Search arXiv⌕ Search

arXiv subjects

Zihao Wang

Publications and source records attributed to Zihao Wang.

At least 91 records · Page 5Linked to original sources
arXiv2026

A Differentiable Composite Approximation Framework for Autonomous Underwater Vehicle Maneuvering Modeling from Sea-Trial Data

Field-based modeling from onboard measurements can produce autonomous underwater vehicle (AUV) maneuvering models that reflect real operating characteristics. From an approximation perspective, conventional maneuvering models use predefined constraint polynomial bases, whereas data-driven models use data-adaptive bases. Motivated by this basis-function view, this paper presents a differentiable composite-approximation formulation, in which the polynomial-basis component and the data-adaptive basis component are treated as differentiable parts of a single predictor and calibrated jointly. A gradient-based co-calibration method is developed for full-scale AUV maneuvering prediction, where a sensitivity-aware mechanism regulates bounded polynomial updates while the neural residual captures remaining nonlinear discrepancies under a shared prediction objective. To account for ocean-current effects in field data, a turning-motion-based current estimation and compensation procedure is incorporated to construct current-compensated learning targets for training and rollout. The framework is evaluated using sea-trial data collected from a 7-meter AUV under multiple maneuvering conditions. Results show that the proposed method improves recursive trajectory and velocity prediction compared with polynomial-only, neural-only, and frozen-prior hybrid baselines, demonstrating its applicability to field-data-based AUV maneuvering modeling.

cs.RO↗
arXiv2026

Rubric-Guided Self-Distillation: Post-Training Without Rubric Verifiers

Rubrics have emerged as an alternative to RLVR in open-ended domains where a single ground-truth final answer is not available. Existing rubric-based training methods rely on an LLM verifier that scores each rollout against rubrics. This introduces substantial training-time overhead, exposes optimization to verifier-specific biases, and reduces rubric feedback to a sparse end-of-trajectory signal. We propose Rubric-Guided Self-Distillation (RGSD), a verifier-free training method in which the base policy, conditioned on the rubric, serves as the teacher for the unconditioned student. RGSD distills the rubric-conditioned teacher distribution into the student token-by-token, replacing sparse trajectory-level rewards with dense per-token learning signals and removing the LLM judge from the training loop entirely. Across Qwen-2.5 (3B, 7B) and Qwen3-Thinking (4B, 8B) models on medical and science domains, RGSD achieves rubric satisfaction comparable to judge-based GRPO while using one on-policy rollout per prompt and no training-time verifier calls. Ablations show that raw rubrics provide a stronger teacher enrichment signal than self-generated reference responses, while a stronger GRPO judge can outperform RGSD in some settings, positioning RGSD as a complementary verifier-free alternative when verifier cost or reliability is the bottleneck.

cs.LG↗
arXiv2026

Neural Scalable Symbolic Search Framework for Complex Logical Queries with Multiple Free Variables

Complex Query Answering (CQA) is a fundamental knowledge representation and reasoning task over incomplete knowledge graphs (KGs). Answering existential first-order queries with $k$ free variables (i.e., $\text{EFO}_k$ queries) is a crucial yet challenging problem, as it requires ranking answer tuples in $\mathcal{E}^k$, where $\mathcal{E}$ denotes the entity set of a KG. This quickly becomes intractable as $k$ grows. Consequently, existing benchmarks and methods rely on marginal rankings over individual variables; however, marginal rankings are a poor proxy for the true joint ranking of tuples. Building on neural symbolic search for $\text{EFO}_1$ queries, we propose Neural Scalable Symbolic Search (NS3), a budgeted framework that approximates joint ranking without enumerating $\mathcal{E}^k$. NS3 (i) answers marginalized sub-queries to obtain necessary candidate sets, (ii) merges multiple free variables into hypernodes whose domains are pruned and controlled by a dynamic budget $B$, and (iii) progressively reduces an $\text{EFO}_k$ query to an $\text{EFO}_{k-1}$ query over a budgeted reduced domain. Across three standard KG datasets, NS3 substantially improves joint ranking performance while retaining strong marginal accuracy. We further release a joint-ranking benchmark that extends existing $\text{EFO}_1$ datasets to $k=3$, enabling systematic evaluation of multi-variable queries. Our code is provided in https://github.com/HKUST-KnowComp/NS3_KDD2026.

