arXiv subjects

Xin Xu

Publications and source records attributed to Xin Xu.

At least 19 recordsLinked to original sources

arXiv2026

Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic Systems

Improving an agent through successive revisions requires balancing exploration and evaluation under a limited budget. A narrow search can miss useful sequences of edits, while fully evaluating every intermediate version often exhausts the budget before those sequences yield a verified improvement. Inspired by the draft--verify principle of speculative decoding, we propose Speculative Recursive Self-Improvement (SpecRSI), a two-level framework coupling inner-loop agent search with outer-loop improver learning. SpecRSI drafts successors before their predecessors are fully evaluated, while adopting only fully evaluated, admissible versions. In the inner loop, Verification-Aware Tree Search (VATS) allocates the budget between further edits and outstanding checks using the predicted verified return of a continuation. In the outer loop, Group Verification Optimization (GVO) updates the improver from completed searches while keeping the real evaluator fixed. We evaluate SpecRSI on scientific optimization (including active queue management (AQM) for network infrastructure), coding, and general agent-system benchmarks. Under matched evaluation-call budgets, SpecRSI improves scientific terminal gain over Janus by 60.0%--96.3% and area under the budget curve by 32.1%--91.6%, while achieving higher means across all 50 scientific backbone--task pairs and outperforming MetaRSI-v1 by 1.30--2.70 percentage points (pp) in system score across twenty backbones. On instance-level comparisons, it further improves PyBaMM terminal gain by 142.6% and SWE-Pro success by 6.0 pp at 7.0% and 9.2% lower token cost, respectively. These results demonstrate the efficacy of the proposed approach.

cs.LG↗

arXiv2026

Two-point residual step size gradient methods for linear systems

Kaczmarz-type methods often converge slowly when the coefficient matrix is ill-conditioned. In particular, when the smallest singular value is small, every row of the coefficient matrix is nearly orthogonal to the corresponding right singular direction, so that the iteration stalls along that direction. To break this slow-convergence regime, we borrow the delayed-step-size idea of the Barzilai-Borwein (BB) method and, within the framework of the residual-based surrogate-hyperplane Kaczmarz method, propose two new step sizes for gradient-descent-type methods solving linear systems: a delayed residual step size $α_k^{\mathrm{DR}}$ and a residual-difference step size $α_k^{\mathrm{RD}}$, which correspond, respectively, to the BB1 and BB2 step sizes in convex quadratic optimization. We prove that $α_k^{\mathrm{RD}}\leα_k^{\mathrm{DR}}$, and we analyze the convergence of the resulting method in a canonical model with two singular values: although the step sizes are non-monotone, the residual norm is shown to decrease after finitely many steps. Numerical experiments on some ill-conditioned linear systems illustrate the effectiveness of the proposed methods.

math.NA↗

arXiv2026

Thought-deciphering via hierarchical EEG-to-text alignment

Communication is important for our well-being, but it can be impaired by disabilities, such as locked-in syndrome (LIS), cerebral palsy (CP), and dysarthria. Non-invasive brain-to-text brain-computer interfaces (BCIs) offer a promising solution by providing an alternative communication pathway for people with communication disabilities. However, most recent breakthroughs have relied on large, cumbersome, and costly neuroimaging modalities, such as functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). Numerous studies have attempted decoding using the lighter and cheaper electroencephalography (EEG), but the results remain inferior compared to the other modalities. More importantly, we do not understand how these modalities are related to text at the letter, word, sentence, and higher semantic levels. Many prior studies have focused on decoding below the sentence level because it is considerably easier to design an experiment around phonemes, letters, and words. However, during communication, sentences play a much more important role than words and letters. This is because sentences carry information at such a high level that they can be understood even if some words and letters are missing. However, within sentence-level decoding, sentence classification is not as well investigated as sentence generation, making it difficult to decide whether a sentence should be retrieved or generated in the context of EEG-to-text decoding. This paper strives to investigate sentence-level EEG-to-text classification via a novel hierarchical alignment framework, where the main goal is to retrieve a read sentence using its context and EEG as inputs.

