arXiv subjects

Xin Wang

Publications and source records attributed to Xin Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Orbital cancellation and reentrant localization in a quasiperiodic trimer ladder

Uniform orbital detuning can restore spatial extent even as quasiperiodic modulation strengthens. In a trimer ladder with fixed hoppings and orbital modulation signs, increasing detuning and modulation amplitude together drives successive avoided crossings that redistribute orbital weights, suppressing the projected onsite potential in both lower molecular bands while leaving the third strongly modulated. Complete-band participation scaling supports two reentrant high-participation windows, with wave-packet propagation corroborating spatial recovery at representative controls over the studied sizes and times. These diagonal cancellations leave residual interband modulation, which generates a self-energy whose magnitude depends on the molecular level separation. Gaussian quadrature improves convergence of closed-walk determinant estimates used to cross-check energy-resolved attenuation. Comparing the coupled and decoupled ladders under the same finite-size participation criterion shows that restoring the full off-diagonal modulation narrows both sampled windows, with a larger absolute contraction of the second.

cond-mat.dis-nn↗

arXiv2026

RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models

Video-based interactive world models aim to enable real-time, user-controllable environments. Yet most existing systems are trained and rolled out at a fixed temporal rate, limiting both runtime-rate adaptation and physical-horizon coverage -- a limitation we term temporal-rate coupling. Specifically, when runtime generation throughput differs from the learned temporal rate, the generated world evolves unintentionally more slowly or quickly. Moreover, using a single temporal rate imposes an inherent trade-off between temporal granularity and physical horizon: higher rates capture finer motion but cover a shorter duration under a fixed context window. To address both limitations, we introduce Rate-Adaptive Rollout (RAR), a framework that enables interactive world models to roll out at controllable temporal rates. At its core, Stride-RoPE aligns latent frames sampled at different rates on a shared physical-time coordinate, enabling consistent multi-rate rollout, unseen-rate generalization, and online rate switching without discarding history. Meanwhile, RAR uses multi-stride video sampling to enable joint learning of fine-grained dynamics and long-horizon evolution. Across action- and camera-controlled environments, RAR achieves temporally consistent multi-rate rollout and online rate switching while preserving high-rate generation quality and extending physical horizons under matched training budgets. RAR also transfers to pretrained world models while preserving perceptual quality.

cs.CV↗

arXiv2026

AdaStep: Adaptive Step Credit Weighting for Agentic Reinforcement Learning

Long-horizon LLM agents are typically trained with sparse outcome rewards, making trajectory-level objectives too coarse to distinguish the contribution of individual decisions. Step-level credit assignment provides finer-grained supervision, but its estimates can be unreliable because observed returns also depend on subsequent actions, environment transitions, and trajectory length. We propose AdaStep, an Adaptive Step-credit weighting method that controls how strongly each group-derived local advantage modifies the trajectory-level signal. We formulate this weighting as a mean-squared-error estimation problem for the latent step advantage and, under an explicit conditional sampling assumption, derive an optimal per-state shrinkage coefficient. The coefficient admits a signal-to-total-variance interpretation: it preserves local credit when return variation is attributable to the selected action and suppresses it when variation is dominated by downstream randomness. AdaStep requires only lightweight scalar computation, with no critic, additional rollouts, or extra model inference. Experiments with three model backbones on ALFWorld, WebShop, and ScienceWorld show consistent improvements over baselines at low computational cost.

cs.LG↗

arXiv2026

Approximation Algorithms for Dual-Memory Request Scheduling in Disaggregated LLM Inference

Disaggregated LLM inference executes the prefill and decode phases on separate resource pools, reducing interference between the two phases. However, this architecture introduces a coupled scheduling problem because a request that completes prefill retains its prompt KV cache in prefill memory until it is admitted to the decode pool. Prefill and decode admission decisions therefore jointly determine memory occupancy and request completion times. To the best of our knowledge, this problem lacks a formal formulation, and no approximation guarantees are known for request scheduling in disaggregated LLM inference. We formalize this setting as the dual-memory request scheduling (DMRS) problem, whose objective is to minimize the sum of request completion times, and prove that it is strongly NP-hard. We develop two deterministic polynomial-time approximation algorithms. The first applies to instances satisfying $α<1$ and $β<1$ and achieves an approximation ratio of $8+\frac{4}{1-α}+\frac{8}{1-β}$, where $α$ and $β$ are the maximum normalized prefill and peak decode memory demands among all requests, respectively. In particular, it yields a $32$-approximation when $α,β\leq 1/2$. The second algorithm applies to all feasible instances and achieves a uniform $36$-approximation. Finally, evaluations on two H100 GPUs using trace-derived offline workloads show that the proposed algorithms reduce the sum of request completion times relative to multiple vLLM deployment baselines.

