arXiv subjects

Han Wu

Publications and source records attributed to Han Wu.

At least 19 recordsLinked to original sources

arXiv2026

Skill$^{\infty}$: Recursive Self-Improvement for Agent Skill Optimization

Recursive self-improvement (RSI), in which a system also improves the way it improves, is a long-standing aspiration on the path to AGI. Language agents increasingly store procedural knowledge in reusable skills. Optimizing skills offers a direct way to achieve RSI, because the optimization procedure itself can be encoded as a skill. Existing skill optimizers, however, follow a hand-designed procedure. Even methods that evolve a meta-skill confine it to a fixed pipeline and relearn it for each task. We present Skill$^{\infty}$, which makes skill optimization recursively self-improving: a meta agent follows a meta-skill to optimize task skills and optimizes the meta-skill itself using the same procedure. The optimization is agent-led. The meta agent diagnoses execution traces, proposes candidates, requests independent reviews, and allocates the search budget. The meta-skill defines these steps, allowing self-improvement to rewrite them. This meta-skill is also decoupled from task skills, refined on multiple skill optimization tasks and then reused to optimize task skills across diverse domains. As in deployment, skills at both levels are standard packages loaded through native skill discovery. On 83 SkillsBench tasks, Skill$^{\infty}$ reaches a pass rate of 60.7%, compared with 56.4% for human-curated skills and 54.5% for SkillEvolver.

cs.AI↗

arXiv2026

GeoReason: Bridging Logical Reasoning and Spatial Fidelity in Remote Sensing Segmentation

Multimodal Large Language Models (MLLMs) have significantly advanced instruction-driven perception through the "Embedding-as-Mask" paradigm. However, extending this capability to Remote Sensing (RS) reasoning segmentation remains exceptionally challenging under cluttered geographical contexts and extreme scale variations, often leading to Spatial Fidelity Degradation and Target Misidentification. We identify a critical yet underexplored bottleneck: the sequential alienation of the segmentation token [SEG] from its initial visual grounding within the autoregressive generation paradigm. Produced only at the end of long reasoning chains, [SEG] relies on visual evidence that has been progressively diluted by linguistic abstraction and autoregressive noise. This information decay is particularly catastrophic for RS targets with minuscule pixel footprints, where even marginal fidelity loss leads to localization failure. To address these bottlenecks, we propose GeoReason, a novel framework designed to fortify logical deduction and preserve spatial integrity. First, we introduce an Anticipatory Prior mechanism that injects the [SEG] token directly into the initial user query, shifting it from a terminal output to a primary condition. This paradigm shift transforms the token into a persistent spatial anchor, preventing information decay during long-form linguistic inference. Second, we enhance visual reasoning via a Location-aware Chain-of-Thought (CoT) that enforces coarse-to-fine localization using bounding boxes, coupled with a Latent Patch-to-Token Alignment (PTA) Loss that explicitly aligns the latent segmentation token with physical RS textures. Extensive experiments demonstrate that GeoReason consistently outperforms state-of-the-art methods on the Earthreason and RRSIS-D benchmarks. Code is available at https://github.com/Fanc0126/GeoReason.

cs.CV↗

arXiv2026

Interlock: From Local Correctness to Task Completion in Long-Horizon Agents

Large language model agents increasingly undertake long-horizon tasks in complex environments characterized by fragmented evidence, partially completed states, and dependencies across resources and artifacts. Existing agents may satisfy individual requirements while failing to maintain these relationships, producing locally correct but incomplete outcomes. These failures arise because task completion depends on relationships among environmental evidence, consequential decisions, and required effects. To address this completion gap, we propose Interlock, a framework for task synthesis that represents each task as a typed completion graph. Grounding edges connect public evidence to decisions, coupling hyperedges impose joint constraints across components, and realization edges connect decisions to all required downstream effects. Interlock compiles this graph into a resettable workspace, a public request, and an evaluation rubric. A source-coverage review checks whether the original requirements survive generation, while a public-input review checks whether the rubric is supported by the materials available to the solver. An agent-as-judge inspects execution artifacts and runs verification scripts, retaining trajectories judged to satisfy all mandatory requirements. To evaluate training utility, we fine-tune two Qwen3.6 backbones on their own accepted trajectories. Across the two backbones, training improves AutomationBench task pass rates by 4.00-4.50 percentage points, JobBench weighted scores by 3.20-6.64 points, and Harvey LAB criterion pass rates by 3.89-4.16 percentage points. These results demonstrate the training utility of the synthesized experience across external task environments.

