arXiv subjects

Xilu Wang

Publications and source records attributed to Xilu Wang.

At least 19 recordsLinked to original sources

arXiv2026

ValuSpec: Plug-and-Play Candidate Valuation before Target Verification for Tree-Based Speculative Decoding

Tree-based speculative decoding accelerates large language model inference by verifying multiple candidate branches in parallel. However, in practical serving scenarios, similar requests repeatedly induce overlapping candidate branches. While retrieval-based methods mitigate this by reusing historical drafts, the resulting candidate trees often contain many low-quality branches as the candidate tree grows, which reduces verification efficiency and wastes target-side computation. To address this, we propose ValuSpec, a plug-and-play candidate valuation mechanism that filters low-quality branches from a supplied candidate tree before expensive target-side verification. ValuSpec introduces transfer-based supervision to align filtering with the target model's behavior. A filtering threshold is estimated from the target model's greedy trajectories on a calibration split, without any parameter training. The target model then verifies the filtered tree under the output-identical greedy rule, so accepted tokens are guaranteed to match standard autoregressive decoding exactly. Experimental results on HumanEval, GSM8K, Dolly, and SpecBench show that, in our main evaluation, ValuSpec achieves up to 2.38x end-to-end speedup in retrieval-based tree settings. Furthermore, integrating ValuSpec into the EAGLE-3 dynamic tree framework yields up to 2.78x end-to-end speedup without modifying its tree construction strategy, demonstrating that ValuSpec serves as a generally applicable and plug-and-play candidate valuation module.

cs.AI↗

arXiv2026

Robust Soft Skill Adaptation via Behavioral Localization in Tool-Using Agents

Text skills can extend the capabilities of tool-using agents, but long skill descriptions increase computational costs. Soft prompts provide a compact alternative, yet learning from entire successful trajectories may also introduce unnecessary behavioral changes. We find that the useful effect of a text skill is highly concentrated in a small subset of trajectory positions. Based on this observation, we propose Behavioral Localization, which compares the frozen agent with and without the text skill to identify positions where the skill both increases the likelihood of successful tokens and meaningfully changes the next token distribution. We use these localized positions to supervise a short soft prompt. Experiments on SpreadsheetBench with a frozen Qwen3.6-35B-A3B agent show that the top 10\% of trajectory positions capture 96.12\% of the total positive skill gain. In the shared prompt setting, selective supervision improves success from 30.36\% with full trajectory to 36.79\%. Under task-specific conditions, using the top 10\% localized positions achieves 57.38\%, outperforming both the full text skill and the on-policy OPCD baseline. These results show that a small set of behaviorally relevant trajectory positions can provide effective supervision of soft skills.

cs.AI↗

arXiv2026

Beyond Correctness: Compute-Responsive Data Selection for Reasoning Model Post-Training

Reasoning language models increasingly rely on test-time compute scaling, yet most existing work studies how to allocate inference computation while treating the model itself as fixed. We ask a complementary question: can reasoning models be post-trained to make better use of a given amount of test-time compute, and which examples are most useful for this goal? Our key observation is that reasoning problems differ substantially in their response to additional computation. Some are already solved under limited compute, others remain unresolved even after further reasoning, while a distinct subset changes from complete initial failure to success when additional reasoning is provided. We call these compute-responsive (CR) examples and introduce Compute-Responsive Data Selection (CRDS), a post-training data-selection principle that prioritizes examples with empirically demonstrated marginal utility from additional reasoning. On Qwen3-8B, SFT on CR examples recovers 45.3\% of held-out low-compute failures, compared with 40.6\% when training on already-solved examples. The advantage extends beyond the selected CR split: SFT-CR reaches 51.8\% overall accuracy on an independently sampled test distribution, with the gain concentrated on compute-responsive problems. CR training also shifts the compute--accuracy frontier across inference budgets from 512 to 4,096 tokens, outperforming matched non-CR training throughout and reaching 48.1\% recovery at the largest budget. The same pattern is supported across multiple data sources, remains visible under on-policy distillation, and extends to Llama-3.1-8B-Instruct when CR sets are reconstructed separately for that model. Better test-time compute utilization can therefore be learned through post-training, with recoverable responses to additional reasoning providing a particularly informative signal for selecting training examples.

