Search arXiv⌕ Search

arXiv subjects

Xiang Zhang

Publications and source records attributed to Xiang Zhang.

At least 19 recordsLinked to original sources
arXiv2026

System noise temperature measurements of a 19-element phased array feed

We present multi-epoch system noise temperature measurements of a 19 element hexagonal room-temperature phased array feed (PAF) using a modified Y-factor method. A Python-based pipeline with iterative channel calibration and maxSNR beam forming is developed to do data reduction, and a covariance-based decomposition scheme is proposed to extract mutual coupling noise temperature Tmc. Long-term observations from March to June 2026 capture temporal receiver noise drift mainly induced by LNA temperature variations. Within the 1.37-1.47 GHz operating band, the measured beam equivalent system noise temperature of the PAF is about 73-89 K. The measured Tmc baseline ranges from about 2 K to 8 K, while under thermal loading, the resonant peak of Tmc can increase to more than 20 K. The baseline of Tmc is consistent with the S-parameter mea surements, which show the Tmc of the full array ranges less than 10.7 K. The baseline fluctuation of Tmc most likely arises from the superposition of thermal and non-thermal components of the spatial electromagnetic fields. A detachable stainless steel shield is deployed to quantify sidelobe spillover noise, and comparative tests show unshielded en vironmental interference increases receiver noise temperature by over 60 K. The proposed measurements and data processing framework provides a practical reference for noise per formance characterization of ground-based radio telescope PAF prototypes.

astro-ph.IM↗
arXiv2026

Position: Agentic Evolution is the Path to Evolving LLMs

As Large Language Models (LLMs) move from curated training sets into open-ended real-world environments, a fundamental limitation emerges: static training cannot keep pace with continual deployment environment change. Scaling training-time and inference-time compute improves static capability but does not close this train-deploy gap. We argue that addressing this limitation requires a new scaling axis-evolution. Existing deployment-time adaptation methods, whether parametric fine-tuning or heuristic memory accumulation, lack the strategic agency needed to diagnose failures and produce durable improvements. Our position is that agentic evolution represents the inevitable future of LLM adaptation, elevating evolution itself from a fixed pipeline to an autonomous evolver agent. We instantiate this vision in a general framework, A-Evolve, which treats deployment-time improvement as a deliberate, goal-directed optimization process over persistent system state. We further propose the evolution-scaling hypothesis: the capacity for adaptation scales with the compute allocated to evolution, positioning agentic evolution as a scalable path toward sustained, open-ended adaptation in the real world. Our code is publicly available at https://github.com/A-EVO-Lab/a-evolve.

cs.AI↗
arXiv2026

Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents

LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear whether a model's base capability in task-solving predicts its capabilities in harness self-evolution: which models produce useful harness updates, and which actually benefit from them? We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving. Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains; even Qwen3.5-9B's updates yield gains comparable to those of Claude Opus~4.6. Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier. We trace low gains at the weak tier to two failure modes: weak-tier models may fail to activate relevant harness artifacts, or activate them but fail to follow them faithfully. These findings suggest investing capability budget in the task-solving agent rather than the evolver, and targeting harness invocation and long-horizon instruction following in agent training. Our source code is publicly available at https://github.com/A-EVO-Lab/a-evolve/tree/release/harness-evolution.

cs.AI↗
arXiv2026

αDepth: Learning Scene-Level Soft Boundary Decomposition for Stereo Conversion

Accurately modeling soft boundaries, e.g., hair and defocus blur, is a fundamental challenge in stereo conversion due to the ambiguous blending of foreground and background. Existing depth models primarily predict single-layer depth, leading to ambiguity in depth correspondence at soft boundaries. While matting techniques can capture opacity for layered modeling, they often struggle in complex scenes with multiple targets and usually require user intervention. This paper introduces αDepth, a layered representation that decomposes soft boundaries for high-fidelity stereo conversion. Specifically, we first resolve mixed color and depth ambiguity by estimating layered color and depth values at soft boundaries. Considering complex multi-target scenes, we design a Circular Alpha Representation (CAR) that shifts the paradigm from global target extraction to local boundary decomposition. Unlike prior matting methods restricted to a single foreground/background, CAR enables efficient scene-level inference without manual guidance. Extensive evaluations demonstrate that αDepth achieves state-of-the-art performance in stereo conversion, eliminating background bleeding and structural distortions at soft boundaries.