cs.AI↗
arXiv2026

Cooperative Long Rope Skipping via Multi-Agent Reinforcement Learning

Humans exhibit remarkable motor agility, enabling a wide range of dynamic skills such as running and jumping, which highlights the great potential of humanoid robots for athletic locomotion. Among athletic sports, long rope skipping requires two rope turners to cooperatively swing the rope while adapting to a player under different jumping rhythms, making it a meaningful yet challenging task for humanoid robots. Although existing methods for humanoid sports have achieved success in single-agent and interaction-free settings, such as running, dancing, and parkour, task scenarios that require precise coordination among multiple participants remain largely unexplored. To this end, we propose Marope, a multi-agent reinforcement learning (MARL) framework for cooperative long rope skipping with multiple humanoid robots. Specifically, Marope adopts a hierarchical reinforcement learning framework for policy training. At the lower level, it learns decentralized rope manipulation policies through MARL, while at the upper level, a centralized scheduling policy is trained to coordinate the execution of the lower-level policies. To improve generalization across different player behavioral styles, Marope further incorporates diverse jumping policies into cooperative game training. We evaluate our approach on Unitree G1 humanoid robots in both simulation and real-world settings. Experimental results demonstrate that Marope outperforms various baselines, achieving more efficient and stable rope manipulation as well as more robust and adaptable cooperation with varied players.

cs.RO↗
arXiv2026

Training One Model to Master Cross-Level Agentic Actions via Reinforcement Learning

The paradigm of agentic AI is shifting from engineered complex workflows to post-training native models. However, existing agents are typically confined to static, predefined action spaces-such as exclusively using APIs, GUI events, or robotic commands. This rigidity limits their adaptability in dynamic environments where the optimal granularity of interaction varies contextually. To bridge this gap, we propose CrossHA, a unified agentic model that masters heterogeneous action spaces and autonomously selects the most effective interface for each step of a trajectory. We introduce a comprehensive training pipeline that integrates cold-start supervised fine-tuning with a Multi-Turn Group Relative Policy Optimization (GRPO) algorithm. This approach enables the agent to learn adaptive action switching-balancing high-level efficiency with low-level precision-without human-specified rules. Extensive experiments on over 800 tasks in the open-world Minecraft environment demonstrate that CrossHA achieves state-of-the-art performance. By dynamically leveraging the strengths of diverse action spaces, our model significantly outperforms fixed-action baselines, exhibiting superior generalization and efficiency in long-horizon reasoning. All code and models are available at https://github.com/CraftJarvis/OpenHA.

cs.LG↗
arXiv2026

$\mathbb{R}^{2k}$ is Theoretically Large Enough for Embedding-based Top-$k$ Retrieval

This paper studies the Minimal Embeddable Dimension (MED): the least dimension in which there exists a configuration of $m$ object vectors so that every subset of size at most $k$ is exactly retrieved by score comparison. Our result shows MED is $Θ(k)$, independent of $m$, for inner product, Euclidean distance, and cosine similarity. We then consider Robust MED (RMED), where all vectors are unit normed and an $ε$ gap of scores is required. We derive the $m$-dependent feasibility ceiling $ε_\star(m,k)=m/\sqrt{k(m-1)(m-k)}$, which approaches $1/\sqrt{k}$ when $m\gg k$, and a Gaussian centroid construction gives a robust witness upper bound in the feasible margin regime. Numerical simulation on synthetic top-$2$ retrieval with cyclic polytope and centroid query optimization confirmed our theoretical claims. Experiments on LIMIT and LIMIT-small datasets also show that simple embedding-based retrieval baselines can overfit and outperform the reported single-vector LLM embedding baseline. Both theoretical and empirical findings rule out the lack of exact geometric capacity as the obstruction.

cs.LG↗
arXiv2026

Efficient and Scalable Neural Symbolic Search for Knowledge Graph Complex Query Answering