cs.LG↗

arXiv2026

Learning from Fleet Experience: Collective Experience Learning Across 96 Robots

Robot deployment generates experience that can support policy improvement, but continued reliance on expert corrective actions makes learning difficult to scale across growing fleets. We propose learning from fleet experience: using deployed robots' own executions as a source of supervision for further policy improvement. We call this paradigm Collective Experience Learning (CEL), in which robots pool execution experience to update a shared policy that returns to the fleet. Following expert-supervised initialization, we realize CEL through failure-guided experience retrieval fine-tuning (FERT). FERT uses failure contexts to retrieve relevant segments from successful trajectories and prioritizes their recorded actions for supervised updates, without collecting additional expert demonstrations or corrective actions. Operators provide binary episode-outcome labels and environment resets. A fault-tolerant asynchronous training system (FATS) coordinates robot execution, cloud inference and policy optimization despite local interruptions. Across five real-world manipulation tasks, the resulting policies achieve success rates of 93.3-99.2%, evaluated over 120 trials per task on robots excluded from training data collection. With collection time per robot held fixed, larger fleets provide more experience and generally improve performance across the evaluated sizes. On multi-stage phone packaging, learning with 96 robots raises success from 79.2% to 98.3%. These results support CEL as a scalable approach to improving robot policy reliability through experience generated during deployment.

cs.RO↗

arXiv2026

FARSim: a fast RF-chain-aware trigger-screening surrogate for radio detection of ultra-high-energy cosmic rays

Radio arrays provide a scalable route to detecting extensive air showers from ultra-high-energy cosmic rays, but trigger studies for candidate layouts are expensive when every energy, arrival direction, core position and trigger configuration is evaluated with full radio simulations. We present FARSim, a fast surrogate framework that reuses a reduced library of ZHAireS reference footprints to reconstruct ground-plane radio emission and to estimate trigger-relevant observables. The method combines vector geomagnetic and charge-excess field decomposition, geomagnetic-angle and energy scaling, geometrical projection, contour-based core sampling and event-rate integration. We evaluate an independent validation set of 1999 ZHAireS showers; 941 events satisfy the predefined radial-support and station-count requirements and yield a median event-level relative RMSE of 9.4% and a median Pearson coefficient $r_{\mathrm{P}}=0.965$ for the reconstructed geomagnetic lateral profiles. We further extend the peak-field surrogate to time-domain electric-field synthesis by combining the predicted three-component peak-field vector with geometry-dependent normalized pulse templates. Propagating these traces through an RF-chain response enables voltage-domain threshold and L1-trigger diagnostics. When evaluated at the true peak amplitude, the time-domain extension reaches a median vector-waveform $R^2$ of 0.986 over 2112 ZHAireS traces held out from pulse-shape model fitting. FARSim is therefore intended as a rapid, physics-informed screening layer for array-layout and trigger studies; absolute exposure predictions and detector commissioning remain the role of full end-to-end simulations.

astro-ph.IM↗

arXiv2026

Corrected detector-level reconstruction of inclined radio air showers from trigger-level ADC waveforms

Autonomous radio arrays record air-shower radio pulses as trigger-selected analogue-to-digital converter (ADC) voltage traces. We reconstruct the primary arrival direction and energy while accounting for noise, direction-dependent detector responses and atmospheric refraction in inclined showers. Waveform cross-correlation corrects relative trigger times; plane- and spherical-wavefront fits estimate the initial direction and effective emission point. A modified voltage-domain angular-distribution fit uses source-end tangents of refracted rays to refine the direction and distinguish the particle core from the radio-footprint centre, reducing the median opening-angle error by approximately 50% relative to straight source-to-station viewing angles. Antenna and radio-frequency-chain inversion recovers the electric field, and early-late corrections give its lateral peak-amplitude distribution. An event-to-reference peak-amplitude correction is calibrated on 2,500 simulated electric-field showers to account for 50-200 MHz coherence loss; a semi-analytic Fresnel parameterisation examines its finite-source interpretation. Proton and iron reference lateral-distribution fits give two energy estimates, averaged with fixed equal weights. Validation uses 6,000 independent showers. In the field-level test with simulation-input geometry, either correction reduces the relative energy root-mean-square error (RMSE) from 11.90% to 8.31%. Across three waveform treatments, 3,073 common detector-level events yield mean relative energy residuals of -2.30% to -1.86%, standard deviations of 8.46%-8.96%, and median opening-angle errors of 0.0315-0.0349 degrees. The method is intended for sparse radio arrays detecting highly inclined showers, such as the Giant Radio Array for Neutrino Detection (GRAND).

