Search arXiv⌕ Search

arXiv subjects

Hao Fang

Publications and source records attributed to Hao Fang.

At least 19 recordsLinked to original sources

Plug-and-Play Verifier for Decoupling Process Reward in Stabilizing Long-Horizon Agentic Reasoning

Stabilizing long-horizon reasoning requires intelligent agents to sustain useful progress across successive reasoning steps and tool interactions. Such agents are typically trained with reinforcement learning (RL), where actor-critic RL uses a learnable critic network to estimate step advantages, while critic-free methods such as GRPO eliminate critic models and rely mostly on the group variants of sparse outcome-only rewards. Both reward types leave partially useful failed trajectories either implicit or indistinguishable from entirely unproductive ones, and as reasoning horizons grow, it becomes difficult to maintain consistent reasoning trajectories. In this work, we propose a \textit{plug-and-play verifier} framework for stabilizing long-horizon reasoning through explicit verification of intermediate reasoning artifacts. Specifically, the framework decouples process verification through a unified interface that extracts intermediate reasoning artifacts from agent running trajectories, evaluates them using domain-specific verifiers, and aggregates the verification scores as bounded process reward. Then, we combine our dense verified process rewards and the sparse outcome rewards through a two-stage post-training schedule. We showed that with plug-and-play verifiers, we further stabilize the reasoning steps as last-mile self-reflection and self-consistency. Typical evaluation on HotpotQA using Qwen3.5-4B achieves a superior accuracy (60.18\%), outperforming the standard GRPO baseline (57.73\%). Our results suggest that plug-and-play verifiers provide a modular approach to learn more stable and reliable reasoning behaviors over long horizons.

cs.LG↗

Seeing Through the Chain: Understanding and Mitigating Hallucinations in Multimodal Large Reasoning Models

While multimodal reasoning models (MLRMs) have exhibited impressive capabilities, they remain prone to hallucinations, and effective solutions are still underexplored. In this paper, we experimentally analyze the hallucination cause and propose C3PO, a training-based mitigation framework comprising \textbf{C}hain-of-Thought \textbf{C}ompression and \textbf{C}ontrastive \textbf{P}reference \textbf{O}ptimization. Firstly, we identify that introducing reasoning mechanisms exacerbates models' reliance on language priors while overlooking visual inputs, which can produce CoTs with reduced visual cues but redundant text tokens. To this end, we propose to selectively filter redundant thinking tokens for a more compact and signal-efficient CoT representation that preserves task-relevant information while suppressing noise. In addition, we observe that the quality of the reasoning trace largely determines whether hallucination emerges in subsequent responses. To leverage this insight, we introduce a reasoning-enhanced preference tuning scheme that constructs training pairs using high-quality AI feedback. We further design a multimodal hallucination-inducing mechanism that elicits models' inherent hallucination patterns via carefully crafted inducers, yielding informative negative signals for contrastive correction. We provide theoretical justification for the effectiveness and demonstrate consistent hallucination reduction across diverse MLRMs and benchmarks.

cs.CV↗

GaussTrap: Stealthy Backdoor Attacks on 3D Gaussian Splatting for Targeted Scene Misperception

3D Gaussian Splatting (3DGS) has recently emerged as a powerful paradigm for real-time scene representation and novel view synthesis, gaining traction in safety-critical applications such as autonomous driving and AR/VR systems. However, the security vulnerabilities of 3DGS remain largely unexplored. In this paper, we conduct a comprehensive study of backdoor threats in 3DGS pipelines. We uncover an inherent security vulnerability in the adaptive density control mechanism of 3DGS, originally designed for detailed reconstruction, which enables attackers to induce localized gaussian proliferation, effectively embedding malicious backdoors into the scene geometry while remaining invisible from other viewpoints. To expose this threat, we propose GaussTrap, a novel stealthy poisoning framework tailored for 3DGS. GaussTrap operates in three stages (adversarial injection, geometric stabilization, and global fidelity restoration) to embed imperceptible yet adversarial Gaussian components into the reconstruction. It ensures consistent poisoning effects at specific viewpoints while preserving photorealism in benign views, thereby achieving high attack efficacy with low detectability. Extensive experiments on both synthetic and real-world benchmarks demonstrate that GaussTrap reliably induces targeted visual corruption under trigger conditions while maintaining high rendering fidelity elsewhere, highlighting a critical security blind spot in modern 3D rendering systems. The code is available at https://github.com/acang425/GaussTrap.

