Search arXiv⌕ Search

arXiv subjects

Lei Feng

Publications and source records attributed to Lei Feng.

At least 37 records · Page 2Linked to original sources
arXiv2026

Primordial black hole clustering from spectator fields for interpreting the JWST observations

The observations by the James Webb Space Telescope (JWST) have revealed unexpectedly massive galaxy candidates at high redshifts, posing a significant challenge to the $Λ$CDM model. In this work, we investigate whether primordial black holes (PBHs) with spatial clustering, generated by a light spectator field during inflation, can accelerate early structure formation. We adopt the galaxy candidates with inferred stellar mass $10^9\,M_\odot\leq M_*^{\rm obs}\leq10^{11}\,M_\odot$ at redshift $7 \leq z \leq 10$ reported by the CEERS program as a benchmark. Two different mechanisms are considered, through which PBH clustering can influence structure formation: the PBH-induced isocurvature perturbations that enhance the matter power spectrum on linear scales, and the localized seed formation and accretion by compact PBH clusters on nonlinear scales. We find that, when adopting the cosmic microwave background (CMB) isocurvature constraint $β_{\rm iso}<0.035$ at the benchmark pivot scale $k_*=0.002\,{\rm Mpc}^{-1}$, PBH clustering can produce a cumulative stellar mass density consistent with the JWST observations while satisfying the relevant isocurvature constraint. However, the allowed enhancement of structure formation is strongly suppressed when the constraint at $k_*=0.1\,{\rm Mpc}^{-1}$ is imposed, indicating a significant dependence on the choice of the pivot scale. In contrast, the localized seed effect of compact PBH clusters is strongly constrained by the CMB isocurvature bounds, while isolated supermassive PBHs produce stellar mass densities far below those inferred from the JWST observations. Our results show that PBH clustering induced by a spectator field can substantially accelerate early structure formation, but whether it can fully account for the JWST-inferred stellar mass density depends sensitively on the pivot scale adopted for the CMB isocurvature constraint.

astro-ph.CO↗
arXiv2026

Are Human-Aligned Models Models of Humans? A Turing-Test Gap in Preference Alignment

Human-feedback alignment has made language models useful assistants and is commonly described as aligning them with humans. However, the responses people prefer from an AI need not be the responses they themselves would give. We distinguish alignment with human preferences from alignment with human behavior, and show that alignment with human preferences can make model behavior less human-like even when both preferences and responses come entirely from humans. We call this the Turing-test gap. We show that preference alignment preserves the human response distribution only under a restrictive condition, and find no consistent evidence that real human preferences satisfy it. Empirically, the loss of human-response likelihood increases with the strength of preference weighting, regardless of its direction, and the gap also appears under standard DPO. These results establish human-likeness as an explicit dimension of alignment rather than something assumed to follow from preference alignment.

cs.AI↗
arXiv2026

ALP-mediated inelastic dark matter and the LUX-ZEPLIN high-recoil candidate event LZ230616

The LUX-ZEPLIN (LZ) Collaboration has reported a high-energy candidate event LZ230616 with a reconstructed nuclear recoil energy $E_R=248\pm23_{\rm stat}\pm23_{\rm sys}~{\rm keV}$. We investigate a possible interpretation in terms of inelastic scattering between two Majorana dark matter states mediated by an axionlike particle coupled to gluons. The positive mass splitting suppresses low-energy recoils, while the momentum dependence of the interaction reshapes the high-energy spectrum. We treat the dark-sector and gluonic couplings independently and retain the momentum-dependent nucleon form factors and xenon nuclear responses. Using an approximate single-event likelihood, we find that, for $m_a=0.3~{\rm GeV}$, a narrow spectrum near the candidate energy arises at $m_χ\simeq0.35~{\rm TeV}$ and $δ\simeq330~{\rm keV}$, although this configuration requires a large coupling product and is highly sensitive to the Galactic halo speed cutoff. Our analysis establishes the kinematic and coupling requirements for subsequent tests using the thermal relic abundance and laboratory constraints on the mediator.