astro-ph.IM↗

arXiv2026

MS-Bench: A Condition-Stratified Multi-Speaker ASR Benchmark

Multi-speaker automatic speech recognition (MSASR) has advanced rapidly with end-to-end speech large language models. However, existing benchmarks cover limited ranges of specific multi-speaker conditions and scenarios, making it difficult to reveal system limitations under diverse conditions and scenarios. We introduce MS-Bench\footnote{\href{https://aslp-lab.github.io/MS-Bench/}{Demo page: https://aslp-lab.github.io/MS-Bench/}}, a condition-stratified benchmark for fine-grained MSASR evaluation. MS-Bench integrates both public and internal benchmarks and characterizes each recording along five dimensions: speaker number, overlap ratio, speaker similarity, speaker turn interval, and acoustic quality. We stratify these dimensions into condition ranges and construct MS-Bench with broad coverage of conditions, scenarios, languages, and recording devices. Evaluating representative commercial, open-source, and cascaded systems, we find that overlap affects both lexical recognition and speaker attribution, while other factors exhibit more task-specific effects on recognition, attribution, or temporal prediction. These results demonstrate that aggregate metrics alone obscure important MSASR failure modes and highlight the value of condition-stratified evaluation. MS-Bench aims to reveal condition-specific limitations of current MSASR systems and guide future improvements.

eess.AS↗

arXiv2026

Few-shot Classification as Multi-instance Verification: Effective Backbone-agnostic Transfer across Domains

We investigate cross-domain few-shot learning under the constraint that fine-tuning of backbones (i.e., feature extractors) is impossible or infeasible---a scenario that is increasingly common in practical use cases---and show how representing few-shot classification as a series of multiple instance verification (MIV) tasks enables competitive classification performance using the static embeddings produced by frozen, "black-box" backbones. More specifically, based on this novel representation, we introduce a new approach to few-shot domain adaptation named the "MIV-head": a classification head that is backbone-agnostic and computationally efficient. We show that the core components of the MIV-head, when trained on few-shot data from a target domain, collectively yield strong performance on test data from that domain. Importantly, they do so without fine-tuning the backbone, requiring only the "meta-testing" phase. In experiments under various settings on an extension of the Meta-Dataset benchmark for cross-domain few-shot image classification, using representative off-the-shelf convolutional neural network and vision transformer backbones pretrained on ImageNet1K, we demonstrate that the MIV-head achieves highly competitive accuracy when compared to state-of-the-art "adapter" (or "partial fine-tuning") methods applied to the same backbones, while incurring substantially lower adaptation cost. Moreover, compared to our approach, well-known "classification head" approaches achieve a substantially lower level of accuracy. We share our code at: https://github.com/xxweka/MIV-head.