cs.CV↗
arXiv2026

Language Models Learn to Improve Language Models

Large language models (LLMs) are increasingly used to automate AI research, including language-model (LM) development, offering a potential path toward recursive self-improvement. However, turning experimental feedback into a learned ability to propose effective improvements across models faces two challenges. First, training such an ability with real feedback is expensive: reinforcement learning (RL) requires evaluating tens of thousands of proposals, each potentially requiring a full model-training run. Even at ten GPU-hours per run, reward collection alone would consume hundreds of thousands of GPU-hours, motivating a cheaper surrogate. Second, surrogate feedback can become unreliable as policy optimization favors overestimated proposals when the proposal distribution shifts. To address these challenges, we propose LM2LM, which trains LLMs with RL to generate effective LM improvement proposals. To provide scalable feedback, we use a research world model (RWM), a frozen LLM that predicts proposal gains from real experimental records provided in context, as the reward model for RL. To maintain feedback reliability, we periodically execute current-policy proposals ranked highly by the RWM and add their measured outcomes to its context. Across eight autoregressive and diffusion LM development environments, real-execution evaluations on held-out tasks show that LM2LM enables Qwen3.8-27B to outperform GPT-6-sol, even when GPT-6-sol is given 300 real experimental records in context, while requiring 33% lower data-generation cost. LM2LM also generalizes to a model family unseen during training.

cs.CL↗
arXiv2026

AR-RRG: Anatomy-Guided Retrieval-Driven Radiology Report Generation

Radiology report generation (RRG) streamlines clinical workflows and alleviates the burden of radiologists' documentation. Standard vision-language models capture global visual cues and lack fine-grained alignment between anatomical regions and diagnostic text. Most retrieval-augmented methods retrieve complete historical reports, introducing redundant noise and severe vision-text mismatches that trigger clinically risky hallucinations and missed lesions. In this paper, we introduce AR-RRG, an anatomy-guided retrieval-driven method for the generation of radiology reports. AR-RRG consists of two core modules: the anatomy-aware self-training lesion detection network (ASLDNet) and the retrieval-driven radiology report generation (RRRG). ASLDNet leverages anatomy-aware spatial verification (ASV) and class-adaptive dynamic thresholding (CDT) via self-training, delivering accurate lesion localization to support subsequent generation. As the generative branch of AR-RRG, RRRG eliminates clinical hallucinations through dual-granularity retrieval (DR) and complements missing lesions with a neighborhood consensus (NC) mechanism. A parallel continuous visual perception (PCVP) stream converts radiographs into visual tokens through fixed medical encoding and resampling. Multi-scale text and visual features are fused, followed by LoRA-efficient large language model tuning to generate clinically consistent, visually grounded radiology reports. Extensive evaluations on MS-CXR, MIMIC-CXR, and IU X-Ray demonstrate that AR-RRG consistently outperforms state-of-the-art methods across mainstream metrics. Our code will be made publicly available.

cs.CV↗
arXiv2026

EquiContact: A Hierarchical SE(3) Vision-to-Force Equivariant Framework for Spatially Generalizable Contact-rich Tasks

