Searcharxiv⌕ Search

arXiv subjects

Hongwei Zheng

Publications and source records attributed to Hongwei Zheng.

At least 19 recordsLinked to original sources
arXiv2026

Topology-induced Operators Reveal Complementary Graph Representations without Training

Graph representation learning has largely focused on designing increasingly sophisticated models to transform graph topology into vector representations, or embeddings. However, the extent to which embedding quality depends on model learning, rather than on the underlying topological transformations, remains unclear. Here, we show that informative embeddings can be derived without complicated model design and gradient-based training. Propagating random features through implicit hierarchical structures induced by random walks and anonymous walks yields embeddings that capture node proximity and structural role, respectively. These two training-free embeddings preserve complementary aspects of graph organization and perform competitively with classic and recent methods across various node-, edge-, and graph-level tasks. They often require substantially less computation, resulting in a favorable quality-efficiency trade-off. Combining the two types of embeddings further improves inference quality of some tasks compared with using either embedding type alone. Our results suggest that informative graph embeddings can arise from carefully chosen topological transformations before any learning operation is applied.

cs.LG↗
arXiv2026

Control of Harmful Information Spreading on Adaptive Higher-Order Networks via Group Dissolution

Curbing harmful information contagion remains a critical challenge, motivating platform-level interventions such as group dissolution to sever transmission chains. However, in practice, users affected by dissolution often exhibit adaptive behavior, rewiring to form new groups. Yet, it remains unclear how these two mechanisms jointly shape information contagion and whether group dissolution remains effective in suppressing it. Here, we develop an adaptive higher-order contagion model that integrates platform-induced group dissolution with user adaptive rewiring, and derive a theoretical framework. Notably, we reveal an effective window for group dissolution, bounded by a critical infection rate. Above this threshold, dissolution backfires and amplifies information prevalence. Within this window, dissolution acts non-monotonically, initially exacerbating prevalence before eradicating contagion via a discontinuous transition beyond a critical dissolution rate. We further show that higher-order reinforcement expands this infection-rate window over which dissolution remains effective, whereas rewiring homophily substantially narrows it. Simulations on empirical hypergraph also validate these findings. Our work highlights the interplay between top-down platform interventions and bottom-up user adaptation, underscoring the need to account for adaptive responses when designing strategies to curb harmful information without unintended amplification.

physics.soc-ph↗
arXiv2026

Toward Annotation-Efficient Continuous Emotion Arousal Quantification via Group-Level EEG Dynamic Neural Synchrony

Continuous emotional arousal quantification remains bottlenecked by time-consuming and labor-intensive manual annotation. This work investigates group-level EEG dynamic neural synchrony (DNS) as a principled signal for continuous arousal quantification that bypasses per-subject manual labeling. Using Correlated Component Analysis (CorrCA) with sliding-window computation across four EEG datasets spanning 142 subjects and over 207 hours, we systematically evaluate DNS as a group-level marker for emotional arousal dynamics. Three key findings emerge. First, DNS exhibits significant emotion information from valence-dependent differences (all p<0.003), with positive emotions eliciting higher synchrony. Second, DNS correlates more strongly with the first-order derivative of arousal than with raw arousal values, revealing that neural synchrony captures the rate of emotional change rather than static intensity. Third, we provide the first systematic characterization of how DNS-arousal coupling depends on key methodological choices, finding that moderate windows (10-30 s), positive lags (0-10 steps), and First-order Difference feature of EEG from the dominant CorrCA component yield consistently strong coupling. Subject-split replication and block permutation tests confirm these associations are not statistical artifacts. Our findings establish DNS as an empirically validated group-level marker toward annotation-efficient continuous emotional arousal quantification.

cs.HC↗
arXiv2026

Immunization on Temporal Higher-Order Networks

Network immunization is a powerful tool for controlling contagion processes ranging from infectious diseases to misinformation diffusion. While prior works have focused on pairwise or static networks, immunization dynamics in temporal higher-order networks remain poorly understood. Here, we introduce immunization strategies and develop a theoretical framework tailored for such temporal systems. Firstly, we reveal bistability and discontinuous transitions in prevalence as the immunization fraction varies. This implies that immunization effectiveness depends on the initial prevalence, marking a fundamental departure from pairwise networks. Building on this prevalence-dependent behavior, we propose the High Infection Contribution (HIC) strategy, demonstrating its superior performance over all evaluated heuristic strategies. Furthermore, we introduce egocentric strategies by leveraging solely local observations. Notably, the optimal egocentric strategy shifts with the contagion prevalence. Our work advances the understanding of network immunization, paving the way for effective contagion control in temporal higher-order networks.