cs.CV↗

ATLAS: Expert-Guided Sustainable Satellite Backhaul for Off-Grid Wildlife Monitoring

Remote wildlife monitoring increasingly depends on high-fidelity cameras and sensors deployed far from grid power and terrestrial networks. Low Earth orbit satellite links can provide connectivity in these settings, but their high power draw and variable bandwidth, combined with intermittent on-site solar harvesting, make continuous backhaul infeasible. Current systems therefore filter at the edge, with motion triggers that waste limited energy and bandwidth on environmental artifacts, or fixed vision models that cannot follow evolving expert objectives. We present \textsc{ATLAS}, an expert-guided edge-to-cloud system that treats satellite backhaul as a scarce resource for off-grid wildlife monitoring. \textsc{ATLAS} uses foundation models to compile natural-language expert intent into semantic scoring logic, combines semantic relevance with environmental and energy state, and uses a constrained reinforcement-learning scheduler to decide which clips to upload and when. As a representative case study, we deployed \textsc{ATLAS} for over six months across 18 off-grid salmon-monitoring river sites. \textsc{ATLAS} achieves a mean clip-selection accuracy of $87.4\%$, exceeding mean baseline accuracy by $26.4$ percentage points, while maintaining near-continuous operation through low-solar periods in which baselines incur 16.0\% downtime.

eess.SY↗

Ideal Gårding polynomials

We introduce ideal Gårding polynomials, a convexity-enhanced subclass of Gårding polynomials whose Gårding components are recursively convex under partial differentiation. This class strictly contains real stable polynomials and, after translation and homogenization, lies in the Lorentzian class. Our main result is that ideal Gårding polynomials still admit a robust structure theory despite this additional convexity: they are preserved under polarization, satisfy natural closure properties, and support a linear preserver theory. A key contribution of this paper is a universal model for univariate Gårding polynomials, described by monotone root sequences and equivalently by volume polynomials of Pitman--Stanley polytopes. We establish quotient concavity, and Newton--Maclaurin type inequalities, which leads to the polarization theorem, and suggests further connections with convex geometry and Lorentzian polynomials.

math.CO↗

Spectral Gårding Elliptic Equations

We study spectral Gårding elliptic equations on compact Kähler manifolds. Smooth solvability is equivalent to the existence of a smooth representative in the admissible domain. The concavity theorem for ideal Gårding polynomials gives a fixed concave formulation to which Székelyhidi's second-order argument applies in every degree $2\le d\le n$, extending Lin's top-degree solvability theorem. Combined with the numerical criteria of Fu--Li--Zhang--Zhang and Chen--Nie--Xu, this yields a Nakai--Moishezon type criterion for solvability on compact Kähler manifolds. A solution converts the local ideal Gårding structure into a global Lorentzian structure. Its mixed derivatives give a strict nonlinear Hodge--Riemann hierarchy from the degree of the equation down to degree two, and the homogenization of the intersection polynomial is Lorentzian in nef class directions. Numerical criteria then produce explicit Hodge--Riemann classes, Lefschetz isomorphisms in bidegree $(1,1)$, and weighted Hodge-index inequalities.

math.AP↗

When History Helps and Hurts: Selective History Use across Multimodal Turns

Reliable multimodal interaction depends on selective use of conversational history: an earlier question may remain relevant while its previous answer is outdated, whereas a current request may depend on historical evidence despite conflicting new observations. Existing multi-turn evaluations rarely separate these history-use demands from underlying question difficulty. To address this gap, we introduce ReTurn, a benchmark of 7,000 base tasks spanning visual and audio evidence for evaluating selective history use. For task-carrying history, Reconfirm/Reground require applying a historical question to current media while varying historical agreement; for evidence-carrying history, Retrieve/Rebind require answering a current question using historical media while varying current-media competition. Each pair preserves the target question, media, and answer. Tasks support open-ended and multiple-choice evaluation, with matched single-turn counterparts serving as answerability references. Across 13 omni-modal, vision-language, and audio-language models, median model-level open-ended accuracy falls from 93.7% with direct input to 72.3% in conversation. Behavioral probes show that high question recall can coexist with weaker task application, while competing media can redirect answers away from historical targets. Supervised adaptation yields only partial gains. ReTurn provides a controlled framework for assessing whether multimodal models select and use the historical information required by each request.