cs.DS↗

arXiv2026

Fast continuous ion transport through X-junctions with low motional excitation for QCCD architectures

Trapped ion quantum computing based on the quantum charge coupled device (QCCD) architecture uses functional zoning and ion transport to enable high fidelity computation and all to all qubit connectivity. As trapped ion quantum processing units (QPU) scale, 90° X-junction turns account for an increasing fraction of total transport time. Frequent junction crossings, therefore, make transport speed and final phonon excitation direct determinants of quantum algorithm execution efficiency. However, complex electric fields in X-junctions complicate waveform design, making it difficult to achieve high speed and low final phonon excitation simultaneously. Here, we develop a waveform design method for low-phonon-excitation high-speed X-junction(LHX-T) transport. It integrates low barrier pseudopotential channel design using multiple radio frequency (Multi-RF), multidimensional shortcut to adiabaticity dynamics, and constrained DC voltage optimization over the entire transport sequence. This framework connects waveform design, control, and computational validation, with smooth multichannel waveforms computed rapidly through a Karush-Kuhn-Tucker(KKT) linear system. The resulting waveform transports a single $^{171}Yb^+$ ion through a 90° turn in $5μs$ at an average speed of 14.6 m/s. Over 50 consecutive round trips, the final equivalent coherent phonon number consistently stays below 0.001, with no noticeable accumulation over time. System level evaluations at the QCCD scale show that the median duration of a single execution cycle is reduced by about $30\%$ across various circuit types. The fraction of execution time spent on transport decreases from approximately $70\%$ to $55\%$. This approach offers a means of designing fast ion transport waveforms with low phonon excitation for scalable QCCD based QPUs and provides a technical basis for efficient execution of quantum algorithms.

quant-ph↗

arXiv2026

Remembering What Matters for Long-Horizon Robotic World Models

Autoregressive robot video world models based on discrete visual tokens remain unreliable over long rollout horizons. Retaining the entire generated history does not necessarily improve prediction quality. We investigate a possible memory-role conflict in Full Context decoding: persistent scene information, recent visual-action states, and stale generated history compete for attention despite their different predictive roles. This perspective motivates ReCAP (Receding Context with Anchor Preservation), a training-free context-scheduling method for long-horizon robot video world models. ReCAP keeps the pretrained tokenizer, world model, and visual decoder frozen. It preserves the initial scene anchor and a contiguous window of recent complete visual-action blocks while evicting older generated history. The schedule preserves within-block structure and temporal order, requires no additional modules or parameter updates, and bounds active-context growth independently of rollout length. We evaluate ReCAP on RT-1, BridgeV2, LIBERO-90, and CALVIN using frozen autoregressive world models. Across evaluated environments and horizons, ReCAP improves visual fidelity, action-conditioned consistency, and robot structure preservation relative to Full Context. On BridgeV2, it reduces LPIPS by 63.3%. In the system benchmark at H=64, it reduces the logical KV footprint by 73.3% and p50 latency by 6.51%. Attention analyses show increased attention to retained recent states, while history interventions associate larger amounts of stale context with higher prediction error. These observations are consistent with the memory-role-conflict hypothesis, although they do not isolate it from context-length and position-handling effects. Our results support inference-time context scheduling as a practical approach to improving long-horizon robot video prediction and computational efficiency without retraining.

cs.RO↗

arXiv2026

TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models

Reinforcement learning (RL) for post-training large language models (LLMs) incurs substantial computation and memory overhead during rollout generation, which motivates low-precision rollout for efficient RL training. However, existing FP4 RL methods suffer from a key limitation: they primarily optimize quantization accuracy on the training and rollout paths independently rather than directly reducing the discrepancy between the two quantized execution paths. In this work, we propose TRACE (Train-Rollout Quantization Alignment via Compact GuidancE), an FP4 quantization framework for RL training of Mixture-of-Experts (MoE) language models that addresses the limitation of existing FP4 RL methods. TRACE incorporates rollout-guided quantization-aware training that uses rollout-side quantization outcomes to guide training-side FP4 rounding decisions, directly reducing train-rollout discrepancy. Moreover, TRACE adopts an efficient quantization-information caching scheme that selectively retains mantissa and scale information from deeper layers to reduce the storage and communication overhead introduced by rollout guidance. We evaluate TRACE on four large-scale MoE language models across reasoning, coding, and long-horizon RL tasks. Our results demonstrate that TRACE enables joint FP4 weight/activation and FP4 KV-cache rollout with RL performance comparable to BF16 rollout, while achieving up to 5.4xrollout speedup and strong final FP4 performance compared with post-hoc FP4 quantization of BF16-trained policies.

cs.LG↗

arXiv2026

What Personal Information Improves LLM-Based Next-Location Prediction?