cs.AI↗

arXiv2026

Not All Errors Are Equal: Consequence-Aware Reasoning Compute Allocation

Test-time compute has emerged as an effective paradigm for improving large language model capability at inference time. Existing allocation strategies primarily prioritize tasks according to difficulty, uncertainty, or expected performance gain, implicitly treating prediction errors as equally costly. This assumption is often misaligned with real deployment, where failures can differ substantially in their downstream tasks. To address this limitation, this paper introduces consequence-aware test-time compute allocation by formulating a cost-weighted scheduling problem where the priority of a task is its failure consequence with the marginal gain of additional compute. In practice, however, marginal gain is difficult to predict before execution, so we propose a deployable scheduler that uses consequence as the routing signal. The scheduler predicts task consequence from pre-solution inputs and allocates the available premium compute to the corresponding top-ranked tasks. Experiments on the SWE bench Lite show that consequence provides information beyond task difficulty and can be predicted before solving. Under a fixed compute budget, consequence-aware routing achieves the best high-consequence task success, while overall accuracy remains competitive. A controlled within-model experiment further confirms the same advantage when only inference attempts are reallocated.

cs.AI↗

arXiv2026

SE-ADD: Self-Evolving Audio Deepfake Detection with Mistake-Driven Supervision

Audio deepfake detection (ADD) must remain effective when new spoofing attacks emerge after deployment. Emerging audio language model (ALM)-based ADD methods are built on predefined supervision from ground-truth labels or verified forensic rationales. However, this paradigm overlooks an ALM's own mistakes, which indicate where targeted supervision is most needed. To this end, we first introduce evolving spoofing environments for ALM-based ADD, where a new attack becomes dominant while previously observed attacks persist. Motivated by the above learning-from-mistakes perspective, we further propose SE-ADD, a self-evolving framework that iteratively adapts an ALM via low-rank adaptation (LoRA) using mistake-driven supervision built from its verdicts and self-generated forensic cues. All training samples receive direct authenticity supervision, while misclassified ones receive additional cue-augmented supervision. As verdicts and cues are regenerated by the updated ALM, the resulting supervision evolves accordingly. Experiments on two ALMs demonstrate the effectiveness of SE-ADD in generalizing to unseen attacks, reducing the equal error rate (EER) from $36.72\%$ to $7.52\%$ for Qwen2-Audio and from $19.93\%$ to $3.97\%$ for MOSS-Audio.

cs.SD↗

arXiv2026

SEAR: Spoofing Evidence-Grounded Audio Reasoning Benchmark for Audio Language Models

Audio language models (ALMs) are increasingly used for audio deepfake detection (ADD), yet existing benchmarks assess their verdicts or rationale plausibility without verifying the underlying acoustic evidence. To address this issue, we first introduce spoofing evidence-grounded audio reasoning (SEAR), a four-task AQA benchmark to evaluate ALM-based ADD through acoustic evidence identification and quantification, deepfake detection, and forensic rationale generation. We further propose a bona-fide-based acoustic evidence agent (BAEA), which equips a frozen ALM with controlled acoustic tools under \textsc{fixed} or \textsc{adaptive} evidence-acquisition policies. Experiments with six ALMs reveal a clear gap between plausible rationales and verifiable acoustic evidence reasoning, while BAEA-\textsc{Fixed} improves final verdicts and forensic rationales on both evaluation partitions. Controlled interventions further show that misleading evidence degrades both detection and grounding performance.

cs.SD↗

arXiv2026

You Cannot Pick a Provider From the Price List: Market-Aware Routing for Open-Weight LLM Inference