hep-ph↗
arXiv2026

Improving Generalizability and Undetectability for Targeted Adversarial Attacks on Multimodal Pre-trained Models

Multimodal pre-trained models (e.g., ImageBind), which align distinct data modalities into a shared embedding space, have shown remarkable success across downstream tasks. However, their increasing adoption raises serious security concerns, especially regarding targeted adversarial attacks. In this paper, we show that existing targeted adversarial attacks on multimodal pre-trained models still have limitations in two aspects: generalizability and undetectability. Specifically, the crafted targeted adversarial examples (AEs) exhibit limited generalization to partially known or semantically similar targets in cross-modal alignment tasks (i.e., limited generalizability) and can be easily detected by simple anomaly detection methods (i.e., limited undetectability). To address these limitations, we propose a novel method called Proxy Targeted Attack (PTA), which leverages multiple source-modal and target-modal proxies to optimize targeted AEs, ensuring they remain evasive to defenses while aligning with multiple potential targets. We also provide theoretical analyses to highlight the relationship between generalizability and undetectability and to ensure optimal generalizability while meeting the specified requirements for undetectability. Furthermore, experimental results demonstrate that our PTA can achieve a high success rate across various related targets and remain undetectable against multiple anomaly detection methods.

cs.CV↗
arXiv2026

AgentBrew: Offline Tool-Use Agent Learning from Raw Real-World Trajectories

LLM-based agents are increasingly deployed in real-world applications through tool-use APIs, yet training them for specific environments remains fundamentally difficult: real-world applications provide no pre-defined tasks or verifiers, no faithful simulators, and limited budget for large-scale environment interaction. In this paper, we propose \textbf{AgentBrew}, an offline training framework that learns effective tool-use policies from a single batch of raw interaction trajectories, without task verifiers or iterative on-policy rollouts. The agent first explores the target environment to collect a raw trajectory corpus without quality filtering. To extract training signal from this noisy corpus, \emph{retrospective task inference} reconstructs an aligned instruction for each trajectory based on its actual outcome, and \emph{PMI-Based credit assignment} decomposes the trajectory's total information about the inferred instruction into additive per-action credits via pointwise mutual information (PMI). These credits weight the policy training objective, amplifying informative actions while suppressing ineffective ones. On three real-world MCP applications (GitHub, Notion, PostgreSQL), AgentBrew improves Qwen3-32B by +8.7 Acc / +9.7 Score on average, surpassing Qwen3-235B (+2.3 / +4.4) and outperforming rejection sampling (+5.9 / +10.3). These results demonstrate that fine-grained offline learning can recover useful supervision from raw trajectories that filtering-based approaches would discard. The code is available at https://github.com/alphatogo/AgentBrew

cs.AI↗
arXiv2026

DE-Venus: A Data-Efficient RLVR Framework for Large Language Models

Reinforcement learning with verifiable rewards (RLVR) improves large language model reasoning, but its practical scaling is constrained by expensive on-policy rollouts and the cost of obtaining reliable targets at scale. Existing methods address sample selection, incomplete supervision, or noisy labels separately, often entangling supervision logic with distributed training and hindering controlled comparison and reuse. We present DE-Venus, a unified framework for data-efficient RLVR that treats supervision as evolving state across data preparation and policy optimization. It organizes this lifecycle into three modules: Active Data Selection allocates training and annotation budgets; Weak Supervision Construction derives learning signals from unlabeled examples; and Training-Time Supervision Refinement filters or corrects unreliable supervision. DE-Venus supports seven representative methods and a data-selection pipeline by expressing method-specific decisions as offline dataset transitions or online transformations of targets, rewards, batches, and advantages while preserving verl's distributed execution contracts. Across public benchmarks and three business scenarios, separate configurations preserve or improve model quality with only 10% of labels or as little as 13% of relevant data; selected business configurations also reduce observed convergence steps by 63%--75%. DE-Venus thus reduces annotation and training costs without sacrificing scalable RL execution.