physics.soc-ph↗
arXiv2026

Identifying vulnerable nodes for hypergraph dismantling via higher-order competition dynamics

Network dismantling aims to identify a node removal sequence that can rapidly destroy network connectivity, which is an important problem for understanding the structural fragility of complex systems and designing intervention strategies. Existing studies mainly focus on pairwise networks or assume weak-deletion rules where node removal only causes hyperedges to shrink in higher-order networks. However, in many real higher-order systems, the failure of one participant may cause the entire group interaction to fail, i.e., the strong-deletion mechanism. Such a mechanism cannot be fully captured by projected networks or methods based on weak-deletion rules. To address this challenge, we propose hyper-Vulnerability-weighted Dominance rank (hyper-VDrank), a higher-order centrality method for hypergraph dismantling under strong deletion. Hyper-VDrank constructs a higher-order competition dynamics mediated by hyperedge-induced environments, where a node does not compete only with individual neighbors but responds to the collective pressure formed by other nodes in the same hyperedge. It further introduces a hyperedge vulnerability weight based on redundancy and size effects to capture the vulnerable structures, facilitating the distinction of critical nodes. Experiments show that hyper-VDrank reduces the largest connected component more rapidly, collapses the hypergraph earlier, and produces greater structural fragmentation than classical and recent methods. On 14 real-world hypergraphs, hyper-VDrank improves dismantling efficiency by 23.65% and reduces the collapse threshold by 27.63% on average compared with the baselines. In summary, hyper-VDrank offers an effective hypergraph dismantling tool and a new perspective on identifying vulnerable structures in higher-order complex systems.

physics.soc-ph↗
arXiv2026

Emergent dilemma and periodic oscillation in the nonlinear interplay between epidemic and behavior

Human behaviors, particularly non-pharmaceutical interventions (NPIs), are dynamically coupled with epidemic spreading. While prior studies mainly assume a linear interplay, real-world behavioral evolution is driven by nonlinear responses and social influence. Here, we incorporate these multifaceted mechanisms into a co-evolutionary model and analytically derive the critical thresholds. Notably, as the infection rate grows, NPI compliance initially rises but then abruptly drops to zero. This paradoxical decline indicates an emergent social dilemma: at high infection rates, abandoning NPIs is individually optimal but detrimentally triggers an explosive surge in epidemic prevalence. We further show that socially induced overestimation of the infection rate can counterintuitively prompt individuals to abandon NPIs. Moreover, the interplay with social influence induces periodic oscillations, reflecting a tragic cycle of recurrent epidemic waves. Furthermore, we validate the robustness of this NPI-abandonment dilemma in networked population. Our work illustrates rich emergent phenomena in the co-evolution of epidemic and behavior, challenging traditional views on this coupled dynamics.

physics.soc-ph↗
arXiv2026

PACT: Learning Diverse Diagnostic Strategies via Privileged Synthesis and Branch Consensus

Clinical diagnosis requires flexible use of multiple reasoning paradigms under incomplete patient information. Existing LLM-based medical agents show strong medical reasoning ability, but single-paradigm or naively mixed dialogue supervision makes these paradigms difficult to learn without interference. We propose \textbf{PACT} (Periodic Anchor Consensus Training), a framework that couples supervised multi-paradigm dialogue synthesis with consensus-based Branch training. At the data level, \textbf{DPS} (Doctor-Patient-Supervisor) uses complete electronic medical records (EMRs) for quality control while keeping the doctor agent restricted to patient-visible information. This produces validated dialogues under four diagnostic reasoning paradigms without leaking hidden clinical answers. At the training level, PACT trains one paradigm-specific LoRA Branch per paradigm and periodically aggregates Branches into a shared Anchor through sign consensus. We further construct a dynamic multi-turn Chinese medical diagnosis benchmark for interactive consultation. Experiments show that PACT achieves state-of-the-art performance among compared proprietary, medical-specialized, and task-adapted baselines on diagnostic outcome and consultation-process metrics.

cs.CL↗
arXiv2026

Stereotyping by strategy standing diversifies cooperation patterns in indirect reciprocity