This paper presents a framework for learning vision-based robotic policies for contact-rich manipulation tasks, which generalize spatially across task configurations. We propose EquiContact, a novel hierarchical framework composed of a high-level vision planner, such as a Diffusion Equivariant Descriptor Field (Diff-EDF) and a low-level compliant visuomotor policy, such as a Geometric Compliant Action Chunking Transformer (G-CompACT). G-CompACT, operating using localized observations from wrist-mounted RGB images, end-effector pose feedback in the form of a geometrically consistent error vector (GCEV) and force-torque readings, produces actions defined on the end-effector frame. We show that the entire EquiContact pipeline is $\SE$-equivariant under equivariant reference-frame estimation and left-invariant visual features, from perception to force control. Once the robot learns to execute a contact-rich task, such as inserting a peg-in-hole (PiH), it can execute similar tasks at different locations and orientations without retraining of the low-level policy. We also describe three key components for attaining spatially generalizable contact-rich policies: geometrically consistent compliance, the use of localized policies, and induced equivariance. Real-world peg-in-hole (PiH) experiments demonstrate near-perfect success rates under spatially out-of-distribution configurations, validating the proposed framework and principles, while additional screwing and surface-wiping experiments evaluate the transfer of the localized policy design beyond insertion. The experimental videos and more details can be found on the project website: https://equicontact.github.io/EquiContact-website/

cs.RO↗
arXiv2026

From Evaluation to Evolution: Benchmarking Image Editing and Evolving Image Models

Real-world image editing requires coordinating multiple visual transformations with content preservation and reference constraints, yet aggregate scores can readily obscure unmet local requirements. Effective evaluation must therefore identify local failures and translate them into priorities for subsequent model improvement. We introduce WeGenEditBench, comprising 2,740 evaluation instances and 148 task types across single-image, text-intensive, and multi-reference editing. Pre-generated checklists shared across models decompose each request into atomic requirements; item-level deductions and visual evidence are established first and then summarized into scores for edit execution, preservation, visual quality, and task-specific correctness. To support the experience-acquisition component of recursive self-improvement (RSI), we develop WeGenEvoStudio, a unified multi-agent harness for RSI-oriented learning that integrates model-conditioned failure diagnosis, retrieval-grounded data construction, supervised adaptation, and re-evaluation. Among 15 evaluated configurations, GPT-Image-2 leads single-image editing with an Overall score of 4.076, while Seedream-5.0-Pro leads text-intensive editing at 4.291. Evolution experiments across four generation and editing benchmarks demonstrate substantial targeted gains: on their report-selected Top-5 single-image targets, FLUX.2 [dev] improves Overall from 1.083 to 3.517, while Qwen-Image-Edit-2511 gains 13.2%. Together, our benchmark and harness provide a foundation for developing more reliable image models through interpretable evaluation and feedback-driven learning.

cs.CV↗
arXiv2026

Yb/Sr optical clock comparison at 2.2E-18 uncertainty over a 1350-km fiber link

Redefining the International system of Units (SI) second requires independent frequency-ratio measurements between clocks based on different laboratories, with an overall uncertainty below 5E-18, to validate individual clock evaluations through consistency checks. Reaching this uncertainty over long distances has remained challenging. So far, comparisons at this level have been achieved only a few times, between co-located clocks, and their results disagree markedly. Previous high-precision comparisons have therefore been limited to local or kilometre-scale links. Here we compare a ytterbium optical lattice clock in Wuhan with a strontium optical lattice clock in Shanghai over a phase-stabilized fiber link spanning 1350 km, and determine the Yb/Sr frequency ratio to be 1.207 507 039 343 337 716 0(26), with a Bayesian statistical uncertainty of 1.14E-18 and a total uncertainty of 2.18E-18. To our knowledge, this is the first intercity long distance frequency-ratio measurement, whose total uncertainty reaches the level required by the redefinition roadmap. The measured ratio exhibits a significant correlation with Earth tides and its response agree with the model within the statistical uncertainty. At intercity scale, this comparison demonstrates the capability of an optical-clock network for frequency comparison and Earths gravitational-potential sensing, a key step toward a wide-area optical-clock network.

physics.optics↗
arXiv2026

Don't Rewrite the SID Map: Topology Preservation for Generative Retrieval in Dynamic E-Commerce Search