cs.LG↗
arXiv2026

Searching for Solar-Basin Axionlike-Particle Decay with XMM-Newton Blank-Sky Observations

Axion-like particles (ALPs) bound in the solar gravitational field form the so-called ALP solar-basin. Since the two-photon decay of non-relativistic particles is approximately isotropic, this population can be searched for using observations in the anti-solar direction. In this work, we propose a search strategy for narrow decay-line signals from the ALP solar basin using \textit{XMM-Newton} blank-sky observations (XMM-BSOs) stacked spectra data taken in directions opposite to the Sun. By jointly fitting the signal and background model, we obtain limits on $g_{aγγ}^{95}$ in the mass range $m_a=1.4\text{--}16~{\rm keV}$, with typical sensitivities of $g_{aγγ}\sim10^{-10}\text{--}10^{-11}~{\rm GeV}^{-1}$. We have implemented the first anti-solar search for the solar basin, demonstrating that this strategy can exploit the stacked exposure of a large number of X-ray observations and provide a scalable analysis framework for future searches.

hep-ph↗
arXiv2026

Synchrotron self-Compton process for constraining sub-GeV dark matter in Omega Centauri via SKA

The search for the particle identity of dark matter (DM) continues to be a primary objective in modern physics. In this field, the sub-GeV mass range of DM detection remains a crucial yet challenging window. We investigate synchrotron self-Compton (SSC) emission from electrons and positrons produced by MeV-scale DM annihilation as a novel indirect detection channel. Focusing on the globular cluster Omega Centauri and the sensitivity of the Square Kilometre Array, we derive constraints on the annihilation cross section reaching $\langleσv\rangle \sim 10^{-29}\,\rm{cm}^{3}\,\rm{s}^{-1}$ in the tens-of-MeV range. Furthermore, constraints can reach below $\langleσv\rangle \sim 10^{-30}\,\rm{cm}^{3}\,\rm{s}^{-1}$ under favorable parameter choices. Although the derived limits depend on the uncertain propagation parameters, the SSC channel remains competitive with existing indirect constraints over a representative range of astrophysical assumptions, establishing SSC emission as a promising probe of sub-GeV DM.

astro-ph.CO↗
arXiv2026

When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills

Persona skills distill personal interaction histories into portable and executable artifacts for downstream agents. While enabling flexible personalization, this process concentrates fragmented personal signals, amplifies their impact through reuse, and challenges defenses designed for individual records or retrieval-based memory. To systematically investigate the safety of the persona-skill pipeline, we introduce AntiSkillBench, an end-to-end benchmark for evaluating risks and defenses across the persona-skill pipeline. It comprises: (i) a dataset of 7,500 persona-grounded dialogue traces, constructed from 50 behaviorally rich profiles spanning diverse task scenarios; (ii) an evaluation suite that measures skill-level privacy leakage and agent-level attribute disclosure and behavioral impersonation across three skill-distillation strategies; and (iii) a defense evaluation covering four configurations across online and post-hoc interventions, including active risk suppression and passive provenance protection. Experiments across three frontier agents show that persona-skill risks persist across agent backbones and distillation protocols, extending from explicit attributes to communication styles and personality traits. Existing defenses exhibit limited and distillation-dependent effectiveness, failing to generalize across risk and distillation strategies. These results highlight AntiSkillBench as a challenging benchmark for developing privacy-preserving and authenticity-aware persona skills.

cs.CR↗
arXiv2026

When Collaboration Becomes a Trigger: Collective Evidence-Threshold Backdoors in Multi-Agent Systems