Indirect reciprocity explains how cooperation evolves through social reputations. People observe others, assign reputations, and condition their future actions on these assignments. This process is cognitively demanding, and stereotyping offers a simpler alternative by replacing individual-level reputation with group-level information. Theoretical models commonly implement stereotyping through exogenously given group labels. In real societies, however, group-level impressions may be associated with observable patterns of behavior. Here we propose a framework of stereotyping by strategy standing, in which mutants may condition their actions on the overall reputation level associated with a resident strategy rather than on the recipient's reputation. We show that this form of stereotyping can diversify stable cooperation in indirect reciprocity. As the strength of stereotyping increases, additional cooperative evolutionarily stable norm-strategy (ESS) pairs emerge in substantial numbers. In particular, we identify eight highly cooperative ESS pairs that become stable under very weak stereotyping. These pairs, which we call the counterparts of the leading eight, share the same social norms as the classical leading eight and differ only in how they prescribe behavior between bad individuals. They are unstable without stereotyping because they can be invaded by their corresponding leading strategies, but they become stable once stereotyping exceeds a critical threshold. Our results suggest that group-level impressions based on strategy standing can provide a coarse-grained informational route to stable cooperation and offer a more behaviorally grounded perspective on how stereotyping affects indirect reciprocity.

physics.soc-ph↗
arXiv2026

Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models

Vision-Language Models (VLMs) have broad potential in privacy-sensitive domains such as healthcare and finance, yet strict data-sharing constraints render centralized training infeasible. Federated Learning mitigates this issue by enabling decentralized training, but practical deployments face challenges due to client heterogeneity in computational resources, application requirements, and model architectures. Under extreme model and data heterogeneity, replacing parameter aggregation with preference-based collaboration offers a more suitable interface, as it eliminates the need for direct parameter or data exchange. Motivated by this, we propose MoR, a federated alignment framework that combines GRPO with Mixture-of-Rewards for heterogeneous VLMs. In MoR, each client locally trains a reward model from local preference annotations, capturing specific evaluation signals without exposing raw data. To combine these heterogeneous supervision signals, MoR introduces a Mixture-of-Rewards mechanism with learned routing, which adaptively fuses client reward models according to the input and alignment objective. The server then optimizes a base VLM using GRPO with a KL penalty to a reference model, enabling preference alignment without requiring client models to share architectures or parameters. Experiments on diverse public vision-language benchmarks demonstrate that MoR consistently outperforms federated alignment baselines in generalization and cross-client adaptability. Our approach provides a scalable solution for privacy-preserving alignment of heterogeneous VLMs under federated settings.

cs.AI↗
arXiv2026

Indirect reciprocity beyond pairwise interactions

Cooperation in groups underpins collective responses to challenges from climate governance to public goods provision, yet how moral evaluation sustains it remains poorly understood. Indirect reciprocity -- cooperating to build a good reputation -- is well characterized for pairwise interactions, but real collective action requires individuals to be judged against the reputational profile of an entire group. Here we develop a general framework for multiplayer indirect reciprocity and show that stable group cooperation obeys a simple organizing principle: `all good, help; one bad, halt'. This rule is both necessary and sufficient for cooperation to emerge, and it recovers the classical leading eight norms in the pairwise limit. We further show that group structure fundamentally changes reputation dynamics: unlike pairwise models, which are monostable, multiplayer systems exhibit bistability and hysteresis, with a critical tipping point separating cooperative and defective regimes. Assessment of the latent norms of large language models reveals that they shift toward punitive defection when provided with richer social information, yet fail to follow the full logic of `all good, help; one bad, halt'. Our results establish a unifying principle for reputation-based cooperation in groups and provide a benchmark for evaluating cooperative alignment in artificial intelligence.

physics.soc-ph↗
arXiv2026

IceBreaker for Conversational Agents: Breaking the First-Message Barrier with Personalized Starters