Existing LLM routers choose among models using static per-model costs. We show that open-weight inference markets introduce a second, largely ignored decision axis: after choosing a model, a client must still choose which provider serves it. Measuring live endpoints across [nummodels] open models, competing providers, multiple task types, and three measurement waves, we find that provider choice cannot be inferred from the price list. The same model can vary sharply in quality, latency, availability, and price across providers; higher-priced providers are consistently faster, but price does not reliably predict quality or availability; and provider feasibility is task-selective, with one deployment nearly normal on knowledge tasks but catastrophically degraded on multi-step reasoning. We formulate same-model provider selection as a price-taker market-aware routing problem. A simple measured-map policy routes to the cheapest provider that is both quality-equivalent and healthy, yielding matched-quality savings while avoiding degraded endpoints. Because the map drifts, we introduce FACET, an online provider router that certifies per-(provider x task) feasibility facets and fails safe to an anchor before serving uncertified arms. Across relaxed deployment assumptions, FACET tolerates imperfect task assignment and sparse feedback, while systematic evaluator bias exposes a quality-signal trust boundary that can be mitigated with ground-truth probes or audits. Live provider runs further confirm that certification can move real traffic from a premium anchor to a substantially cheaper certified endpoint. Our results suggest that market-aware LLM routing must measure not only which model to use, but also who serves it.

cs.AI↗

arXiv2026

HESS J1507-622: A Plausible Young Galactic Kilonova Remnant

HESS J1507$-$622 is an unusual TeV gamma-ray source in the H.E.S.S. Galactic Plane Survey, distinguished by its off-plane location ($b \simeq -3.5^{\circ}$) and compact angular extent ($\sim 0.3^{\circ}$). Its location and morphology challenge conventional pulsar wind nebula and supernova remnant interpretations, and the absence of a detected shell or associated pulsar leaves its origin unresolved. Here, we propose that HESS J1507$-$622 is a young Galactic kilonova remnant with gamma-ray emission of leptonic origin. In this scenario, the source's off-plane location is naturally explained by the natal kick imparted to the progenitor binary neutron star system. We demonstrate that a leptonic kilonova remnant model can successfully reproduce the observed GeV-TeV flux via inverse-Compton scattering from shock-accelerated electrons, while the corresponding synchrotron emission remains below current X-ray and radio upper limits. Based on the source's energetics and angular size, we infer a distance of 3.8-14.3 kpc and an age of 0.2-3.0 kyr. Importantly, our analysis favors a kilonova remnant younger than 1 kyr. The lack of a historical naked-eye record can be explained by the transient's rapid fading, far-southern declination, and potential line-of-sight extinction. We further assess the detectability of unique multi-wavelength signatures that could confirm this hypothesis. We find promising prospects for detecting the unique MeV gamma rays from the decay of $r$-process radioisotopes using future observatories, alongside complementary low-energy signatures from thermal dust emission and a possible optical light echo. These distinct signals offer definitive tests of the kilonova remnant interpretation and motivate targeted observations with current and future facilities.

astro-ph.HE↗

arXiv2026

Grounded Checklist Partial Credit for Agent Skill Trajectories

Language-model agents increasingly tackle long-horizon tasks in interactive environments, yet their evaluation commonly relies on task-level success rates by reducing an entire execution trajectory to whether the task passes an official verifier. This binary score hides partial progress and is particularly limited for procedural agent skill evaluations, since a skill can alter execution without changing the final outcome. While checklists provide finer-grained evaluation by scoring individual task requirements, costly manual authoring and unreliable automatic generation make trustworthy evaluation difficult to scale. To address these challenges, we introduce Grounded Checklist Partial Credit (GCPC), a human-governed and LLM-instantiated partial-credit evaluation of agent trajectories. Humans define reusable rules once, from which an LLM instantiates a task-specific checklist grounded in the task instruction and official verifier. To keep judgment tied to evidence, a judge scores each item from execution log evidence alone and abstains when evidence is missing. A separate scripted step then applies the official verifier outcome to the score. Across a 4,455-trajectory, deduplicated SkillsBench evaluation population, GCPC better discriminates official PASS and FAIL outcomes than holistic judging on the shared subset (AUC 0.689 vs. 0.619). Human evaluation on 96 trajectories from 12 tasks shows that GCPC aligns more closely with human assessments of progress. Applied to 1,946 matched with/without-skill pairs, GCPC exposes the effects hidden by pass@1: among 879 pairs whose binary outcome does not change, 20.9% improve by more than 0.10 while 18.7% regress by the same margin. The GCPC pipeline also transfers to Terminal-Bench and SWE-bench, demonstrating applicability beyond skill-conditioned evaluation.