cs.CL↗

arXiv2026

ClawLoop: Self-Improving Agents via Harness-Native Training Loop

The rapid advancement of Large Language Models (LLMs) has driven their deployment toward general-purpose harness environments, where LLM-based agents operate in sandbox workspaces and interact directly with file systems and shells. While numerous post-training pipelines exist for tool-use agents, their counterparts for harness-based agents remain scarce and are predominantly one-shot and teacher-dependent: a stronger external model synthesizes tasks and collects trajectories, the target agent is trained once, and evaluation results are consumed by human researchers rather than fed back into the loop. Such pipelines are inherently bounded by the capability of the external model and the scope of human prior, leaving no room for self-improvement. To address these limitations, we propose ClawLoop, a framework that enables self-improving without any external teacher model and with minimal prior dependencies. ClawLoop composes three phases into a self-sustaining cycle: Task Synthesis, which generates a diverse pool of training tasks via two complementary construction routines, applies diagnosis-conditioned difficulty growth followed by both agent-free and agent-based task validation; Training, which updates the agent via SFT and/or RL; and Diagnosis and Feedback, which analyzes the agent's rollout behaviors to produce a capability analysis report that will be used as feedback for task synthesis in the next iteration. Using Qwen3.6-27B as the loop model, ClawLoop yields consistent improvements on SkillsBench, Claw-Eval-Live, and ClawGym-Bench. The iteratively refined tasks are further transferred to smaller-scale Qwen3.5-4B and 9B models with gains across all benchmarks, validating data quality independently of the loop model.

cs.AI↗

arXiv2026

Distributed Hypothesis Testing Against Dependence

We study distributed hypothesis testing and establish the exact error exponent in single-letter form for new testing problems. In distributed hypothesis testing, a receiver decides between $\mathcal{H}_0:P_{XY}$ and $\mathcal{H}_1:Q_{XY}$ based on $Y^n$ and a rate-limited description of $X^n$. So far, such single-letter forms are known only for testing against independence, studied by Ahlswede and Csiszár, and testing against conditional independence, studied by Rahman and Wagner. In this paper, we study testing against dependence, where $P_{XY}=P_XP_Y$, and show that its error exponent is given by Han's exponent, which is established by single-letterizing a multi-letter version of Han's exponent. We also disprove a previous conjecture of Han claiming that the error exponent is given by the lautum information, as we show that Han's exponent can strictly exceed the latter. We then consider the Cartesian product of testing against dependence and testing against independence, and derive a single-letter characterization of its error exponent. We next study testing against conditional dependence, which is the dependence-testing counterpart of the setting studied by Rahman and Wagner. We derive a single-letter converse bound for the error exponent by introducing and solving a related setting where the side information is also available to the transmitter. We show that our converse bound is tight in some cases by using a conditional-coding version of the quantization scheme. Finally, we extend our converse method for testing against dependence to the general setting, and recover a recently established single-letter converse bound.

