature reference for the model.

cond-mat.str-el↗

arXiv2026

Approximating the Rural Postman Problem below 3/2

We give an approximation algorithm for the rural postman problem with approximation ratio strictly smaller than $3/2$. We obtain this result by adapting to the rural postman problem the technique of sampling from maximum entropy distributions for the metric traveling salesman problem of Karlin, Klein, and Oveis Gharan. We also observe that, for every fixed $\varepsilon>0$, any $α$-approximation algorithm for the metric traveling salesman problem yields an $(α+\varepsilon)$-approximation algorithm for the rural postman problem; this implication is already implicit in the treatment of edges that must be traversed in the work of Lampis on the inapproximability of the traveling salesman problem.

cs.DS↗

arXiv2026

Learning Representations to Guide Pixel Diffusion Transformers

Visual representations are largely developed for understanding, yet what makes them useful for generation remains unclear. We study this question in pixel diffusion transformers and uncover two key insights. First, better representation quality does not necessarily translate into greater generative utility, as externally aligned semantics can be progressively lost during pixel synthesis. Second, architectural separation with specialized pixel decoders mainly redistributes semantic and spatial modeling, without necessarily yielding representations better suited to generation. These insights suggest that a useful generative representation must be both properly formed and actively exploited. Motivated by these insights, we introduce Representation Guidance (RG), which directly forms and exploits internal representations while retaining a plain Transformer architecture. A low-dimensional bottleneck forces generation-relevant information into a compact representation, which is then explicitly used to guide decoder generation by modifying the flow-matching target according to its induced prediction offset. RG learns clearer semantic and spatial structure while scaling efficiently in both parameters and computation. On ImageNet 256x256, RG-L/16 with only 459M parameters outperforms substantially larger pixel diffusion transformers, while RG-XL/16 achieves an FID of 1.55. RG further reaches an FID of 1.74 at 512x512 and extends to text-to-image generation, achieving 0.89 on GenEval and 87.1 on DPG-Bench.

cs.CV↗

arXiv2026

SharP: Soft and Hard Prompt-Guided Augmentation with LLM for Low Resource Fake News Detection

Fake news detection under low resource conditions is challenged by the scarcity of labeled data and the difficulty of capturing subtle deceptive patterns. In this work, we focus on low resource fake news detection, where only a small labeled set is available for training, while evaluation is performed under a fixed and unchanged test protocol. This setting is motivated by costly, ambiguous fact-checking and emerging topics and domains that are poorly covered by existing benchmarks. Existing data augmentation methods, such as synonym substitution, generative adversarial networks, and large language model (LLM)-based generation, often produce fluent but overly generic content, lacking the task-specific relevance needed for accurate detection. To address this, we propose SharP, a boundary-aware text generation framework guided by soft and hard prompts, which leverages LLMs as tools adapted to produce task-specific synthetic samples. SharP employs a dual prompt mechanism: hard prompts provide structured guidance and factual constraints, while soft prompts introduce learned semantic patterns from the data. These two types of prompts are combined to guide the generation process, enabling the model to produce samples that closely align with the characteristics of news. To further enhance detection performance, we incorporate a boundary-aware strategy that steers generation toward areas where the classifier is less confident, helping to clarify subtle distinctions between real and fake content. In addition, we adopt a dual objective optimization that balances semantic alignment with the source data and classifier feedback. This encourages the generation of samples that are both domain-consistent and helpful for refining decision boundaries, ultimately improving model robustness and generalization in low resource fake news detection.

cs.CL↗

arXiv2026

From band reconstruction to Bogoliubov dispersion: How dz2-band enhances iron-based superconductivity

Conventional understanding of iron-based superconductivity is usually established based on electronic bands at the Fermi level, leaving deeper bands overlooked as potential factor. Here, we expand this view by demonstrating that a deep-lying dz2 band below the Fermi level can be employed to enhance the superconducting paring strength. By using scanning tunneling microscope to mechanically change the crystal lattice of an iron-based superconductor, we observe an upward shift of the deep-lying dz2 band related feature, leading to hybridization with the primary superconducting bands at the Fermi level. We visualize the novel evidence of this band hybridization and resulting enhanced superconducting gap through quasiparticle interference imaging, providing conclusive evidence that engineering such correlations between deep and pairing-active bands is a potent mechanism for amplifying the superconducting gap. This work refines the conventional view of pairing-relevant electronic structures and establishes band hybridization as an effective way to optimize superconducting materials.