Complex Query Answering (CQA) is a crucial reasoning task over Knowledge Graphs (KGs), which aims to answer first-order logical queries from incomplete KGs. While existing neural-symbolic methods achieve strong performance, they face significant complexity bottlenecks: quadratic data complexity scaling with the number of entities, and NP-hard query complexity for cyclic queries. Consequently, these approaches struggle to scale effectively to large knowledge graphs and complex queries. To address these limitations, we propose an efficient and scalable symbolic search method comprising two key components: (1) constraint strategies that drastically reduce the variable search domain, lowering data complexity; and (2) a local search algorithm that approximately solves NP-hard cyclic queries. Experiments on various CQA benchmarks demonstrate that, for tree-form queries, our method achieves 97% relative MRR with a 10$\times$ speedup using only 10% of the search space. Furthermore, it demonstrates robust performance on complex cyclic queries and large-scale KGs, effectively alleviating efficiency and scalability challenges. Our code is provided in https://github.com/HKUST-KnowComp/NLISA_KDD2026.

cs.AI↗
arXiv2026

Towards Brain MRI Foundation Models for the Clinic: Findings from the FOMO25 Challenge

Clinical deployment of automated brain MRI analysis faces a fundamental challenge: clinical data is heterogeneous and noisy, and high-quality labels are prohibitively costly to obtain. Self-supervised learning (SSL) can address this by leveraging the vast amounts of unlabeled data produced in clinical workflows to train robust \textit{foundation models} that adapt out-of-domain with minimal supervision. However, the development of foundation models for brain MRI has been limited by small pretraining datasets and in-domain benchmarking focused on high-quality, research-grade data. To address this gap, we organized the FOMO25 challenge as a satellite event at MICCAI 2025. FOMO25 provided participants with a large pretraining dataset, FOMO60K, and evaluated models on data sourced directly from clinical workflows in few-shot and out-of-domain settings. Tasks covered infarct classification, meningioma segmentation, and brain age regression, and considered both models trained on FOMO60K (method track) and any data (open track). Nineteen foundation models from sixteen teams were evaluated using a standardized containerized pipeline. Results show that (a) self-supervised pretraining improves generalization on clinical data under domain shift, with the strongest models trained \textit{out-of-domain} surpassing supervised baselines trained \textit{in-domain}. (b) No single pretraining objective benefits all tasks: MAE favors segmentation, hybrid reconstruction-contrastive objectives favor classification, and (c) strong performance was achieved by small pretrained models, and improvements from scaling model size and training duration did not yield reliable benefits.

cs.CV↗
arXiv2026

Rethinking Side-Channel Analysis: Automated Discovery and Analysis of Side-Channel Leakage with LLM-Assisted Agents

Side-channel attacks exploit unintended information leakage from system behavior and continue to pose serious privacy risks in modern platforms. Despite extensive prior work, side-channel analysis remains largely manual and fragmented, typically assuming predefined target events and a fixed set of known channels. As systems and applications grow increasingly complex, several fundamental questions remain unanswered: which user or system events are sensitive in practice, how side channels associated with these events can be systematically discovered without exhaustive manual effort, and how their leakage can be analyzed at scale without prohibitive data collection and model training costs. To address these questions, we present SCAgent, an automated framework for side-channel risk analysis. To identify sensitive targets beyond manually specified events, SCAgent performs agent-driven system exploration guided by LLM-based semantic reasoning. To systematically discover side channels while mitigating the risk of LLM hallucination, it reasons over system documentation and incorporates explicit verification to enforce semantic consistency, threat-model feasibility, and per-channel usability. To enable scalable analysis under limited data, SCAgent adopts a few-shot learning paradigm based on foundation models, avoiding the need to train bespoke models for each channel--event pair. To bridge the gap between raw time-series side-channel signals and tabular foundation models, SCAgent further introduces a time-shift--robust feature extraction layer that enables effective downstream analysis. We instantiate SCAgent on iOS as a first step, focusing on OS-level side channels observable by unprivileged applications. Our evaluation spans standard benchmarks such as foreground app and website fingerprinting, as well as newly identified sensitive in-app activities in popular applications.