cs.IT↗

arXiv2026

Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning

Processing long visual token sequences from high-resolution images makes multi-step reasoning computationally expensive for multimodal Large Language Models (MLLMs). Existing one-shot pruning and aggregation methods compress visual tokens into a fixed context before decoding. However, visual evidence needs can shift as reasoning unfolds, making it difficult for a fixed compressed context to retain all the details needed across stages. To address this challenge, we propose ViMoD, a lightweight framework that maintains a compact visual context while preserving access to original fine-grained evidence as reasoning needs evolve. Deformable Aggregation of Region-wise Tokens (DART) learns content-adaptive groups and aggregation capacities, constructing compact Coarse representations linked to recoverable original Fine tokens. Temporal Routing for Adaptive Contextual Evidence (TRACE) integrates decoding history to anticipate upcoming evidence needs and select, retain, or replace active Fine-token groups. Selected Fine tokens augment the persistent Coarse context in the frozen backbone, enabling stage-specific evidence access without continuously attending to all visual tokens. On Qwen3-VL-4B, ViMoD outperforms all evaluated baselines on all eight reasoning benchmarks at a 20% target visual token budget, improving the mean normalized score by 39.0% over the strongest evaluated one-shot baseline. These gains are achieved with only 0.0546% additional trainable parameters relative to the frozen backbone.

cs.CV↗

arXiv2026

Voronoi-Hankel Transforms

Let $π$ be a generic irreducible representation of $\mathrm{GL}_n(\mathbf{F})$ for a local field $\mathbf{F}$. We introduce the Voronoi--Hankel transform $\mathcal{VH}_π$ for $π$, and prove it is equivalent to the $π$-Fourier transform introduced by Jiang--Luo. We give an extension $\widetilde{\mathcal{VH}}_π$ of $\mathcal{VH}_π$ to some larger class of functions on $\mathbf{F}^{\times}$, and prove the multiplicativity with respect to parabolic induction via the above-mentioned equivalence. For non-archimedean $\mathbf{F}$ we give a systematical study of the relevant kernel function, including the asymptotic behavior at $0$ and $\infty$, as well as some simple integral representation based on the local Langlands correspondences for the essentially tame supercuspidals. As an application in the non-archimedean case, we give an effective version of the stability theorem for the twisted local gamma factors.

math.NT↗

arXiv2026

Active Feature Acquisition With Incomplete Training Data

In many prediction tasks, acquiring all features can be a prohibitively expensive or outright impossible task. Further, in many cases a static subset of features may not be enough to solve the problem sufficiently across various instances. Active Feature Acquisition (AFA) addresses these problems by formalizing the trade-off between feature cost and predictive performance during sequential feature selection. However, prior AFA work largely assumes access to complete training data, an assumption that is often violated in practice. Here we study AFA with Incomplete Training Data (AFA-ITD), showing that under missing completely at random (MCAR) data, one-step acquisition values remain unchanged, whereas multi-step values can decrease. We analyze three approaches to learning from incomplete data: aliasing, filtering, and generative restoration. We show that filtering can require a number of training instances scaling exponentially with the dimension, whereas generative restoration scales exponentially with the acquisition budget. We empirically test our theory on a controlled experiment and across common AFA datasets and find that missingness mainly damages methods that exploit multi-step acquisitions and that generative restoration is able to recover lost performance in many experiments. Code is available at https://github.com/Linusaronsson/AFA-Benchmark/tree/missing-data.

cs.LG↗

arXiv2026

High-Temperature ferromagnetism from site-selective filling in (Fe,Ni)$_{6-δ}$GeTe$_2$

The discovery of high-temperature ferromagnetism in the metallic van der Waals (vdW) system Fe$_N$GeTe$_2$ has brought two-dimensional (2D) magnets into technologically relevant temperature scales. Specifically at N = 5, dilution of magnetic moments by nickel substitution counterintuitively achieves a record high Curie temperature of 478~K. Unraveling the origin of this nickel-substitution-induced enhancement is complicated by the compound's structural complexity, coexistent itinerant and local magnetic contributions, and mesoscopic compositional domains. Through coordinated structural and electronic characterization, we identify that the high-T$_C$ magnetic phase arises from a strain-stabilized Fe$_6$GeTe$_2$ nano-precipitate. Combining first-principles calculations and spin- and angle-resolved photoemission spectroscopy (ARPES), we uncover a site-specific electronic landscape in which interior iron atoms primarily host localized moments while the outer iron atoms neighboring the tellurium layers produce spin-polarized itinerant carriers that cross the vdW gap. The large energy cost associated with homogeneous nickel substitution is found to favor the spontaneous precipitation of the crystallographically and electronically ``clean'' high-T$_C$ phase. Finally, we compare metal-rich vdW magnets with binary magnetic alloys, and discuss the unifying roles of nano-precipitates in stabilizing otherwise unattainable bulk phases. Our work provides mechanistic insights into the record-high T$_C$ ferromagnetism in (Fe,Ni)$_{5+δ}$GeTe$_2$, establishing a rigorous foundation for the atomic engineering of vdW magnetic metals informed by direct electronic signatures.