cs.CV↗

arXiv2026

A Topology-Guided GeoAI Agentic Framework for Urban Flood Screening

Open-world geographic artificial intelligence (GeoAI) agents can transform public geospatial data into accessible decision support for real-world challenges, but must balance autonomous reasoning with reliable spatial analysis and human oversight. We present a topology-guided GeoAI agentic framework for urban flood screening that takes a place name, drawn region, or uploaded boundary and returns ranked candidate basins with an interactive map and report. A deterministic controller retrieves public elevation, rainfall, soil, land-cover, exposure, and infrastructure layers, extracts terrain depressions with discrete Morse theory, and ranks basins with a relative flood-priority score. The language model acts only at recorded decision points: selecting a persistence threshold from a precomputed candidate table, drafting recommendations from computed evidence, and reviewing each recommendation against the same evidence. Three human-in-the-loop checkpoints let users revise the threshold, factor weights, or recommendation. In a College Park, Maryland case study, the selected threshold reduced 278 raw terrain minima to 70 basins, with 77% of these basins' lowest points also identified as sinks by a separately computed eight-direction (D8) flow-routing model. Across four further study areas, agreement ranged from 61% to 83%. By combining open data, explicit spatial reasoning, traceable evidence, and human oversight, the framework enables transparent first-pass flood screening while keeping consequential recommendations inspectable and user-controlled.

cs.AI↗

arXiv2026

Harnessing Language Models as Optimization Modeling Agents

Large language models (LLMs) can translate natural-language optimization problems into solver programs, but reliably producing correct formulations and implementations remains challenging. Existing approaches improve optimization modeling through agentic workflows or task-specific post-training. This raises a practical question: can a lightweight harness using a single model role achieve competitive performance without additional training? We introduce OPTICYCLE and OPTISTATE, two lightweight inference harnesses for optimization modeling, and evaluate them on nine benchmarks comprising 2,634 problems using 11 Qwen variants spanning 0.6B-27B parameters. Our results show that execution-guided revision consistently improves macro accuracy over direct generation for all evaluated models, with gains of up to 49.46 percentage points. Despite their simplicity, both harnesses outperform six adapted agentic workflows in same-model comparisons and four evaluated post-trained optimization models at comparable model sizes, without additional training. Explicit formulation state provides further but model-dependent improvements, suggesting that additional structure is not universally beneficial.

cs.AI↗

arXiv2026

Incomparable Minima in Agent Tool Authorization

An agent that needs a tool needs authority to call it, and every proposal for governing that authority has to answer one question: of two grants that both let the agent finish a task, which one is smaller? Over-privilege benchmarks score agents against this answer, and permission solvers compute it. We check the question against the authorisation tables that Microsoft and Google publish, covering 22,305 Microsoft Graph endpoint bindings and 23,852 Google API methods, on tasks generated from each table's own resource structure. For 29.8-33.3% of Graph tasks and 3.2% of Google resource tasks the question has no answer, because the minimal sufficient grants form an antichain, a set of grants none of which authorises a subset of what another does. In its two main authorisation modes, Microsoft's shipped privilegeLevel rates 67.4-74.4% of these incomparable pairs equal. Google ships no such scale, so on Google we apply a reconstruction of the FORTIS privilege ladder, which rates 90.7-98.7% equal. Applying the same reconstruction to Graph attributes much of that difference to the scale rather than the corpus: its form based on the HTTP verb alone rates 92.0% of Google pairs and 89.0-92.0% of Graph pairs equal. Four risk measures (operations, writes, services reached, administrative surface) order 28.7% of incomparable Google pairs and 15.7-23.4% of Graph pairs in opposite directions, so for those pairs the ranking has no measure-independent answer. Offered pools of minimal and larger grants, 6 frontier models pick a minimal grant on 99.0% of 1,708 decisions, a grant smallest under all four measures on 66.7%, and on 22.1% a grant that another option matches or beats on every measure. We propose that benchmarks report which minimal grant an agent chose and that deployments show the alternatives to whoever approves the grant.

cs.CR↗

arXiv2026

Foundation-Style Pretraining for Scalable and Transferable Distributed PV Forecasting