cs.CR↗
arXiv2026

DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Models

Large language models (LLMs) are commonly adapted to downstream tasks through fine-tuning, but fine-tuning data often contains sensitive information that may be leaked by the resulting model. Differential privacy (DP) offers formal protection against such leakage, yet DP fine-tuning of LLMs still suffers from substantial utility degradation due to gradient clipping and noise injection. Existing work improves this trade-off by combining DP with parameter-efficient fine-tuning methods such as LoRA, which constrain the form of updates. In this work, we study a complementary direction: selective fine-tuning, which constrains where updates are applied. We propose DP-SelFT, a framework for differentially private selective fine-tuning of LLMs. DP-SelFT addresses three DP-specific challenges in parameter selection: avoiding repeated privacy cost, improving stability under noisy estimates, and selecting parameters that remain useful under clipped and noisy updates. It first constructs a lightweight DP synthetic dataset and performs selection only on this synthetic data, so the selection stage incurs no additional privacy cost. It then conducts layer-level selection by temporarily training candidate layer subsets on a synthetic training split and evaluating them on a synthetic validation split. Crucially, this temporary training is performed under a perturbation regime matched to downstream DP fine-tuning, with worst-case perturbations of the same scale as DP noise. This favors layer subsets that are not only learnable but also robust to noisy private updates. Experiments on benchmark tasks show that DP-SelFT consistently improves the privacy--utility trade-off over existing DP fine-tuning baselines under the same privacy guarantees.

cs.LG↗
arXiv2026

Reward Hacking in Rubric-Based Reinforcement Learning

Reinforcement learning with verifiable rewards has enabled strong post-training gains in domains such as math and coding, though many open-ended settings rely on rubric-based rewards. We study reward hacking in rubric-based RL, where a policy is optimized against a training verifier but evaluated against a cross-family panel of three frontier judges, reducing dependence on any single evaluator. Our framework separates two sources of divergence: verifier failure, where the training verifier credits rubric criteria that reference verifiers reject, and rubric-design limitations, where even strong rubric-based verifiers favor responses that rubric-free judges rate worse overall. Across medical and science domains, weak verifiers produce large proxy-reward gains that do not transfer to the reference verifiers; exploitation grows over training and concentrates in recurring failures such as partial satisfaction of compound criteria, treating implicit content as explicit, and imprecise topical matching. Stronger verifiers substantially reduce, but do not eliminate, verifier exploitation. We also introduce a self-internalization gap, a verifier-free diagnostic based on policy log-probabilities, which tracks reference-verifier quality, detecting when the policy trained using the weak verifier stops improving. Finally, in our setting, stronger verification does not prevent reward hacking when the rubric leaves important failure modes unspecified: rubric-based verifiers prefer the RL checkpoint, while rubric-free judges prefer the base model. These disagreements coincide with gains concentrated in completeness and presence-based criteria, alongside declines in factual correctness, conciseness, relevance, and overall quality. Together, these results suggest that stronger verification reduces reward hacking, but does not by itself ensure that rubric gains correspond to broader quality gains.

cs.AI↗
arXiv2026

Misrouter: Exploiting Routing Mechanisms for Input-Only Attacks on Mixture-of-Experts LLMs

Mixture-of-Experts (MoE) architectures have emerged as a leading paradigm for scaling large language models through sparse, routing-based computation. However, this design introduces a new attack surface: the routing mechanism that determines which experts process each input. Prior work shows that manipulating routing can bypass safety alignment, but existing attacks require model modification and thus apply only to locally deployed models. By contrast, real-world LLM services are remotely hosted and accessible only through input queries. This raises a fundamental question: can MoE routing be exploited through input-only attacks to induce stronger unsafe behaviors in real-world services? Our key insight is to optimize attacks in a white-box setting on open-source surrogate MoE models and transfer the resulting adversarial inputs to public API services within the same model family. This setting presents three main challenges: routing can be influenced only indirectly through input perturbations, routing control and output generation are tightly coupled, and even a successful safety bypass may still produce low-quality responses. To address these challenges, we propose Misrouter, an input-only attack framework that jointly targets routing behavior and expert functionality. Misrouter identifies weakly aligned experts that are willing to produce target harmful content by analyzing expert activations under harmful queries paired with unsafe continuations. It then optimizes adversarial inputs to steer routing toward these experts and away from strongly aligned ones. It further biases routing toward highly capable general-purpose experts identified from benign question-answering tasks. Finally, because routing and output objectives can conflict, Misrouter uses a two-phase optimization strategy that first steers routing and then optimizes harmful outputs while preserving routing stability.