Generative retrieval assigns each item a discrete semantic identifier (SID) and retrieves items by autoregressively generating SID sequences. In dynamic environments such as e-commerce, new items arrive continuously, requiring periodic updates to the SID-to-Item mapping and model to maintain retrieval timeliness. Each update, however, can invalidate search paths already learned by the retriever, and the quantitative relationship between mapping change and retrieval quality has not been systematically characterized. We study this problem with extensive controlled experiments over 67 days of large-scale e-commerce search logs and obtain three core findings. (1) SID semantic drift causes severe retrieval degradation that may be masked by large beam sizes: mapping updates reduce HitRate by 41.78-62.05% at beam size 1 but only 8.33-10.85% at beam size 1000, as redundant candidate generation conceals the true severity of drift. (2) The fraction of changed SIDs is strictly monotonically related to retrieval quality: as the remapped fraction increases from 0% to 72.75%, HitRate decreases monotonically, dropping up to 42.66% at beam size 1. Full remapping remains substantially worse than the preservation policy under warm-start training, indicating that short-budget continued training alone cannot repair topology disruption. (3) The SID topology-preserving principle makes even the simplest warm-start strategy highly effective: fixing old-item SIDs, inserting new-items, and warm-starting from the previous model improves old-item HitRate by 10% and new-item HitRate by 109%, validated across two non-overlapping update cycles. These findings distill into a concise deployment principle: insert new-items, freeze old-items; warm-start when possible, otherwise freeze; never change SIDs the model has already learned. A three-week online A/B test on TmallAPP shows UCTCVR +0.29% and GMV +1.16%.

cs.IR↗
arXiv2026

Taming Subpacketization without Sacrificing Communication: A Packet Type-based Framework for D2D Coded Caching

Finite-length design is essential for making coded caching practical, as the optimal communication gains of existing schemes often require prohibitively large subpacketization. This paper studies rate-optimal device-to-device (D2D) coded caching with reduced subpacketization. We propose a packet type-based (PT) framework that exploits the geometric structure induced by user grouping. Under this structure, subfiles, packets, and multicast groups are classified into types, allowing the originally symmetric Ji-Caire-Molisch (JCM) design~\cite{ji2016fundamental} to be systematically relaxed without sacrificing the optimal D2D communication rate. The key feature of the PT framework is that subpacketization reduction is achieved through two complementary mechanisms: \emph{subfile saving}, by excluding redundant subfile types, and \emph{further-splitting saving}, by assigning type-dependent further-splitting factors to subfiles through transmitter selection. The type-dependent splitting factors are then coordinated across multicast group types to produce a globally consistent file-splitting structure. Based on this framework, we construct several classes of rate-optimal D2D coded caching schemes that strictly improve upon the JCM subpacketization. The proposed schemes achieve either order-wise reductions in the number of users or constant-factor reductions over broad memory regimes, while preserving the optimal rate. These results reveal a structural distinction between D2D and shared-link coded caching: unlike in the shared-link setting, full symmetric subpacketization is not necessary for rate-optimal D2D caching.

cs.IT↗
arXiv2026

When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation

Label-free evaluation methods, such as self-consistency filtering, majority-vote aggregation, and agreement-based reliability scoring, rely on an implicit assumption that consistency can serve as a proxy for correctness when ground-truth labels are unavailable or expensive to obtain. In this paper, we test this assumption directly. Using four financial QA datasets (multiple-choice and free-form), ten consistency dimensions, and two frontier model families, we measure the joint distribution of consistency and accuracy at the level of individual examples rather than in aggregate. We find that 60% of all incorrect answers fall in the high-consistency region (consistency $\geq$ 0.8), and that 77.7% of these show complete answer collapse, i.e. identical wrong answers across all perturbation variants. This high-consistency/incorrect quadrant cannot be detected by any evaluation signal that is built on agreement alone, as the model appears maximally reliable while being wrong. We show that the phenomenon is robust to the choice of consistency threshold t (it persists from t=0.50 to t=1.00), is shared across model families, and is concentrated in specific intersections of perturbation dimension and domain. We map where this quadrant occurs, analyze its stability properties, and discuss the implications for evaluation pipelines that use agreement-based signals in place of human labels.