cs.SE↗

arXiv2026

From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based LLM Agents

Agent skills are increasingly used to equip large language model (LLM) agents with reusable procedural knowledge. Although recent work has substantially improved skill retrieval due to the increasing skill libraries, retrieving a plausible skill bundle does not guarantee that executing it is worthwhile. Since every skill-conditioned rollout is computationally expensive, deciding whether a retrieved bundle should be executed has become an increasingly important challenge. To this end, we introduce the Reward-Aware Dynamic Execution Gate (RADEG), a lightweight, retriever-agnostic decision layer between skill retrieval and agent execution. RADEG learns a low-cost surrogate model that predicts the execution utility of a query--bundle pair before the expensive rollout is launched. To obtain informative supervision while controlling for task difficulty, we locally perturb each retrieved bundle by deleting, adding, or replacing one skill, producing matched same-query rollouts that isolate the effect of bundle composition on verifier reward. During deployment, RADEG updates only a warm-started logistic head as new verifier feedback becomes available, enabling inexpensive adaptation of the execute/skip boundary without retraining either the retriever or the agent. Under a query-level held-out evaluation on 288 collected rollouts, RADEG substantially reduces unnecessary agent executions while preserving a large fraction of the downstream verifier reward. It consistently outperforms relevance-based and random gating across different execution budgets, demonstrating that execution-aware surrogate modeling provides a practical and cost-effective complement to skill retrieval.

cs.AI↗

arXiv2026

Not All Visual Tokens Are Equally Safe to Remove:Consequence-Sensitive Visual Token Compression

Visual token compression for vision--language models (VLMs) has largely relied on criteria such as attention, redundancy, and uncertainty to maximize average accuracy under a fixed compute budget, implicitly assuming that all errors carry equal cost. However, the consequence of an incorrect prediction on downstream tasks is rarely symmetric: misreading an invoice amount can be far more costly than misclassifying a background color. Motivated by this, we introduce consequence-sensitive visual token compression, which allocates visual computation across requests according to their potential error costs. Our method follows a calibrate-then-allocate procedure, estimating consequence-specific error-budget curves offline and applying the calibrated token budgets online using consequence signals available from question or task information. On a controlled within-task benchmark, high- and low-consequence questions are drawn from the same document images, so content alone cannot reveal which questions are costly to get wrong. In this setting, our method reduces high-stakes errors from 0.300 to 0.133 under the same total token budget, whereas a content-driven allocator performs no better than uniform allocation. Measuring how error rates change with token budget across different cost ratios, we derive an allocation frontier: uniform allocation is optimal when errors are equally costly, and token transfer toward high-consequence questions becomes increasingly beneficial as the cost gap grows. This allocation principle generalizes well across three dense vision-language benchmarks, two budget realization mechanisms (token deletion and resolution reallocation), two VLM architectures, and multiple token selection strategies. On a realistic mixed workload, consequence-sensitive allocation reduces cost-weighted error by 38% while achieving approximately 21% lower latency than full-resolution inference.

cs.CV↗

arXiv2026

CRIP: Channel Level Representation Injection for Personalized One-Shot Federated Learning

One-shot federated learning (OSFL) has emerged as a promising collaborative model learning framework with only a single round of communication, offering significant advantages in communication efficiency and privacy preservation. However, OSFL often faces inherent limitations under severe domain heterogeneity across clients due to the lack of iterative knowledge exchange. Most existing OSFL methods require an auxiliary public dataset for knowledge distillation or leverage statistical information for parameter-level aggregation, overlooking feature shift caused by domain heterogeneity. To address these challenges, we propose CRIP, a personalized OSFL framework that operates in the representation space via channel-level feature alignment. To achieve this, each client uploads its feature extractor to the server, which broadcasts all extractors back to every client. Since not all source clients share compatible feature distributions with the target client, indiscriminate fusion of cross-client features would introduce domain-specific noise. Therefore, CRIP effectively measures the channel-wise representational similarity between the target client and each source client on a small local mini-batch, and selectively fuses only the most compatible features. Extensive experiments on domain-heterogeneous benchmarks such as DomainNet, PACS, and Office-Home demonstrate that CRIP consistently outperforms local models and state-of-the-art baselines, validating the effectiveness of representation-space personalization under extreme domain heterogeneity.