Large language models (LLMs) are increasingly used for individual next-location prediction, with personal information easily added to prompts alongside mobility history. Yet the incremental predictive value of such information remains unclear. Using linked sociodemographic records and mobility traces from 5,000 Shenzhen residents, this study separates model responsiveness from predictive value. GPT-5 is the primary model, with GPT-5.5 and Claude Opus 4.6 used for replication. In 1,000 paired prediction instances, models rank 100 candidate destinations with and without age, gender, occupation and income while all other inputs are held fixed. Behavioural history raises top-1 accuracy from 5.6% to 18.5% as prior history increases from zero to six days. By contrast, sociodemographic attributes produce no detectable overall gain, although replacing correct attributes with those of another person reduces accuracy by 5.4 percentage points. Candidate construction also matters, removing distance raises accuracy by 7.7 points under proximity sampling but lowers it by 22.3 points under popularity sampling, with the reversal reproduced across all three LLMs. These findings identify behavioural history as the clearest source of incremental value and show that personal information should be evaluated under matched, explicitly specified conditions before its privacy and governance costs are justified.

cs.CY↗

arXiv2026

Beyond Report Imitation: Clinically Aware Multi-Image Ultrasound Report Generation from Visible Evidence

Generating ultrasound reports from multiple images requires aggregating clinical evidence across views, yet archived key frames capture only part of the dynamic examination. Raw-report imitation is therefore misaligned with visual supervision: content that is clinically valid for the full examination may be unverifiable from the images available to a model. This gap creates a clinical behavior alignment problem. A model must preserve visible findings, avoid diagnostic reversals and unsupported completion, and not collapse into conservative templates. We propose CAMEO, a Clinically Aware Multi-image Evidence-grounded Orchestration framework for ultrasound report generation. Stage I learns ultrasound visual-language primitives; Stage II performs Cross-View Evidence Grounding by distilling trusted visible report points into multi-image QA and report-style supervision; and Stage III performs Clinically Aware Preference Alignment using clinical-error-oriented preference pairs. From USReport, we construct USReport-Distilled with 17,670 evidence-grounded paired-image training instances and USReport-Pref with 21,869 preference pairs; we additionally use 25,631 PubMedVision-US ultrasound instruction samples for domain adaptation and multi-image instruction tuning. On the primary USReport-Distilled benchmark, CAMEO improves over EchoVLM from 0.25 to 0.40 BLEU-1, 0.28 to 0.45 ROUGE-1, and 0.27 to 0.43 METEOR, while raising ClinicalScore from 55.02 to 74.20. These results underscore the value of evidence-grounded supervision, clinically aware alignment, and clinically structured evaluation for reliable ultrasound report generation.

cs.CV↗

arXiv2026

Nearly Optimal T-Count for Symmetry-Based Quantum State Purification

Protecting quantum information from noise is essential for reliable quantum computation and quantum memories. Quantum state purification addresses this challenge by processing multiple noisy copies of an unknown quantum state to produce an output state of higher fidelity. Here we establish nearly tight upper and lower bounds on the non-Clifford resources required to implement the symmetric-projection purification protocol. For $n$ copies of a $d$-dimensional system, we construct a Clifford+T circuit that block-encodes the symmetric projector with operator-norm error at most $ε$ using $\widetilde{O}(n)$ T gates, where the tilde-$O$ notation suppresses polylogarithmic factors in $n$, $d$, and $1/ε$. Conversely, using stabilizer nullity, we prove that any Clifford+T implementation of such a block encoding with error $ε<\frac{1}{2}\sqrt{1-1/n}$ requires at least $2n-2$ T gates, even allowing arbitrary positive normalization and clean stabilizer ancillas. Taken together, these results establish the nearly optimal scaling of non-Clifford cost with the number of copies and uncover the fundamental resource requirements for coherent symmetry projection in quantum state purification.