cond-mat.supr-con↗

arXiv2026

SDFlow: Similarity-Driven Flow Matching for Time Series Generation

Vector quantization (VQ) with autoregressive (AR) token modeling is a widely adopted and highly competitive paradigm for time-series generation. However, such models are fundamentally limited by exposure bias: during inference, errors can accumulate across sequential predictions, leading to pronounced quality degradation in long-horizon generation. To address this, we propose SDFlow ($\textbf{S}$imilarity-$\textbf{D}$riven $\textbf{Flow}$ Matching), a non-autoregressive framework that operates entirely in the frozen VQ latent space and enables parallel sequence generation via flow matching. We tackle three key challenges in making this transition: (1) eliminating exposure bias by replacing step-wise token prediction with a global transport map; (2) mitigating the high-dimensionality of VQ token spaces via a low-rank manifold decomposition with a learned anchor prior over the latent manifold; and (3) incorporating discrete supervision into continuous transport dynamics by introducing a categorical posterior over codebook indices within a variational flow-matching formulation. Extensive experiments show that SDFlow achieves state-of-the-art performance, improving Discriminative Score and substantially reducing Context-FID, particularly for challenging long-sequence generation. Moreover, SDFlow provides significant inference speedups over autoregressive baselines, offering both high fidelity and computational efficiency. Code is available at https://github.com/William-Liwei/SDFlow

cs.AI↗

arXiv2026

The maximum size of simple solid matching covered graphs

A connected graph with at least two vertices is matching covered if each of its edges is contained in a perfect matching. A matching covered graph is solid if every separating cut in it is a tight cut. A matching covered graph which is free of nontrivial tight cuts is a brick if it is nonbipartite. Every bipartite matching covered graph is solid. Lucchesi and Murty conjectured that there exists a positive integer $N$ such that, for every integer $n\ge N$, the maximum number of edges in a simple solid matching covered graph on $2n$ vertices is $n^2$. In this paper, we disprove this Conjecture, and show that the maximum number of edges of a simple solid matching covered graph on $2n$ ($n\ge2$) vertices is $n^2+2$. Moreover, we characterize the graphs attaining this bound. In addition, we prove that every simple solid brick of order $2n$ has at most $n^2$ edges for $n\ge4$.

math.CO↗

arXiv2026

Unimodality of Forest Independence Polynomials

For a finite forest $F$ let $i_k(F)$ be the number of independent sets of $F$ with $k$ vertices. Zhang and Li proved that the sequence $i_0(F),i_1(F),\dots,i_{α(F)}(F)$ is unimodal for every finite forest $F$, which answers Erdős Problem 993. We give a second proof. It starts from their decomposition relative to a fixed independent set and from the bounds of Zhang and Li and of Fang, Lu, Nevo, Yao and Zheng that confine a valley of the sequence to an explicit window of ranks. For a forest with at least $25$ vertices, one moment argument excludes a valley at every rank of the window: at the activity where the hard-core mean equals the rank, the size of a random independent set is a mixture of binomial laws over an independent set of maximum weight, a valley is a moment inequality for this mixture, and it is excluded by duality given three bounds that hold for every forest, on the variance of the number of free vertices and on its Laplace transforms, and on the variance ratio. The variance bound is proved by hand up to finitely many interval checks and the other two bounds are verified by computer on finite interval-arithmetic coverings; on the resulting parameter domain a valley is excluded by exact tests on finitely many rational boxes while the mean number of free vertices is below an explicit starting mean between $19$ and $50$, and above it by one inequality, with explicit constants, for the fibers of a weighted valley kernel, proved by hand up to a finite list of explicit checks and averaged over the mixture. Forests with at most $24$ vertices are treated by exact counting, by hand except for exact rational evaluations of two explicit formulas at $43$ parameter triples. No forest is enumerated. A formal proof of the theorem in Lean 4 accompanies the paper.

math.CO↗