cs.LG↗

arXiv2026

TSegAgent: Zero-Shot Tooth Segmentation via Geometry-Aware Vision-Language Agents

Automatic tooth segmentation and identification from intra-oral scanned 3D models are fundamental problems in digital dentistry, yet most existing approaches rely on task-specific 3D neural networks trained with densely annotated datasets, resulting in high annotation cost and limited generalization to scans from unseen sources. Thus, we propose TSegAgent, which addresses these challenges by reformulating dental analysis as a zero-shot geometric reasoning problem rather than a purely data-driven recognition task. The key idea is to combine the representational capacity of general-purpose foundation models with explicit geometric inductive biases derived from dental anatomy. Instead of learning dental-specific features, the proposed framework leverages multi-view visual abstraction and geometry-grounded reasoning to infer tooth instances and identities without task-specific training. By explicitly encoding structural constraints such as dental arch organization and volumetric relationships, the method reduces uncertainty in ambiguous cases and mitigates overfitting to particular shape distributions. Experimental results demonstrate that this reasoning-oriented formulation enables accurate and reliable tooth segmentation and identification with low computational and annotation cost, while exhibiting strong generalization across diverse and previously unseen dental scans.

cs.CV↗

arXiv2026

BudgetDraft: Acceptance-Aware Multi-View Training for Sparse-KV Speculative Decoding

Speculative decoding speeds up autoregressive decoding by using a drafter to propose multiple tokens that a verifier validates in parallel. In resource-constrained deployments, the drafter uses a sparse KV cache to limit peak GPU memory and end-to-end latency under a fixed KV budget, while the verifier keeps a full KV cache. Mid-to-long context inference (4K--16K context length) is common in real applications. However, naive sparse/full speculative decoding suffers from the sparse/full mismatch as context length grows, causing the acceptance rate to drop quickly. We propose BudgetDraft, a multi-view sparse training method for sparse drafting in mid-to-long inference. The drafter is exposed to multiple sampled KV budgets during training and learns to align each sparse view with one shared full-cache teacher target. BudgetDraft combines an acceptance-aware loss on a full-cache branch with a multi-view loss on a sparse-cache branch, producing a single budget-robust drafter that recovers acceptance across sparsity levels without extra inference-time components. Experimental results on PG-19, LongBench, and LWM show that BudgetDraft achieves up to 6.55x, 4.46x, and 2.10x end-to-end speedup vs AR at 4K, 8K, and 16K context lengths, while keeping the inference pipeline memory-friendly.

cs.LG↗

arXiv2026

Identifying observable MeV lines from the decays of weak and main $r$-process isotopes in mergers

We consider predictions for the MeV gamma-ray spectrum emitted by the $β$ decays of freshly synthesized isotopes from a neutron star merger at timescales of relevance for post-merger (days) and remnant (years) emission. We develop a search algorithm to identify observable spectral peaks and then determine if a specific isotope has a dominant emission line producing the spectral feature. We predict emission spectra using nucleosynthesis calculations which consider nuclear models with distinct masses, $β$-decays, and fission properties as well as variations on main ($A>130$) and weak ($A<130$) $r$-process astrophysical conditions. We tabulate all lines from decaying isotopes that our procedure identifies and provide the predicted range in time over which each line could be visible. We find that Rh-106 presents a unique opportunity to distinguish between main and weak $r$-process emission, as our calculated spectrum above $\sim 1$ MeV for an event dominated by the weak $r$ process is identical to the Rh-106 emission spectrum from $\sim$ 0.2 to $\sim$17 years. We further find emission from species such as Hf-181, Ta-182, Ta-184, and Re-188 offers the potential to be able to distinguish between nuclear models. We investigate whether the 2.6 MeV strong gamma-ray line from Tl-208 is predicted to be robustly observable across calculation variations on both timescale of days and years. We find Tl-208 to consistently shine through on the order of years, though it can face competition from Ga-72 and La-140 at early times ($\sim$ days). We additionally highlight numerous isotopes of interest for observation and nuclear experiment.