cs.CL↗

Improving Reasoning Ability via Asynchronous On-Policy Self-Distillation under Positive Rollouts

Distillation and reinforcement learning through verifiable rewards (RLVR) have achieved progress in enhancing the reasoning ability of large language models (LLMs). However, we note that negative rollouts may admit no gradation of failure severity, and the combinatorial vastness makes penalizing a few sampled negatives unlikely to cover a meaningful reward signal under sparse binary rewards. In this work, we propose Positive-Only Policy Optimization (POPO), an on-policy self-distillation integrated RLVR framework in which learning occurs exclusively on online positive rollouts. Specifically, POPO utilizes bounded importance sampling over the positive rollout set. Thus, no disjoint negative rollouts are used for gradient guidance during post-training. We show that implicit negative gradients can emerge naturally through reinforcing the positive probability via rollout redistribution. Next, POPO stabilizes the policy optimization through self-distillation. First, it applies a Siamese policy network with a momentum-based adaptation law for asynchronous policy evolution. Second, we replace the KL-divergence with a bounded similarity penalty term in the Siamese representation space. We conduct extensive experiments using publicly available, well-established text-LLM models across all-level mathematical benchmarks (MATH-500, AMC23, AIME 2024/2025, and Olympiad). Our experiment demonstrates that POPO achieves superior performance compared to GRPO. Notably, we show that POPO can achieve 36.67% in AIME 2025 with Qwen-Math-7B, outperforming GRPO 30.00%. Our ablation and sweep studies further illustrate the necessity and robustness.

cs.CL↗

CHILLGuard: Towards Fine-Grained Chinese LLM Safety Guardrail with Scalable Data Construction and Model-aware Preference Alignment

Malicious content generated from large language models (LLMs) could pose severe safety risks and ethical concerns. While existing LLM safety guardrails excel in English or multilingual settings, they lack adaptation to Chinese-specific regulatory policies, cultural context, and linguistic nuances, failing to support fine-grained risk classification for diverse deployment needs. In this paper, we introduce a 5-macro, 31-micro category fine-grained risk taxonomy for Chinese scenarios, and build CHILLGuard: a dedicated Chinese LLM content safety guardrail. To address the critical scarcity of high-quality annotated Chinese safety data, we propose a scalable multi-stage data construction pipeline: we expand multi-source corpus via retrieval-augmented generation, generate implicit harmful samples through prompt engineering rewriting, and refine high-quality data via multi-model voting-based label calibration. Based on this, we build CHILLGuardTrain, a large-scale training set with 405,007 samples, and CHILLGuardTest, a rigorously curated annotated test set with 51,745 samples. We then train CHILLGuard on CHILLGuardTrain under a generator-classifier collaborative framework via Model-aware Direct Preference Optimization. Extensive experiments under multiple settings demonstrate the state-of-the-art performance of CHILLGuard, e.g., a 15.92% relative improvement of F1 score over Qwen3Guard-8B-Strict on our benchmark. We release our resources at https://github.com/cswbyu/CHILLGuard.

cs.CL↗

Gårding Polynomials

We introduce Gårding polynomials, a class of real multivariate polynomials characterized by positivity regions that are invariant under translation by positive vectors and closed under strictly positive affine transformations. We prove that this geometric formulation is equivalent both to a reduction to the multi-affine setting via polarization and to a recursive criterion in terms of partial derivatives. The class of Gårding polynomials strictly extends that of real stable polynomials while preserving many of their structural properties. In particular, multi-affine Gårding polynomials with nonnegative coefficients satisfy the Rayleigh property, and their positive univariate specializations have ultra log-concave coefficient sequences. The Gårding property for several matroid generating functions is preserved under natural matroid operations. As applications, we derive new negative dependence results for generating functions associated with various classes of matroids and graphs, including examples previously beyond the scope of real stability and Lorentzian methods. We further obtain analogous results for characteristic polynomials arising from certain matrix classes.

math.CO↗

Harmony in Diversity: Multi-domain Contrastive Policy Optimization for Large Reasoning Models