cs.CL↗
arXiv2026

Beyond Retrieval Failure: Understanding Why LLMs Miss Evidence in Multi-Hop RAG

Multi-hop retrieval-augmented generation (RAG) requires an agent to retrieve evidence for a sequence of dependent sub-questions. When a wrong answer lacks a required passage, the passage may be inaccessible to the retriever, or the agent may have failed to issue a query that would retrieve it. Standard retrieval evaluations conflate these two causes. We study how much apparent retrieval failure is instead caused by search drift: the search no longer tracks an unanswered sub-question. Holding the retriever, corpus, and generation model fixed, we introduce ExecutionRAG, a training-free requirement register that records open sub-questions and their supporting evidence. The register changes search guidance but not evidence access, so each gold passage it recovers certifies that the passage was reachable by the original retriever. On 1,000 MuSiQue questions, the number of wrong answers with unretrieved required evidence falls from 382 to 234 with all-MiniLM-L6-v2 and from 383 to 274 with Qwen3-Embedding-0.6B. These counts correspond to 69% to 44% and 66% to 48% of each system's wrong answers, respectively. Crucially, evidence recovery does not produce a proportional reduction in total error: premature resolutions and abstentions increase as missing-evidence failures decrease. The central finding is therefore a shift in the remaining bottleneck, from guiding search toward required evidence to verifying that the retrieved evidence is sufficient.

cs.CL↗
arXiv2026

ChairNav: Cross-Embodiment Pre-training and Personalization for Long-Horizon Wheelchair Navigation

Developing autonomous electric wheelchairs capable of long-horizon sidewalk navigation is critical for accessible personal mobility, yet existing approaches either focus on indoor map-based navigation or rely on expensive, complex sensor stacks to achieve full autonomy. In this work, we propose to tackle outdoor sidewalk navigation by leveraging visual navigation foundation models and large-scale navigation datasets, which offer strong generalization using lightweight hardware. To this end, we present ChairNav, a novel framework that adapts a general visual navigation model into a personalized electric wheelchair navigation system using a three-stage data pyramid. To address the scarcity of in-domain data, ChairNav first pre-trains on large-scale cross-embodiment datasets transformed into the wheelchair domain, followed by mid-training on a small number of wheelchair demonstrations to eliminate remaining domain gaps. Finally, a post-training stage refines the deployed model from user interventions to personalize its behavior. We deploy ChairNav on a commercial electric wheelchair equipped with an RGB-D camera and an onboard NVIDIA Jetson compute module. Experimental results validate the effectiveness of each stage in our data pyramid: pre-training reduces lateral ADE by 18%, and post-training increases the mean distance between user interventions from 29.2 m to 57.4 m. Further evaluation on kilometer-scale urban sidewalk routes shows that our system operates autonomously for 93.4% of total driving time, demonstrating its potential for everyday accessible mobility. Demo video and more details are at https://vail.cs.ucla.edu/ChairNav/.

cs.RO↗
arXiv2026

MMKG-R1: Constructing Multimodal Knowledge Graphs via Reinforcement Learning

Multimodal knowledge graphs (MMKGs) organize information across different modalities to support multimodal Graph Retrieval-Augmented Generation (RAG), which aims to improve the performance of multimodal large language models (MLLMs). The effectiveness of these multimodal Graph RAG methods largely depends on the quality of the constructed MMKGs. However, most existing methods construct MMKGs without explicit guidance on graph quality, while reinforcement learning (RL)-based methods that rely on final-answer correctness may provide only indirect feedback on MMKG quality. To address these limitations, we propose MMKG-R1, a novel MMKG-quality-driven RL framework for training an MLLM as a multimodal knowledge constructor. To be specific, MMKG-R1 introduces multiple quality-aware rewards to directly evaluate both the quality of information extracted from individual knowledge sources and the quality of the assembled MMKG. These rewards are then used to guide the RL training. Experiments across multiple datasets show that MMKG-R1 constructs MMKGs that improve downstream question-answering accuracy.