The rapid growth of distributed photovoltaic (PV) generation is increasing the demand for accurate forecasting across large numbers of geographically dispersed stations. Existing approaches are commonly developed for individual sites or specific forecasting horizons, requiring repeated training and maintenance as new stations are integrated and limiting direct reuse at previously unseen sites. To address this scalability gap, this paper presents a domain-specific foundation-style pretraining framework that learns a shared forecasting model from heterogeneous multi-site and multi-horizon PV data. NWP-assisted forecasting is formulated as a sequence-completion problem that unifies historical PV power, meteorological observations, future numerical weather prediction (NWP) forecasts, and time encodings. Three pretraining objectives are designed and systematically evaluated, including forecast-aligned reconstruction, generic reconstruction, and mixed reconstruction. The mixed objective combines the operational structure of PV forecasting with random sequence reconstruction and provides the most balanced performance across forecasting accuracy, degraded NWP conditions, and downstream adaptation. Experiments on 479 distributed PV stations use 340 stations for model development and 139 spatially held-out stations for zero-shot evaluation. Across horizons from 6 hours to 7 days, the multi-site pretrained MR model outperforms representative station-specific and shared baselines, as well as generalist time-series models, including a recent large-scale pretrained foundation model. The pretrained model can also be directly deployed at new PV stations and subsequently adapted to region-specific operational objectives when local data become available, providing a scalable forecasting paradigm for large distributed PV fleets.

cs.LG↗

arXiv2026

Thinking-Free Policy Initialization Makes Distilled Reasoning Models More Effective and Efficient Reasoners

Reinforcement Learning with Verifiable Reward (RLVR) effectively solves complex tasks but demands extremely long context lengths during training, leading to substantial computational costs. While multi-stage training can partially mitigate this, starting with overly short contexts often causes irreversible performance degradation, ultimately failing to reduce overall training compute significantly. In this paper, we introduce Thinking-Free Policy Initialization (TFPI), a simple yet effective adaptation to RLVR that bridges long Chain-of-Thought (CoT) distillation and standard RLVR. TFPI employs a simple ThinkingFree operation, explicitly discarding the thinking content via a direct append, to reduce token usage during inference. Training with ThinkingFree-adapted inputs improves performance and lowers token consumption, even in the original slow-thinking mode. Extensive experiments across various benchmarks have shown that TFPI accelerates RL convergence, achieves a higher performance ceiling, and yields more token-efficient reasoning models without specialized rewards or complex training designs. With TFPI only, we train a 4B model to reach 89.0% accuracy on AIME24 and 65.5% on LiveCodeBench using less than 4K H20 hours.

cs.LG↗

arXiv2026

Composition-RL: Compose Your Verifiable Prompts for Reinforcement Learning of Large Language Models

Large-scale verifiable prompts underpin the success of Reinforcement Learning with Verifiable Rewards (RLVR), but they contain many uninformative examples and are costly to expand further. Recent studies focus on better exploiting limited training data by prioritizing hard prompts whose rollout pass rate is 0. However, easy prompts with a pass rate of 1 also become increasingly prevalent as training progresses, thereby reducing the effective data size. To mitigate this, we propose Composition-RL, an offline data-augmentation approach that composes existing prompts before RL to address the pass-rate-1 bottleneck without modifying the RL optimizer. The composed dataset is generated and verified once from the fixed source pool and then used for RL training. Extensive experiments across model sizes from 4B to 30B show that Composition-RL consistently improves reasoning capability over RL trained on the original dataset. Performance can be further boosted with a curriculum variant of Composition-RL that gradually increases compositional depth over training. Additionally, Composition-RL enables more effective cross-domain RL by composing prompts drawn from different domains. All our code, data, and models are publicly available to facilitate future research.

cs.CL↗

arXiv2026

Constructed regional photovoltaic generation fields improve ultra-short-term distributed solar forecasting

Ultra-short-term forecasting of densely distributed photovoltaic (PV) systems remains challenging because external meteorological and satellite products may not adequately resolve rapid local variations, while heterogeneous PV systems exhibit different generation responses under similar atmospheric conditions. Meanwhile, dense PV fleets provide spatially distributed measurements that reveal regional generation dynamics. Here, we reformulate fleet-level PV forecasting from modeling discrete station time series to predicting explicit, spatially structured regional PV generation fields. We develop SatPV-Field, a regional-to-local forecasting framework that transforms distributed PV measurements into image-like generation fields and models their spatiotemporal evolution. By combining historical PV fields with satellite-derived future shortwave radiation guidance and meteorological forecasts, the framework predicts future regional generation states, which subsequently provide location-specific guidance for individual-station forecasting. Experiments involving 496 distributed PV systems, including 446 source stations and 50 held-out target stations, demonstrate that these regional fields can be effectively forecast and used to improve station-level predictions. The proposed approach reduces station-level MAE by approximately 10.0% relative to the best-performing single-station temporal baseline and by 2.5% relative to a graph-based regional forecasting approach using the same source fleet. These findings establish an explicit spatial-field modeling paradigm that connects regional generation dynamics with individual-site forecasting.