cond-mat.mtrl-sci↗

arXiv2026

Comparing local-global obstructions on algebraic stacks

Let k be a number field. For quotient stacks [X/G] with X a smooth quasi-projective geometrically integral k-variety and G a linear k-group, we extend several relations between local-global obstructions previously known for varieties, showing that étale-Brauer obstruction is the finest among various obstructions (such as iterated descent, finite descent, descent, Brauer-Manin).

math.AG↗

arXiv2026

A Real-Time Tsetlin Machine-based Non-intrusive Load Monitoring System on MCUs

Non-Intrusive Load Monitoring (NILM) systems estimate individual appliance energy consumption from a single aggregate meter, without requiring separate sensors for each device. By installing a single meter that measures a building's total electricity consumption, NILM algorithms can determine the active status of each appliance. However, traditional NILM systems use computationally intensive optimization algorithms to process offline data, limiting their capability for on-device deployment, where sensitive household data must be processed locally. This paper proposes a Tsetlin Machine (TM)-based NILM framework, targeting real-time applications on resource-constrained microcontrollers (MCUs), enabling privacy-preserving edge deployment. The problem is reformulated as a classification task, and the proposed approach achieves an average precision of 90% and recall of 96% for two-appliance classification, and 77% precision and 80% recall for four appliances on the REDD dataset. The trained model occupies only 17 KB of flash memory and achieves an inference latency of 0.43 ms on an ESP32, demonstrating its suitability for NILM applications on MCUs.

cs.LG↗

arXiv2026

Taming foundation model with invariance-oriented pre-training for broad-spectrum EEG analysis across signal-level, brain-state, and brain-health tasks

Electroencephalography (EEG) is a widely used window into human brain function, but most EEG models remain tied to a one-dataset-one-model supervised paradigm. Recent EEG foundation models offer a route toward reusable representations, but most remain reconstruction-centered, assuming that EEG content predictable from local context is necessarily transferable neural information. Here we present INCEPT, an invariance-oriented EEG foundation model trained on over 11,000 hours of unlabelled clinical EEG. Rather than prioritizing signal recovery alone, INCEPT learns representation-level stability across correlated EEG observations, separating stable neural structure and essential subject-sensitive information from the nuisance variability that dominates scalp recordings while preserving subject-, state- and condition-discriminative information. We evaluate INCEPT on a broad-spectrum benchmark of ten datasets spanning three levels of post-acquisition EEG analysis: signal-level assessment, brain-state decoding, and brain-health evaluation. INCEPT ranks first among recent EEG foundation models on 26 of 30 linear-probing metrics and 24 of 30 fine-tuning metrics, and also surpasses strong task-specific specialist encoders across diverse downstream settings. Objective ablations and representation analyses further show that invariance-oriented pre-training improves transfer and organizes subject-sensitive neural representations beyond reconstruction alone. These results establish invariance learning as a promising principle for building reusable EEG foundation models.