LLM-based multi-agent systems (MAS) extend LLM capabilities through iterative communication and shared contexts. However, this collaboration introduces a vulnerability: backdoor behavior can be activated when peer evidence reaches a hidden threshold, rather than being determined by any single message. We introduce a collective evidence-threshold backdoor paradigm for MAS and Boundary-Conditioned Backdoor Injection (BCBI), which constructs counterfactual boundary pairs to separate benign behavior before the threshold from the adversarial objective after it, and learns latent progression aligned with evidence. To mitigate this threat, we propose LAtent Transition Test-time Evaluation (LATTE), a clean-only latent-transition defense that learns benign communication dynamics and quarantines anomalous agent updates before their responses propagate. Across several benchmarks, BCBI yields selective activation with little premature activation; without knowing the attack target or trigger, LATTE limits propagation with minimal disruption.

cs.MA↗
arXiv2026

PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective

Real-time bidding is central to computational advertising, comprising three elements: Supply Side Platform (SSP) selling ad impressions, Demand Side Platform (DSP) bidding for advertisers, and Ad Exchange conducting auctions between them. Traditional auto-bidding algorithms focus solely on the DSP side, maximizing advertiser conversions by adjusting bids against competitors. However, current big ad platforms, such as social media and e-commerce companies, now integrate SSP, DSP, and Ad Exchange functions internally. From such ad platforms' perspective, the goal of the auto-bidding algorithms is not only to maximize the advertisers' conversions, but also the total revenue of the platform. Given the lack of platform-centric evaluation frameworks and the pressing need to advance auto-bidding research, we propose PlatformBid - the first comprehensive benchmark designed from a unified ad platform's perspective. To accurately reflect the real-world auto-bidding scenarios, we define three representative settings: (1) homogeneous competition with identical algorithms across advertisers, (2) heterogeneous competition with diverse algorithmic strategies, and (3) promotional competition where some advertisers surge budgets for boosting sales during promotional events like Black Friday. We systematically evaluate a broad spectrum of existing auto-bidding methods across these settings, encompassing classical control methods, RL-based methods, and recent generative methods. Besides these methods, we further propose a novel auto-bidding method based on flow-matching, termed BidFlow, which leverages the flow-matching method's expressive policy representation to effectively handle dynamic competitive environments. Online experiments on Kuaishou further show a +0.68\% improvement in target cost, providing deployment evidence for the offline-online consistency of PlatformBid.

cs.LG↗
arXiv2026

Evidence for a $\sim 43$ GeV $γ$-ray line signal in a stacking analysis of the Virgo, Fornax, and Ophiuchus Galaxy clusters

As the largest gravitationally bound objects in the Universe, galaxy clusters have provided the first piece of evidence for the presence of dark matter and may be suitable targets for indirect dark matter searches. Among various signals, the GeV-TeV $γ$-ray line has been taken as the smoking-gun signal of the dark matter annihilation and decay since no known astrophysical or physical process(es) could generate such a peculiar spectrum. With 15.5 years of Fermi-LAT P8R3 publicly available data, we search for the $γ$-ray line emission in the directions of 13 massive galaxy clusters at redshifts $z \leq 0.028$ with an unbinned likelihood analysis. A $γ$-ray line signal at $\sim 43.2$ GeV has a net test statistic (TS) value of $\approx 30$ if we take into account only the data in the directions of the Virgo, Fornax, and Ophiuchus clusters, three massive clusters with the highest J-factors expected to generate the dark matter annihilation signal. The signal still presents when the data of 10 other nearby massive clusters have also been included, though the TS value decreases to $\approx 21$, likely because of their lower signal-to-noise ratios. The absence of this signal in the inner Galaxy disfavors both the instrumental effect and the canonical dark matter annihilation interpretation, and a more sophisticated dark matter model or very peculiar astrophysical scenario might be needed. This $γ$-ray line signal, if intrinsic, could be unambiguously verified by the Very Large Area $γ$-ray Space Telescope in its first two years of performance.