cs.CR↗
arXiv2026

Rethinking Traffic Matrix Completion: Estimate the Process, Not the Entries

Traffic matrix measurement is fundamental for datacenter operations, but obtaining complete traffic matrices at scale remains challenging due to the prohibitive cost of global fine-grained measurement and partial observations resulting from network faults. Although existing matrix completion methods (reduce cost) achieve satisfactory performance in specific scenarios, their reliance on restrictive assumptions or black-box mappings results in a lack of interpretability and an inability to characterize uncertainty. In this paper, we propose Utimac, an uncertainty-aware traffic matrix completion for data center networks. Our analysis shows that, within a locally stationary window, log-domain traffic can be decomposed into a principal statistical component and a sparse deviation component. Based on this insight, we formulate traffic matrix completion as a parameter inference problem: multiple partially observed frames within a window are used to infer shared parameters and recover missing entries. To avoid the intractability and boundary degeneracy of the original integral-form marginal likelihood, we construct a regularized surrogate objective and solve the resulting joint optimization problem with block coordinate descent. Utimac consistently outperforms all baselines on data center networks datasets in both overall and burst scenarios, with its advantage becoming more pronounced as observations grow sparser. All code is publicly available in an anonymous repository: https://anonymous.4open.science/r/Utimac-0551/

cs.NI↗
arXiv2026

Hey, That's My Data! Token-Only Dataset Inference in Large Language Models

Large Language Models (LLMs) rely on massive training datasets, often including proprietary data, which raises concerns about unauthorized usage and copyright infringement. Existing dataset inference methods typically require access to log probabilities or other internal signals, but many modern LLMs restrict such access, motivating token-only inference approaches. We propose CatShift, a token-only dataset inference framework based on catastrophic forgetting, where models overwrite prior knowledge when trained on new data. Fine-tuning an LLM on a subset of its training data induces larger output shifts than fine-tuning on unseen data. CatShift compares these shifts against those from a known non-member validation set to infer whether a dataset was included in training. Experiments on both open-source and API-based LLMs show that CatShift remains effective without logit access, enabling practical protection of proprietary datasets.

cs.CL↗
arXiv2026

Quantum Mpemba-like effect in Unruh thermalization

We revisit the thermal nature of the Unruh effect within a quantum thermodynamic framework. For a Unruh-deWitt (UDW) detector in $n$-dimensional Minkowski spacetime, we demonstrate that its irreversible thermalization to a Gibbs equilibrium state follows distinct trajectories on the Bloch sphere, which depend on the types of fields the detector interacts with, as well as the spacetime dimensionality. Using thermodynamic process functions, particularly quantum coherence and heat that form the quantum First Law, we characterize the Unruh thermalization through a complementary time evolution between the trajectory-dependent rates of process functions. Grounded in information geometry, we further explore the kinematics of the detector state as it "flows" along the trajectory. In particular, we propose two heating/cooling protocols for the UDW detector undergoing Unruh thermalization. We observe a quantum Mpemba-like effect, characterized by faster heating than cooling in terms of Uhlmann fidelity "distance" change. Most significantly, we establish the maximum fidelity difference as a novel diagnostic that essentially distinguishes between Unruh thermalization and its classical counterpart, i.e., classical thermal bath-driven thermalization of an inertial UDW detector. This compelling criterion may serve as a hallmark of the quantum origin of the Unruh effect in future experimental detection and quantum simulation. Finally, we conclude with a general analysis of Unruh thermalization, starting from equal-fidelity non-thermal states, and demonstrate that the detectors' fidelity and "speed" of quantum evolution still exhibit a Mpemba-like behavior.