Post-training via Reinforcement Learning (RL) has enabled Large Reasoning Models (LRMs) to achieve strong performance in individual domain. However, real-world applications increasingly require general-purpose reasoners rendering strong performance across diverse domains. Mixed-domain post-training aims to achieve this goal by jointly training with mixed domain data, but this often induces capability compromise and degradation among different domains. Existing methods attribute this performance degradation to harmful cross-domain interactions and propose various strategies to mitigate them, but these strategies may also impede beneficial knowledge sharing across domains and in turn fail to match or surpass single-domain performance. To address this problem, we propose \textbf{M}ulti-domain \textbf{C}ontrastive \textbf{P}olicy \textbf{O}ptimization (MCPO), which uses contrastive learning to utilize both positive and negative cross-domain interactions for knowledge sharing and competition. Specifically, we partition each rollout generated by LRMs according to its underlying reasoning structures and use reasoning segments to capture these structures. We thus formulate positive and negative pairs of reasoning segments as mutually augmented examples, which provide supportive and competing signals for knowledge sharing. Subsequently, we design complementary contrastive objectives for cross-domain knowledge sharing and intra-domain knowledge consolidation, targeting compatibility across domains and discriminability within each domain to form a harmonious reasoning space. Experimental results across a broad range of domains show that MCPO alleviates performance degradation caused by mixed-domain training and outperforms single-domain training in most cases.

cs.CL↗

MergeHEIR: Mitigating Multimodal Hallucinations as the Tax of Model Merging

Model merging consolidates task-specialized experts into a single deployable model. However, we show that such capability consolidation incurs a merging tax of increased hallucination: across 8 model-merging methods, every merged checkpoint exhibits a higher hallucination rate than the average of its constituent experts. An intuitive approach is to adapt existing hallucination-mitigation methods to the post-merge model, yet this unconstrained adaptation disrupts inherited capabilities, creating a tension between hallucination mitigation and expertise retention. To tackle this challenge, we introduce MergeHEIR, a post-merge adaptation framework designed to reduce this merging tax while preserving expertise inherited from initial experts. Using small expert-task calibration sets, MergeHEIR constructs layer-wise null-space projectors via SVD from task-specific activations collected from the merged checkpoint, and periodically projects the accumulated post-merge displacement onto the resulting null spaces to preserve inherited expertise. Theoretically, we establish minimum-distortion and maximum-dimensionality guarantees, characterize the threshold-controlled adaptation-retention trade-off, and extend perturbation guarantees beyond finite calibration data. Across 24 paired comparisons spanning three MLLM configurations and 8 model-merging methods, MergeHEIR consistently mitigates hallucination while largely preserving inherited expertise, demonstrating a more favorable hallucination-retention trade-off.

cs.AI↗

The $J$-Null Locus and Local Regularity of Weak Solutions of the Semistable $J$-Equation

For the semistable $J$-equation, we prove that the numerical $J$-null locus coincides with the ambient $C^2$-singular locus of Murakami's weak solution. Consequently, this singular locus is a proper analytic subset with finitely many irreducible components, and the weak solution is locally smooth outside the $J$-null locus. The proof relies on two analytic ingredients. First, we establish a regularization theorem for singular $J$-subsolutions, showing that Demailly's global regularization preserves quantitative strict cone conditions. Thus, singular strict subsolutions with prescribed analytic poles can be replaced by smooth strict subsolutions away from their pole sets. Second, we derive relative a priori estimates adapted to these subsolutions: a relative $L^\infty$-estimate from determinant control and a weighted second order estimate yielding uniform $C^2$-bounds on compact subsets of the regular locus.

math.DG↗

Determinantal capacity and $L^\infty$-Estimates

We introduce determinantal ellipticity, or det-ellipticity, a quantitative structural condition for fully nonlinear elliptic operators on compact Kähler manifolds that provides the link between the $L^\infty$-estimates and algebraic/combinatorial properties of a large class of Hessian elliptic operators. On the analytic side, we introduce $\mathcal D$-subsolutions extending the determinant sublevel condition of Sui--Sun. Using the auxiliary comparison method of Guo--Phong--Tong and pluripotential theory for complex Monge--Ampère equations, we obtain relative $L^\infty$-estimates in big cohomology classes under a determinant-entropy bound for viscosity supersolutions and singular reference potentials. Det-ellipticity provides a systematic construction of such subsolutions. On the algebraic side, we study Gårding elliptic polynomial operators of degree $d$. We characterize the determinant increment bound with exponent $d/n$ by positivity of the determinantal capacity of the top homogeneous part. This is equivalent to balanced-point conditions for the Newton polytopes of rank-one scalarizations and to slope semistability of an associated subspace polymatroid. Together with the analytic hypotheses, these criteria yield relative $L^\infty$ estimates for the corresponding fully nonlinear equations.