cs.AI↗

arXiv2026

AlphaPilot: Multi-Turn Tool-Using LLM Agents for Adaptive Alpha Mining with Reinforcement Learning

Formulaic alpha mining seeks interpretable expressions that rank assets by future returns. The task is challenging because it demands simultaneous processing of unstructured market descriptions, historical data, and quantitative evaluation of candidate factors, yet most existing methods exploit only a subset of these signals. We introduce AlphaPilot, a reinforcement-learning framework that trains multi-turn tool-using LLM agents for adaptive alpha mining. The agent conditions on market descriptions, proposes factor expressions, queries real-data evaluator feedback, and updates a retained factor pool across turns. This design jointly leverages market text, LLM financial knowledge, real data, and evaluation feedback. At inference time, AlphaPilot performs day-by-day search and pool updates to adapt to shifting market regimes. Across three independent runs in out-of-sample rolling backtests on CSI300 and CSI500, AlphaPilot achieves mean total returns of 42.93% and 54.19%, with mean Sharpe ratios of 2.46 and 2.86, respectively. It attains the highest mean total return, Sharpe, and RankIC among the evaluated methods on both universes.

cs.CL↗

arXiv2026

USER: A Unified and Extensible System for Real-World Online Policy Learning in Embodied AI

Online policy learning directly in the physical world is a promising yet challenging direction for embodied intelligence. Unlike simulation, real-world systems cannot be arbitrarily accelerated, cheaply reset, or massively replicated, suggesting that real-world policy learning is not merely an algorithmic problem, but inherently a systems problem. We present USER, a \underline{U}nified and extensible \underline{S}yst\underline{E}m for real-world online policy lea\underline{R}ning. On the systems side, USER introduces a hardware abstraction layer for unified robot management and an adaptive communication plane that enables efficient cloud-edge training. On the learning side, USER adopts a fully asynchronous training framework, designs a persistent and cache-aware replay buffer, and provides extensible abstractions for rewards, algorithms, and policies. Experiments in both simulation and the real world demonstrate that USER supports multi-robot coordination, heterogeneous manipulators, cloud-edge training with large models, and long-running asynchronous training. Together, these capabilities establish USER as a unified and extensible systems foundation for real-world online policy learning.

cs.RO↗

arXiv2026

LoopFM: Learning frOm HistOrical RePresentations of Foundation Model for Recommendation

Knowledge distillation (KD) transfers a single scalar prediction from a large foundation model (FM) to compact vertical models (VMs), suffering from diminishing transfer ratio -- the fraction of FM improvement captured by the VM -- as a single scalar cannot convey the rich intermediate knowledge that larger FMs learn. To address this bottleneck, we propose LoopFM (Learning frOm HistOrical RePresentations of FM), a framework that opens a high-bandwidth transfer channel by structuring FM intermediate embeddings as input features (e.g., user history sequence) for downstream VMs, without requiring real-time FM inference at serving and architectural coupling between FM and VM. We provide a theoretical framework for LoopFM with a gain decomposition and transfer-ratio analysis. On three public benchmarks, LoopFM demonstrates strong AUC improvements (e.g., 6%+ on TaobaoAd) and complementary knowledge transfer capability with KD. On industrial-scale systems (billions of examples, trillion-parameter FMs), LoopFM approximately doubles the knowledge transfer ratio on top of KD, delivering a +0.5% conversion improvement in the first half after its initial launch, and +1.03% and +1.22% conversion improvement from two individual launches in the subsequent half. Through systematic experiments, LoopFM demonstrates a scaling law in sequence length, embedding dimension, and upstream FM size.

cs.LG↗