astro-ph.HE↗
arXiv2026

TokenSwap: Backdoor Attack on the Compositional Understanding of Large Vision-Language Models

Large vision-language models (LVLMs) have achieved impressive performance across a wide range of vision-language tasks, while they remain vulnerable to backdoor attacks. Existing backdoor attacks on LVLMs aim to force the victim model to generate a predefined target pattern, which is either inserted into or replaces the original content. We find that these fixed-pattern attacks are relatively easy to detect, because the attacked LVLM tends to memorize such frequent patterns in the training dataset, thereby exhibiting overconfidence on these targets given poisoned inputs. To address these limitations, we introduce TokenSwap, a more evasive and stealthy backdoor attack that focuses on the compositional understanding capabilities of LVLMs. Instead of enforcing a fixed targeted content, TokenSwap subtly disrupts the understanding of object relationships in text. Specifically, it causes the backdoored model to generate outputs that mention the correct objects in the image but misrepresent their relationships (i.e., bags-of-words behavior). During training, TokenSwap injects a visual trigger into selected samples and simultaneously swaps the grammatical roles of key tokens in the corresponding textual answers. However, the poisoned samples exhibit only subtle differences from the original ones, making it challenging for the model to learn the backdoor behavior. To address this, TokenSwap employs an adaptive token-weighted loss that explicitly emphasizes the learning of swapped tokens, such that the visual triggers and bags-of-words behavior are associated. Extensive experiments demonstrate that TokenSwap achieves high attack success rates while maintaining superior evasiveness and stealthiness across multiple benchmarks and various LVLM architectures.

cs.CV↗
arXiv2026

Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making

Recent studies on world modeling for Large Language Model (LLM) agents typically formulate the learning objective as next-observation prediction. However, this objective ties supervision to what a transition happens to reveal, which may omit the dynamics most relevant to the agent's current decision. To bridge this gap, we propose Agent-Authored World Modeling (AAWM), a training procedure that constructs supervision from the policy's own decision needs. Specifically, at each state, the agent identifies what it needs to understand about the environment before acting. These needs drive the retrieval of relevant transition evidence across trajectories, which is then synthesized into training targets that capture decision-oriented dynamics instead of reconstructing the next observation. This aligns the training objective with the dynamics the policy needs before acting, not with the contents of the next observation. Experimental results validate the effectiveness of AAWM across multiple environments and training settings. These results show that decision-aware world-model targets provide a more effective learning signal than next-observation prediction.

cs.CL↗
arXiv2026

CARPRT: Class-Aware Zero-Shot Prompt Reweighting for Black-Box Vision-Language Models

Pre-trained vision-language models (VLMs) enable zero-shot image classification by computing the similarity score between an image and textual descriptions, typically formed by inserting a class label (e.g., "cat") into a prompt (e.g., "a photo of a"). Since the score for a given image-class pair is sensitive to the choice of prompt, existing studies ensemble multiple prompts using a weighting vector to aggregate scores across different prompts. Yet, in current strategies, the weighting vector assigned to each prompt is shared across all classes, implicitly assuming that prompts are conditionally independent of classes, which often does not hold in practice, as a prompt like "an aerial view of" might be apt for "airport" but ill-suited for "apple". To address this, we propose class-aware zero-shot prompt reweighting (CARPRT). This scoring scheme adjusts the weighting vector for each class label by capturing the class-specific relevance of different prompts in a training-free manner. For each class label and every available prompt, we quantify their class-specific relevance by averaging image-text relevance scores over images predicted to that class under the given prompt. These estimates are then normalized to derive class-specific weights. Evaluations on standard image classification benchmarks show that CARPRT outperforms existing class-independent reweighting methods, confirming that modeling prompt-class dependencies is crucial for effective zero-shot prediction and even broader VLM-based application settings that rely on prompt ensembling. Our code is available at https://github.com/tmlr-group/CARPRT.