quant-ph↗

arXiv2026

SWE-NFI: Studying and Benchmarking Coding Agents for Non-Functional Improvements

Although coding agents have achieved impressive performance on correctness-oriented benchmarks, their ability to make behavior-preserving non-functional improvements (NFIs) remains underexplored. In real-world software development, developers continuously improve software quality without changing observable behavior, yet existing benchmarks primarily evaluate functional correctness and provide limited support for assessing these non-functional improvements. In this paper, we present SWE-NFI, a benchmark for evaluating coding agents on NFIs beyond functional correctness. SWE-NFI contains 188 tasks constructed from real merged pull requests in open-source Python projects. We operationalize five developer-oriented NFI aspects into 92 executable rules and develop an evaluation suite that first applies task-specific functional preservation checks to assess whether key properties of the original code are preserved and then performs rule-based NFI evaluation. For outputs that pass these checks, we measure NFI improvement as the change in the normalized rule score for the task's target aspect relative to the original code. We evaluate eight commercial and open-source agent configurations and run each configuration five times on every task. At most 70.0% of valid agent outputs satisfy all task-specific functional preservation checks. Among outputs that pass these checks, NFI performance varies substantially across aspects. In particular, all evaluated configurations achieve mean Logic Patterns improvements between 0.0 and 1.3, below the Human Reference mean of 1.5. Our benchmark and findings provide a reproducible foundation for evaluating coding agents on non-functional improvements and identifying the NFI aspects where further advances are needed.

cs.SE↗

arXiv2026

SkillPoison: Progressive Skill Poisoning via Successful Experiences

Self-improving LLM agents increasingly distill successful experiences into persistent, reusable skills. Existing skill attack methods corrupt this learning pipeline by injecting malicious triggers, behaviors, or false facts into individual experiences or extracted skills. However, such attacks are easily detected, and the injected malicious behaviors often fail to accumulate as persistent skills. In this paper, we show that skill poisoning can arise even from verified successful experiences, without making any individual trajectory malicious. Based on this insight, we propose SkillPoison, a novel framework that progressively poisons skill via successful experiences. SkillPoison first constructs a set of successful experiences that reinforce a target behavior, and then removes the contextual conditions that constrain when the behavior applies. Rather than injecting malicious content, SkillPoison shapes how the skill extractor generalizes, allowing useful behavior to support task success while inducing harmful behavior when they are misapplied. Extensive experiments on three benchmarks show that SkillPoison achieves 95.71% attack success rates, while all injected experiences remain task-correct and pass verification and lexical inspection. Our code, data and implementation details are available for the community at https://github.com/DEEP-JLU/SkillPoison.

cs.CR↗

arXiv2026

Foresight-over-Graph: Reasoning Beyond Local Horizons for Knowledge Base Question Answering

Large language models (LLMs) have demonstrated strong capabilities in question answering, yet they still frequently suffer from hallucinations on knowledge-intensive tasks. Knowledge graphs (KGs) provide LLMs with structured, interpretable, and updatable factual grounding, making them a promising external knowledge source for reliable reasoning. However, existing LLM-guided graph reasoning methods typically rely on hop-wise greedy or beam-style pruning during evidence retrieval. Such local decision processes are inherently myopic: evidence that appears weak near the source may become crucial only after deeper graph context is explored, causing answer-critical branches to be discarded prematurely and making the reasoning chain difficult to recover. To address this limitation, we propose Foresight-over-Graph (FoG), a foresight-aware evidence retrieval framework for knowledge base question answering (KBQA). FoG iteratively constructs a question-relevant evidence subgraph and uses far-to-near feedback to guide path exploration, and maintains a compact memory subgraph to support continued exploration. Extensive experiments on widely used KBQA benchmarks demonstrate that FoG achieves state-of-the-art performance, with a particularly large improvement of 16.58% in Hit on CWQ, while also reducing LLM calls and token usage. Our code is available at https://github.com/yhong7/FoG .

cs.CL↗

arXiv2026

Adaptive Power Sampling for LLM Reasoning

Sequence-level power sampling has recently emerged as a training-free approach to reasoning by sampling from a sharpened output distribution of a base large language model (LLM). Nevertheless, existing methods typically sharpen the base model distribution uniformly across queries, overlooking variations in query difficulty and in how well the base model already handles each query. The goal of this work is to equip power sampling with query adaptivity. Theoretically, we show that the benefits of further sharpening are determined by the self-reward gap between correct and incorrect responses. Based on this insight, we propose \emph{Adaptive Power Sampling} (APS), which adjusts the sharpening exponent on a per-query basis at test time using the relationship between answer agreement and the model's self-reward. Experiments across diverse reasoning tasks, including MATH500, HumanEval, and GPQA, show that APS consistently outperforms power sampling with a fixed sharpening exponent, without additional training.