hep-th↗
arXiv2026

Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies

Goal-conditioned policies enable decision-making models to execute diverse behaviors based on specified goals, yet their downstream performance is often highly sensitive to the choice of instructions or prompts. To bypass the limitations of discrete text prompts, we formulate post-training adaptation as a latent control problem, where the goal embedding serves as a continuous control variable to modulate the behavior of a frozen policy. We propose Preference Goal Tuning (PGT), a framework that optimizes this latent control variable to align the induced trajectory distribution with task preferences. Unlike standard fine-tuning that updates policy parameters, PGT keeps the policy frozen and updates only the latent goal using a trajectory-level preference objective. This approach essentially searches for the optimal conditioning input that maximizes the likelihood of preferred behaviors while suppressing undesirable ones. We evaluate PGT on the Minecraft SkillForge benchmark across 17 tasks. With minimal data, PGT achieves average relative improvements of 72.0\% and 81.6\% on two foundation policies, consistently outperforming expert-crafted prompts. Crucially, by decoupling task alignment (latent goal) from physical dynamics (frozen policy), PGT surpasses full fine-tuning by 13.4\% in out-of-distribution settings, demonstrating superior robustness and generalization.

cs.AI↗
arXiv2026

Knowledge-Graph-Driven Data Synthesis for Low-Resource Software Development: A HarmonyOS Case Study

In low-resource framework development (e.g., HarmonyOS), large language models (LLMs) often lack sufficient pre-training exposure, resulting in poor code generation performance. Although they generally preserve programming logic across languages, they frequently fail on framework-specific APIs and syntax, revealing a gap between learned algorithmic knowledge and unfamiliar framework conventions. Consequently, even advanced models such as GPT-4o struggle to produce correct code without prior exposure. Inspired by these challenges, we propose APIKG4Syn, a framework that leverages API knowledge graphs to synthesize API-oriented question-code pairs without requiring executable environments. It incorporates both single-API and multi-API information, with the latter guided by uncertainty estimation (UE) and Monte Carlo Tree Search (MCTS), to construct high-quality fine-tuning data. For evaluation, we select HarmonyOS as a case study due to its accessible documentation and growing ecosystem, and build the first benchmark for its code generation. Experimental results show that fine-tuning Qwen2.5-Coder-7B with APIKG4Syn achieves a pass@1 of 25.00%, outperforming untuned GPT-4o (17.59%). We further observe that larger volumes of data generated by APIKG4Syn consistently lead to better fine-tuning performance, and that the optimal Single-API to Multi-API ratio is 8:2. Ablation studies also confirm the necessity and effectiveness of each component in our framework. These findings highlight the effectiveness of API-oriented data in enhancing LLM performance for low-resource software development scenarios.

cs.SE↗
arXiv2026

Probing excited-state dynamics of transmon ionization

The fidelity and quantum nondemolition character of the dispersive readout in circuit QED are limited by unwanted transitions to highly excited states at specific photon numbers in the readout resonator. This observation can be explained by multiphoton resonances between computational states and highly excited states in strongly driven nonlinear systems, analogous to multiphoton ionization in atoms and molecules. In this work, we utilize the multilevel nature of high-$E_J/E_C$ transmons to probe the excited-state dynamics induced by strong drives during readout. With up to 10 resolvable states, we quantify the critical photon number of ionization, the resulting state after ionization, and the fraction of the population transferred to highly excited states. Moreover, using pulse-shaping to control the photon number in the readout resonator in the high-power regime, we tune the adiabaticity of the transition and verify that transmon ionization is a Landau-Zener-type transition. We further extend these methods to a typical transmon with $E_J/E_C \approx 55$ and probe the offset-charge dependence of ionization dynamics in a timed-resolved manner. Our experimental results agree well with the theoretical prediction from a semiclassical driven transmon model and may guide future exploration of strongly driven nonlinear oscillators.

quant-ph↗