Conversational agents, such as ChatGPT and Doubao, have become essential daily assistants for billions of users. To further enhance engagement, these systems are evolving from passive responders to proactive companions. However, existing efforts focus on activation within ongoing dialogues, while overlooking a key real-world bottleneck. In the conversation initiation stage, users may have a vague need but no explicit query intent, creating a first-message barrier where the conversation holds before it begins. To overcome this, we introduce Conversation Starter Generation: generating personalized starters to guide users into conversation. However, unlike in-conversation stages where immediate context guides the response, initiation must operate in a cold-start moment without explicit user intent. To pioneer in this direction, we present IceBreaker that frames human ice-breaking as a two-step handshake: (i) evoke resonance via Resonance-Aware Interest Distillation from session summaries to capture trigger interests, and (ii) stimulate interaction via Interaction-Oriented Starter Generation, optimized with personalized preference alignment and a self-reinforced loop to maximize engagement. Online A/B tests on one of the world's largest conversational agent products show that IceBreaker improves user active days by +0.184% and click-through rate by +9.425%, and has been deployed in production.

cs.CL↗
arXiv2026

Reinforcement learning with reputation-based adaptive exploration promotes cooperation

Reinforcement learning provides a framework for studying how individuals adjust their behavior through repeated interaction and feedback in social dilemmas. In Q-learning, exploration controls how often agents choose actions other than those favored by their current learned Q-values. Yet existing models usually treat the exploration rate as a constant parameter. In systems with social evaluation, however, trial-and-error behavior carries different costs and opportunities for agents with different reputations, making exploration dependent on social standing rather than uniform across agents. Herein, we develop a spatial prisoner's dilemma model in which Q-learning agents adapt their exploration rates according to local reputation differences, while reputation is updated through an asymmetric, state-dependent rule. Results show that adaptive exploration and asymmetric reputation updating each promote cooperation, but their combination produces a stronger reinforcing effect than either mechanism alone. Low-reputation agents explore more and can recover reputation through cooperation, while high-reputation agents explore less and avoid reputation losses caused by defection. This mechanism also reorganizes cooperation in space, producing a stable checkerboard-like coexistence at intermediate reputation concern. In addition, cooperation is most vulnerable at intermediate baseline exploration rates, whereas stronger asymmetric reputation updating mitigates this exploration-induced disruption. These results suggest that reputation can act not only as a record of past behavior, but also as a dynamic signal that regulates exploratory behavior during learning and thereby stabilizes cooperation.

physics.comp-ph↗
arXiv2026

HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders

Large Language Models (LLMs) are powerful and widely adopted, but their practical impact is limited by the well-known hallucination phenomenon. While recent hallucination detection methods have made notable progress, we find most of them overlook the dynamic nature and underlying mechanisms of it. To address this gap, we propose HalluSAE, a phase transition-inspired framework that models hallucination as a critical shift in the model's latent dynamics. By modeling the generation process as a trajectory through a potential energy landscape, HalluSAE identifies critical transition zones and attributes factual errors to specific high-energy sparse features. Our approach consists of three stages: (1) Potential Energy Empowered Phase Zone Localization via sparse autoencoders and a geometric potential energy metric; (2) Hallucination-related Sparse Feature Attribution using contrastive logit attribution; and (3) Probing-based Causal Hallucination Detection through linear probes on disentangled features. Extensive experiments on Gemma-2-9B demonstrate that HalluSAE achieves state-of-the-art hallucination detection performance.

cs.CL↗
arXiv2026

Z-Erase: Enabling Concept Erasure in Single-Stream Diffusion Transformers

Concept erasure serves as a vital safety mechanism for removing unwanted concepts from text-to-image (T2I) models. While extensively studied in U-Net and dual-stream architectures (e.g., Flux), this task remains under-explored in the recent emerging paradigm of single-stream diffusion transformers (e.g., Z-Image). In this new paradigm, text and image tokens are processed as a single unified sequence via shared parameters. Consequently, directly applying prior erasure methods typically leads to generation collapse. To bridge this gap, we introduce Z-Erase, the first concept erasure method tailored for single-stream T2I models. To guarantee stable image generation, Z-Erase first proposes a Stream Disentangled Concept Erasure Framework that decouples updates and enables existing methods on single-stream models. Subsequently, within this framework, we introduce Lagrangian-Guided Adaptive Erasure Modulation, a constrained algorithm that further balances the sensitive erasure-preservation trade-off. Moreover, we provide a rigorous convergence analysis proving that Z-Erase can converge to a Pareto stationary point. Experiments demonstrate that Z-Erase successfully overcomes the generation collapse issue, achieving state-of-the-art performance across a wide range of tasks.