math.AP↗

From Bits to Beliefs: Recoverable Semantic Fingerprints for Black-Box Verification of Large Language Models

Open-weight large language models (LLMs) can be copied, modified, and redeployed behind black-box APIs, making post-release ownership verification difficult. Existing black-box fingerprints often rely on secret query-key pairs that reproduce predefined responses, and can therefore be easily disrupted by fine-tuning, pruning, quantization, model merging, and serving-time prompt changes. We propose SimPrint, a recoverable semantic fingerprinting framework for black-box LLM ownership verification. Rather than relying on isolated exact matches, SimPrint encodes a private owner signature into a coded semantic fingerprint domain, distributing ownership evidence across natural binary question-answering probes. It implants only base-deviating probes through a low-interference batch update that preserves the original model behavior, and later recovers the signature by parsing suspect-model responses into reliable bits or erasures with an error-correcting recovery mechanism. Because verification only uses input-output queries, SimPrint remains applicable when model weights or activations are inaccessible. Experiments on three open-weight LLMs show that SimPrint reliably recovers the owner signature in both clean and modified settings, remains robust under fine-tuning, pruning, quantization, model merging, and serving-time perturbations, and maintains comparable downstream utility.

cs.CR↗

Privacy Leakage on DNNs: A Survey of Model Inversion Attacks and Defenses

Deep Neural Networks (DNNs) have revolutionized various domains with their exceptional performance across numerous applications. However, Model Inversion (MI) attacks, which disclose private information about the training dataset by abusing access to the trained models, have emerged as a formidable privacy threat. Given a trained network, these attacks enable adversaries to reconstruct high-fidelity data that closely aligns with the private training samples, posing significant privacy concerns. Despite the rapid advances in the field, we lack a comprehensive and systematic overview of existing MI attacks and defenses. To fill this gap, this paper thoroughly investigates this realm and presents a holistic survey. Firstly, our work briefly reviews early MI studies on traditional machine learning scenarios. We then elaborately analyze and compare numerous recent attacks and defenses on Deep Neural Networks (DNNs) across multiple modalities and learning tasks. By meticulously analyzing their distinctive features, we summarize and classify these methods into different categories and provide a novel taxonomy. Finally, this paper discusses promising research directions and presents potential solutions to open issues. To facilitate further study on MI attacks and defenses, we have implemented an open-source model inversion toolbox on GitHub (https://github.com/ffhibnese/Model-Inversion-Attack-ToolBox).

cs.CV↗

Looking Back and Forward: What Teaching Materials Do Not Remember About Instructional Reasoning

Instructional materials can be passed along to new instructors, yet the instructional reasoning, such as instructional intentions, contextual information, and reflective insights behind their design, is rarely preserved. Traditional Learning Management System (LMS) tools are designed to store instructional artifacts rather than reasoning. To address this issue, we propose Teaching Memory, a design approach that treats instructional reasoning as a primary form of knowledge. We present a design-oriented framework for preserving instructional reasoning in a lightweight manner within instructors' existing workflows, enabling continuity and reuse.

cs.CY↗

Min-Max Regret Task Allocation and Planning of Heterogeneous Multi-Robot System in Partially Known Environments

Efficient task allocation for large-scale Heterogeneous Multi-Robot Systems (HMRS) is critical, yet dealing with complex temporal logic tasks in partially known environment (PKE) remains a computational bottleneck. Existing approaches often struggle to balance exploring uncertain regions and exploiting known resources, while also suffering from exponential computational complexity. To address these issues, this paper presents a robust planning framework that simultaneously handles high-level logical constraints and environmental uncertainty without sacrificing scalability. We formulate the problem as a min-max regret optimization, proposing a Region-Binding Atomic Proposition (RbAP) to capture resource uncertainty within the automaton structure. To solve this, we propose the Extended Planning Decision Tree (E-PDT) equipped with a novel Regret-based Branch-and-Bound (BnB) strategy. Unlike traditional methods that rely on prior probabilities or worst-case analysis, our approach dynamically prunes suboptimal policies, effectively balancing the need for information gathering (exploration) and task completion (exploitation). Theoretical analysis confirms the feasibility and completeness of our approach. Extensive numerical and physical experiments demonstrate that the proposed framework achieves near-linear scalability with respect to the number of robots and types, significantly outperforming MILP-based baselines in both solution quality and computational efficiency.

cs.RO↗