cs.LG↗

arXiv2026

RoRA: Role-Oriented Regional Allocation for Visual Token Pruning in MLLMs

Multimodal large language models (MLLMs) encode images as long visual token sequences, making prefilling and KV-cache storage expensive. Existing training-free pruning methods select tokens by importance, diversity, or spatial coverage, but treat retained tokens as interchangeable and do not explicitly track which object-related regions are already covered. We present RoRA, a training-free framework that casts visual token pruning as role-oriented regional evidence allocation. Given a fixed budget, RoRA partitions tokens into a protected semantic core, complementary context, and fine-grained detail. It first calibrates text-conditioned attention with a positional prior and a prompt-calibrated object prior, then builds Attention-Anchored Regions (AARs) from high-confidence anchors as lightweight proxies for covered object support. Context is explored mainly outside AARs, while a small AAR-guided budget restores local detail; pairwise similarity is used only for context-stage redundancy filtering. Under matched budgets, RoRA consistently outperforms strong training-free baselines across LLaVA and Qwen-VL families, retaining most of the unpruned accuracy even at aggressive pruning ratios, e.g., 96.5% of full performance at 88.9% pruning on LLaVA-1.5, and improving over D2Pruner by about 5% on Qwen3-VL at 75-90% pruning. At a 66.7% pruning ratio, RoRA requires only 0.7 ms for token selection and reduces end-to-end inference time by 24.6%, corresponding to a 1.33x speedup over unpruned inference on an NVIDIA H800.

cs.CV↗

arXiv2026

LeanMem: Simple and Efficient Long-Term Memory for LLM Agents

Long-term memory is essential for LLM-based agents to sustain interactions and reliably leverage distant history. However, existing memory systems typically process heterogeneous dialogue content through a uniform summarization and retrieval pipeline, leading to either excessive token consumption or irreversible loss of fine-grained evidence. We argue that historical dialogue content should be handled differently according to its compressibility, temporal dynamics, and fidelity requirements. Based on this insight, we propose LeanMem, a lightweight long-term memory framework. LeanMem first filters out low-value content, then stores informative segments as compact profile memory, temporally structured event memory, or source-grounded record memory, depending on the nature of the information. During maintenance, only dynamically evolving event memories are selectively updated, avoiding redundant consolidation of stable profiles and immutable records. During inference, LeanMem dynamically selects memory types and allocates retrieval budgets according to query-specific evidence demands, assembling relevant evidence on demand. On LoCoMo and LongMemEval-S with GPT-4.1-mini and Qwen3-8B, LeanMem improves accuracy over the strongest memory-based baseline in every setting, by up to 15.1 points, at the lowest or near-lowest construction cost, inference tokens, and latency. The code and datasets are included in the supplementary materials.

cs.AI↗

arXiv2026

A Unified Kullback--Leibler Divergence Analysis of Generative Diffusion Models via Entropy Production Rate

We introduce a unified framework for the error analysis of generative models based on the entropy production rate of the forward-reverse diffusion process pair. For a pair of continuity equation flows, the rate admits a closed velocity form identity whose time integral decomposes the terminal Kullback--Leibler (KL) divergence into the sum of an initialization error, a score approximation error, and a time-discretization error. By analyzing the entropy production at the level of marginal distributions, rather than in path space, our framework yields a sharp convergence rate of $\mathcal{O}(h^2)$ for the Euler-Maruyama sampler, where $h$ is the step size. This improves upon the $\mathcal{O}(h)$ rates typically obtained from Girsanov's path-space analyses. Furthermore, our framework unifies the analysis of score-based SDEs, probability-flow ODEs, and stochastic interpolants by varying diffusion coefficients within a single inequality, revealing the trade-off between deterministic and stochastic sampling. Numerical experiments confirm the predicted scaling with step size and terminal time.