cs.CV↗
arXiv2026

Lyapunov Probes for Hallucination Detection in Large Foundation Models

We address hallucination detection in Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) by framing the problem through the lens of dynamical systems stability theory. Rather than treating hallucination as a straightforward classification task, we conceptualize (M)LLMs as dynamical systems, where factual knowledge is represented by stable equilibrium points within the representation space. Our main insight is that hallucinations tend to arise at the boundaries of knowledge-transition regions separating stable and unstable zones. To capture this phenomenon, we propose Lyapunov Probes: lightweight networks trained with derivative-based stability constraints that enforce a monotonic decay in confidence under input perturbations. By performing systematic perturbation analysis and applying a two-stage training process, these probes reliably distinguish between stable factual regions and unstable, hallucination-prone regions. Experiments on diverse datasets and models demonstrate consistent improvements over existing baselines.

cs.CV↗
arXiv2026

Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models

VLMs have broad potential in privacy-sensitive domains such as healthcare and finance, yet strict data-sharing constraints render centralized training infeasible. FL mitigates this issue by enabling decentralized training, but practical deployments face challenges due to client heterogeneity in computational resources, application requirements, and model architectures. We argue that while replacing data with model parameters characterizes the present of FL, replacing parameters with preferences represents a more scalable and privacy-preserving future. Motivated by this perspective, we propose MoR, a federated alignment framework based on GRPO with Mixture-of-Rewards for heterogeneous VLMs. MoR initializes a visual foundation model as a KL-regularized reference, while each client locally trains a reward model from local preference annotations, capturing specific evaluation signals without exposing raw data. To reconcile heterogeneous rewards, we introduce a routing-based fusion mechanism that adaptively aggregates client reward signals. Finally, the server performs GRPO with this mixed reward to optimize the base VLM. Experiments on three public VQA benchmarks demonstrate that MoR consistently outperforms federated alignment baselines in generalization, robustness, and cross-client adaptability. Our approach provides a scalable solution for privacy-preserving alignment of heterogeneous VLMs under federated settings.

cs.AI↗
arXiv2026

Indirect Reciprocity with Environmental Feedback

Indirect reciprocity maintains cooperation in stranger societies by mapping individual behaviors onto reputation signals via social norms. Existing theoretical frameworks assume static environments with constant resources and fixed payoff structures. However, in real-world systems, individuals' strategic behaviors not only shape their reputation but also induce collective-level resource changes in ecological, economic, or other external environments, which in turn reshape the incentives governing future individual actions. To overcome this limitation, we establish a co-evolutionary framework that couples moral assessment, strategy updating, and environmental dynamics, allowing the payoff structure to dynamically adjust in response to the ecological consequences of collective actions. We find that this environmental feedback mechanism helps lower the threshold for the emergence of cooperation, enabling the system to spontaneously transition from a low-cooperation state to a stable high-cooperation regime, thereby reducing the dependence on specific initial conditions. Furthermore, while lenient norms demonstrate adaptability in static environments, norms with strict discrimination are shown to be crucial for curbing opportunism and maintaining evolutionary resilience in dynamic settings. Our results reveal the evolutionary dynamics of coupled systems involving reputation institutions and environmental constraints, offering a new theoretical perspective for understanding collective cooperation and social governance in complex environments.

physics.soc-ph↗
arXiv2026

Stable-RAG: Mitigating Retrieval-Permutation-Induced Hallucinations in Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has become a key paradigm for reducing factual hallucinations in Large Language Models (LLMs), yet little is known about how the order of retrieved documents affects model behavior. We empirically show that under a Top-5 retrieval setting with the gold document included, LLM answers vary substantially across permutations of the retrieved set, even when the gold document is fixed in the first position. This reveals a previously underexplored sensitivity to retrieval permutations. Although existing robust RAG methods focus primarily on enhancing LLM robustness to low-quality retrieval and mitigating positional bias to distribute attention fairly over long contexts, neither approach directly addresses permutation sensitivity. In this paper, we propose Stable-RAG, which exploits permutation sensitivity estimation to mitigate permutation-induced hallucinations. Stable-RAG runs the generator under multiple retrieval orders, clusters hidden states, and decodes from a cluster-center representation that captures the dominant reasoning pattern. It then uses these reasoning results to align hallucinated outputs toward the correct answer, encouraging the model to produce consistent and accurate predictions across document permutations. Experiments on three QA datasets show that Stable-RAG improves answer accuracy, reasoning consistency, and generalization across datasets, retrievers, and input lengths compared with strong baselines.

cs.CL↗