cs.LG↗
arXiv2026

Graph Diffusion-Based AeBS Deployment and Resource Allocation in RSMA-Enabled URLLC Low-Altitude Wireless Networks

As a key component of low-altitude wireless networks, aerial base stations (AeBSs) provide flexible and reliable wireless coverage to support 6G ultra-reliable and low-latency communication (URLLC) services. However, limited spectrum resources and severe co-channel interference pose significant challenges to the deployment and resource allocation of AeBSs. To address these limitations, this paper proposes a novel rate-splitting multiple access (RSMA)-enabled transmission design to manage interference and enhance URLLC services in spectrum-constrained multi-AeBS networks. We formulate a joint optimization problem involving AeBS deployment, user association, and resource allocation to maximize the sum rate and coverage of system. Given the NP-hard nature of the problem, we propose a novel alternating optimization framework based on the generative graph diffusion models. Specifically, we model AeBSs and ground users as graph nodes, then we employ a discrete graph generation process solved via denoising diffusion to explore the combinatorial space of deployment and association strategies. Moreover, the successive convex approximation (SCA) is adopted to optimize AeBS beamforming and RSMA rate allocation under finite blocklength constraints. Extensive simulations demonstrate that the proposed algorithm outperforms existing methods in terms of convergence speed, sum rate, and coverage, while also exhibiting robust performance under varying network densities and interference levels.

cs.NI↗
arXiv2026

Post-Selection Probability and Fidelity of Bidirectional Teleportation

Understanding the scrambling of quantum information is central to many areas of quantum physics, including quantum thermalization, entanglement growth, and quantum information processing. Insights from these studies have, in turn, inspired the development of novel quantum protocols and algorithms. Recently, a bidirectional teleportation protocol was proposed to implement a digital SWAP operation between qubits by leveraging chaotic Hamiltonian evolution combined with measurement and post-selection. In this work, we provide a comprehensive study of two central quantities that characterize the protocol, the post-selection probability and the fidelity, taking into account possible errors in time-reversed dynamics. We show that these quantities can be expressed in terms of standard diagnostics in quantum dynamics, including the Loschmidt echo and its subsystem variant. The results unveil (1) the initial-state dependence of the fidelity and (2) the stability of the post-selection probability in integrable models. Our findings offer practical guidance for the implementation of the protocol on realistic quantum devices.

quant-ph↗
arXiv2026

Understanding Diversity Collapse in RLVR via the Lens of Overtraining

Reinforcement learning with verifiable rewards (RLVR) has become a key approach for enhancing the reasoning abilities of large language models. However, RLVR often suffers from \emph{diversity collapse}: Pass@$1$ improves while high-$k$ Pass@$k$ degrades, which is viewed as a narrowing of the model's reasoning boundary. We formalize this diversity collapse through the lens of \emph{overtraining}: once a problem's contribution to the reference metric has effectively saturated, further updates no longer expand what the model can solve but still concentrate probability mass on the trajectories favored by on-policy sampling. Under a standard setup with few rollouts per problem, even a single observed success places a problem in a nearly saturated regime for high-$k$ Pass@$k$, so most updates in standard RLVR are overtraining from the boundary perspective. This perspective also suggests a reading of whether RLVR can expand the model's reasoning abilities beyond the base model: since RLVR is structurally biased against high-$k$ Pass@$k$, its aggregate decline does not by itself mean that no new reasoning gains occurred. Interventionally, restricting updates to problems with zero observed success lifts Pass@$256$ above the base model on difficult benchmarks; observationally, a non-trivial fraction of initially unsolvable problems become solvable during standard RLVR training. Building on these findings, we propose \emph{Bayesian Boundary Gating} (BBG), which redirects optimization away from overtraining by estimating each problem's marginal contribution to the reasoning boundary. Across multiple reasoning benchmarks, BBG improves average Pass@$k$ across a wide range of $k$.

cs.LG↗