cs.CV↗
arXiv2026

Prompt-Segmentation Meets ROI Encryption: Fine-Grained Video Protection Based on Diffusion Isolation

Region-of-interest (ROI) selective encryption in H.265/HEVC aims to protect sensitive content while preserving the utility of non-ROI regions. Existing coding-unit (CU)-level approaches still suffer from substantial encryption redundancy for two reasons. First, existing approaches map object-detection bounding boxes to CUs, producing regular rectangular regions, which poorly match the shape of irregular ROI boundaries, such as facial contours. Second, their coding-tree-unit (CTU)-based diffusion isolation expands the protected footprint in fixed $64\times64$ blocks, causing encryption-induced distortion to spread over a large number of non-ROI pixels. To address the aforementioned shortcomings, we propose a prompt segmentation-guided CU-level ROI encryption algorithm, which mainly consists of two core components. The first component is a prompt segmentation-guided fine-grained ROI encryption pipeline. It maps pixel-level ROI masks to CUs according to an ROI-overlap threshold, refines boundary CUs to $8\times8$ granularity, and applies format-compliant encryption to multiple syntax elements in the mapped ROI CUs. The second is a dual diffusion isolation strategy to address the encryption diffusion problem. For spatial isolation, the encoder refines non-ROI boundary regions into $8\times8$ CUs and uses PCM mode to bypass intra-frame prediction. For temporal isolation, it removes motion-vector candidates whose reference blocks overlap ROI CUs. Experimental results demonstrate that the proposed method generates encryption regions that more closely match irregular ROI boundaries, effectively confines distortion to the ROI.

cs.CR↗
arXiv2026

KV-Skill: Forging Expertise in the Model's Native Language

Task knowledge is commonly stored either as text in the prompt or as an update to model weights. Text is modular but must be interpreted on every use, while weight adaptation makes the resulting capability difficult to load, remove, or share independently. We introduce KV-Skill, a design space of external factorized operators that a frozen language model reads through a lightweight interface. KV-Skill supports two complementary paths. Registration converts an authored text skill into a text-derived operator and trains a shared per-backbone interface. Reward learning develops a compact latent operator directly from task outcomes, with or without an authored skill. Neither path adds positions to the prompt. Across ten benchmarks and four backbones from three model families, converting text to a KV-Skill consistently makes the same procedural knowledge more effective on mathematical reasoning and retrieval tasks. On Qwen3.5-4B LiveMath, registration reaches 77.2 accuracy, compared with 52.0 for the same skill as text (SkillOpt) and 64.5 for SoftSkill. Under matched reward training and parameter budgets, KV-Skill gives the best result in seven of eight matched settings against soft prefixes, prefix tuning, and LoRA. A post-hoc rank analysis further shows that text-derived operators retain nearly all of their benefit with one task-aligned direction per injection layer, while matched random directions fail. Finally, a single shared interface serves eleven diverse skills at once with no loss in average accuracy. These results show that task knowledge can be acquired from text or experience, compressed into an external operator, and deployed separately from the backbone. Code is available at: https://github.com/shawnzhg/KV-Skill

cs.LG↗
arXiv2026

Analytic Liouville Integrable Geodesic Flow Has Zero Entropy

This paper studies the topological entropy of real-analytic Liouville integrable Hamiltonian systems. We prove that the Hamiltonian flow on a compact connected regular energy hypersurface has zero topological entropy whenever a complete family of commuting real-analytic first integrals is generically independent in a connected neighborhood of that hypersurface. The proof uses finite normal expansions along minimal analytic carriers and an ergodic version of Ziglin's leading-term argument to obtain independent invariant normal polynomials, whose Poisson geometry makes all tangent Lyapunov exponents of the carriers vanish. In particular, we prove that a real-analytic Liouville integrable geodesic flow on a closed real-analytic Riemannian manifold has zero topological entropy, giving a positive answer to the conjecture proposed by Paternain in 1992.

math.DS↗