cs.AI↗

arXiv2026

Quantum incapacity from observable shadows

Quantum communication is ruled out when an eavesdropper can reconstruct the receiver's state. Yet states are operationally specified by measurement statistics, raising a sharper question: can access to every expectation value destroy communication? We show that it can. We call unbiased, variance-nonincreasing reconstruction under arbitrary reference extensions \emph{complete observable shadowing} and prove that it forces unassisted private and quantum capacities to vanish. An explicit qutrit channel realizes this mechanism while being neither antidegradable nor PPT, resolving the long-standing questions of whether quantum capacity can vanish outside those two classes and whether antidegradability is the only route to zero private capacity. Notably, it superactivates every finite-dimensional channel with zero quantum but positive private capacity. Statistical access can therefore preclude standalone communication without erasing latent coherent utility.

quant-ph↗

arXiv2026

Collective Quantum State Preparation Is Nonadditive

Quantum state preparation is a basic primitive of quantum computation and control, but its resource cost is usually analyzed for one target state or under an independent-copy construction. Here we study the many-copy problem under weighted Pauli control, where a Pauli-product rotation $e^{iϕP}$ is charged by its rotation angle $|ϕ|$, independently of the weight of $P$. We derive the exact single-qubit preparation cost and show that it is strictly nonadditive: a four-pulse collective protocol prepares two $T$ states at lower cost than two optimal one-copy protocols. We further identify finite regions of the Bloch sphere with collective saving, including trajectories that remain product throughout, showing that transient entanglement is not the only mechanism behind the advantage. Complementarily, a dimension-independent Rényi-entropy speed limit gives an extensive lower bound for every coherent target and an exponential fidelity penalty below a finite preparation rate. These results establish a many-copy geometry of state synthesis in which uncorrelated outputs can nevertheless benefit from collective control.

quant-ph↗

arXiv2026

Closed-Form Arbitrary-Gap Rényi Nonadditivity Above Any Positive Order via Direct Products of Free Groups and High-Girth Graphs

We give a closed-form finite-dimensional construction of quantum channels with real algebraic entries exhibiting arbitrarily large violations of minimum-output Rényi entropy additivity uniformly above any prescribed positive Rényi order, using a different finite realization based on direct products of free groups and fixed-field high-girth graphs. More precisely, for every $p_0>0$ and $g>0$, we construct one channel and one Bell input for which the additivity gap is greater than $g$ simultaneously for all $p\in[p_0,\infty]$. With $r=\lceil2g\rceil$, the construction uses $15r$ output qubits and $O(r2^{30r}+r^2(1+p_0^{-1}))$ input qubits. Compared with the stated parameters of Shou and Gorshkov \cite{ShouGorshkov2026}, our construction uses fewer output qubits for every target gap and has a smaller asymptotic input-qubit bound as the gap grows with the positive cutoff fixed. At the von Neumann order, a separate specialization gives an explicit channel with gap greater than $1/1024$ using $37608898$ input qubits and $8$ output qubits.

quant-ph↗

arXiv2026

EVISKILL: Grounding Skill Evolution in Replayable Evidence

Continual skill evolution enables LLM agents to accumulate and refine reusable procedural knowledge from interaction experience without updating model parameters. Its effectiveness depends on determining not only what to change, but also why a change is justified and when it should become persistent guidance. However, existing experience-driven methods can lose the behavioral evidence and task contexts supporting edits. Moreover, a global validation outcome provides an incomplete judgment of its constituent changes: locally supported corrections may be discarded with a rejected revision, while evidence may require further experience to inform useful updates. To this end, we introduce EVISKILL, an evidence-driven framework that organizes execution observations into Replayable Evidence Cards and synthesizes edits with explicit links to their supporting contexts. Targeted replay verifies these edits through re-execution and provides feedback for correction. Across epochs, EVISKILL preserves evidence and provisionally retains supported edits for further refinement, while global validation governs their incorporation into the final skill. Experiments on three interactive benchmarks across six LLM backbones demonstrate the effectiveness of this approach.

cs.AI↗