astro-ph.HE↗

arXiv2026

ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity

Large Language Models (LLMs) have achieved remarkable capabilities, but their immense computational demands during training remain a critical bottleneck for widespread adoption. Low-rank training has received attention in recent years due to its ability to significantly reduce training memory usage. Meanwhile, applying 2:4 structured sparsity to weights and activations to leverage NVIDIA GPU support for 2:4 structured sparse format has become a promising direction. However, existing low-rank methods often leave activation matrices in full-rank, which dominates memory consumption and limits throughput during large-batch training. Furthermore, directly applying sparsity to weights often leads to non-negligible performance degradation. To achieve efficient pre-training of LLMs, this paper proposes ELAS: Efficient pre-training of Low-rank LLMs via 2:4 Activation Sparsity, a novel framework for low-rank models via 2:4 activation sparsity. ELAS applies squared ReLU activation functions to the feed-forward networks in low-rank models and implements 2:4 structured sparsity on the activations after the squared ReLU operation. We evaluated ELAS through pre-training experiments on LLaMA models ranging from 60M to 1B parameters. The results demonstrate that ELAS maintains performance with minimal degradation after applying 2:4 activation sparsity, while achieving training and inference acceleration. Moreover, ELAS reduces activation memory overhead, particularly with large batch sizes. Code is available at ELAS Repo.

cs.LG↗

arXiv2026

Gardening on the Moon: An Advection-Diffusion Model to Guide the Search for Supernova Debris in the Lunar Regolith

The vertical redistribution of materials in the lunar regolith - ranging from continuously produced space-weathering products to sporadic pulses of supernova- or kilonova-derived isotopes - remains a fundamental problem in planetary science. We present a unified stochastic model of regolith gardening induced by the impact flux. Treating gardening as a competition between impact-driven advection and diffusion predicts the maturity profiles of Apollo cores over more than two orders of magnitude in time ($1.4 \times 10^7$ to $4.5 \times 10^8$ years). This model describes well the depth profiles of live Fe60 in Apollo regolith samples, suggesting that supernova dust capture is independent of native iron abundance, and is consistent with a uniform influx at the latitudes of the Apollo landing sites. We extend our model to predict lunar signals for live r-process species that might originate from supernovae or kilonovae: Pu244 tied to terrestrial detections, and I129, Hf182, and Cm247 based on r-process calculations. The Pu244/Fe60 depth profile can probe the origin of Pu244, motivating searches in Artemis regolith samples down to depths O(100) cm.

astro-ph.EP↗

arXiv2026

High-energy Neutrino Predictions for T Coronae Borealis: Probing Particle Acceleration in Novae

The MAGIC detection of near-TeV gamma rays from the 2021 RS Oph ($2.45$ kpc) outburst has established recurrent novae as TeV particle accelerators. However, the origin of this emission (hadronic vs leptonic) remains unclear due to the lack of coincident neutrinos detected by IceCube. The upcoming outburst of the much closer T Coronae Borealis (T CrB, $\sim0.887$ kpc) offers a unique opportunity to detect these rare nova neutrinos. Here we present the first comparative analysis of the hadronic secondary fluxes expected from the upcoming T CrB outburst and evaluate their detectability across major observatories, considering two proton-acceleration mechanisms: (i) an external shock (ES) at $\sim10^{13}$ cm, and (ii) magnetic reconnection (MR), near the white dwarf surface at $\sim10^{9}$ cm. While the benchmark ES model predicts a gamma-ray flux detectable by current facilities, its corresponding neutrino flux largely remains undetectable. In contrast, the MR scenario generates a robust neutrino flux within the reach of IceCube and KM3NeT. Importantly, as the MR-produced gamma-rays are absorbed, the escaping MR neutrinos will arrive hours before any ES-origin signals. This distinct temporal separation can create a powerful phenomenological signature to disentangle the nova acceleration physics.

astro-ph.HE↗