math.NA↗

arXiv2026

FilmBench: A Film-Grade Benchmark for Cinematic Video Generation

Progress in video generation keeps narrowing the visual gap between AI-generated and professionally produced footage, yet most benchmarks still draw prompts from web sources or LLM templates and score them with untrained, generic multimodal models. More fundamentally, their evaluation taxonomies remain rudimentary (overall visual quality, coarse text alignment and temporal smoothness) rather than the professional Cinematic Language criteria by which films are actually made and judged, so they assess basic video plausibility rather than film-grade craft. We introduce FilmBench, a text-to-video (T2V) and reference-to-video (R2V) benchmark grounded in the professional Cinematic Language of the film- academy tradition and co-developed with directors and faculty from the Beijing Film Academy and the Hujing Digital Media & Entertainment Group film studio. It rests on three choices. First, prompts are reverse-engineered from clips of award-winning films spanning 20 cinematic genres and chosen by professional directors, so every prompt is anchored to a verified live-action reference; the prompts follow real shot lists, and most script multiple shots (1,056 of the 1,169 prompts are multi-shot), unlike prior single-clip benchmarks. Second, evaluation follows a three-level Cinematic taxonomy of 3 axes, 12 components and 35 (T2V) +3 (R2V-only) sub-metrics. Third, we develop an in-house expert-grade automatic evaluation agent and open-source its core suite of Cinematic Language operators (FilmOps). Benchmarking leading video generation models (9 for T2V, 7 for R2V), the evaluator reproduces the human model ranking at model-level Spearman \r{ho} = 0.95 (T2V) and 0.96 (R2V). Scores fall well below prior web-style benchmarks, with two consistent gaps in dynamic aesthetics and a marked single- to multi-shot performance drop that widens for weaker models.

cs.CV↗

arXiv2026

Mid- and long-wavelength infrared computational ghost spectroscopy

Spectral-domain ghost imaging enables high-resolution spectroscopy with a single-pixel detector. The technique does not rely on spectrally resolved detectors, which makes it inherently robust against turbulence and particularly adapted to weak-light conditions. These features are very attractive for spectral imaging in the mid-infrared region which hosts numerous molecular absorption features but lacks highly sensitive detectors. The implementation of spectral ghost imaging in the mid-infrared has however been limited by the absence of suitable light sources and detectors capable of generating and measuring spectral fluctuations in real time. Here, we demonstrate spectral-domain computational ghost imaging in the mid-infrared based on a nonlinear frequency downconversion scheme. Pre-programmed spectral patterns imposed on broadband light at 1.5 mm using a programmable spectral filter are transferred into the mid-infrared through difference-frequency generation in a nonlinear crystal. This enables computational ghost spectroscopy with a spectral resolution of 0.62 cm-1 using a single-pixel mid-infrared detector. The method is flexible, broadly applicable and, as proof of concept, we demonstrate ghost spectroscopy in mid-wavelength infrared and long-wavelength infrared bands using the nonlinear frequency conversion in chirped-poling lithium niobate and ZnGeP2 crystals, respectively. Our approach provides a new avenue for mid-infrared spectroscopy, remote sensing and spectral imaging.

physics.optics↗

arXiv2026

From Question Answering to Task Completion: A Survey on Agent System and Harness Design

LLM-based agents mark a shift from passive question answering to active task completion: they perceive environments, invoke tools, maintain state, and act over extended horizons. As agent systems have evolved from prompt engineering to workflows and context engineering, harness engineering, and agent-native training with co-evolution, a central question has become increasingly important: where does the bottleneck in agent performance reside, in the foundation model, in the execution harness, or in the coupling between them? This survey examines LLM-based agents through a model-harness lens. We first clarify the functional definition of agents and the implementation view of an LLM-based agent as a foundation model coupled with an execution harness. We then analyze the limits of model-centric scaling, trace four paradigms of agent engineering, and decompose the execution harness into six coupled runtime responsibilities: observation, context, control, action, state, and verification. Using this decomposition, we map task properties and domain pressures to harness configurations, review benchmark and evaluation practices, and synthesize model-harness evidence on how runtime design affects long-horizon task completion, efficiency, and reliability. Finally, we identify open challenges in value-aware evaluation, safety, harness generalization, and model-harness co-evolution. Rather than treating agents as models with auxiliary tools, this survey argues that agent quality -- including success, efficiency, safety, and generalization -- emerges from the interaction between model capability, runtime infrastructure, task structure, and evaluation design. A collection of papers discussed in this survey is provided in https://github.com/ggjy/Awesome-Agent-Engineering.

cs.AI↗