Searcharxiv⌕ Search

arXiv subjects

Zhe Cao

Publications and source records attributed to Zhe Cao.

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

T2AV-Compass: Towards Unified Evaluation for Text-to-Audio-Video Generation

Text-to-Audio-Video (T2AV) generation aims to synthesize temporally coherent video and semantically synchronized audio from natural language, yet its evaluation remains fragmented, often relying on unimodal metrics or narrowly scoped benchmarks that fail to capture cross-modal alignment, instruction following, and perceptual realism under complex prompts. To address this limitation, we present T2AV-Compass, a unified benchmark for comprehensive evaluation of T2AV systems, consisting of 500 diverse and complex prompts constructed via a taxonomy-driven pipeline to ensure semantic richness and physical plausibility. Besides, T2AV-Compass introduces a dual-level evaluation framework that integrates objective signal-level metrics for video quality, audio quality, and cross-modal alignment with a subjective MLLM-as-a-Judge protocol for instruction following and realism assessment. Extensive evaluation of 11 representative T2AVsystems reveals that even the strongest models fall substantially short of human-level realism and cross-modal consistency, with persistent failures in audio realism, fine-grained synchronization, instruction following, etc. These results indicate significant improvement room for future models and highlight the value of T2AV-Compass as a challenging and diagnostic testbed for advancing text-to-audio-video generation.

cs.CV↗
arXiv2026

DSA: Dynamic Step Allocation for Fast Autoregressive Video Generation

Video diffusion transformers have achieved state-of-the-art visual quality, but their high inference cost remains a major bottleneck for real-time applications. Recent distillation frameworks produce autoregressive video diffusion models with reduced latency, yet these models still use a fixed number of denoising steps per frame, wasting computation on predictable frames and under-refining challenging ones. We present DSA, a confidence-guided adaptive computation framework for AR video diffusion. DSA introduces a lightweight confidence head, trained jointly with the generator under a distribution-matching distillation objective, to estimate per-frame denoising reliability. At inference, this confidence signal dynamically adjusts the number of diffusion steps: simple frames terminate early for speed, while complex frames receive additional refinement. Our method requires no extra video data, no heuristics, and little architectural modification. Experiments show that DSA achieves real-time autoregressive video generation, reaching 22.63 FPS with sub-second latency on H100 GPUs, while maintaining competitive or superior VBench quality compared to recent autoregressive and bidirectional video diffusion models. Our results demonstrate that confidence-guided adaptive sampling provides an effective and practical path toward interactive video generation.

cs.CV↗
arXiv2026

OmniHalluc-L: Counterfactual Benchmarking and Modality-Perturbation Reliability Calibration for Long-Form Omni Hallucination

Long-video Omni assistants often fail not by inventing content, but by misbinding real evidence: they hear the right utterance and see the right event, yet attach it to the wrong speaker, moment, or modality. These \emph{almost-true} errors evade standard video QA because local evidence remains valid, so item-level scoring can reward both a supported claim and its near-counterfactual. We introduce a counterfactual event-binding protocol that constructs paired supported/counterfactual claims from the same audio-visual event evidence and evaluates them by strict-pair accuracy. We instantiate it as \bench, a benchmark for long-video Omni hallucination, with 3{,}600 single-claim QA items from 638 long-form videos averaging 24.16 minutes and covering 256.87 hours. Under this protocol, open-weight Omni models remain weak at pair-level binding: Qwen2.5-Omni-7B reaches 32.06\% and Qwen3-Omni-Instruct reaches 41.55\%, versus 76.54\% for a closed-source reference. To narrow this gap without updating the backbone, we propose \method, Modality-Perturbation Reliability Calibration, a frozen-backbone framework that selects audio-negative probes within video-level folds and fuses their response shifts with native audio-visual confidence into per-claim support estimates. \method lifts Qwen2.5-Omni-7B to 36.22\% and Qwen3 to 51.09\% on \bench, and improves target-adapted MCQ accuracy on OmniVideoBench ($+$2.20) and WorldSense ($+$1.51) with Qwen3.

cs.MM↗
arXiv2026

DISK: Differentiable Sparse Kernel Complex for Efficient Spatially-Variant Convolution

Image convolution with complex kernels is a fundamental operation in photography, scientific imaging, and animation effects, yet direct dense convolution is computationally prohibitive on resource-limited devices. Existing approximations, such as simulated annealing or low-rank decompositions, either lack efficiency or fail to capture non-convex kernels. We introduce a differentiable kernel decomposition framework that represents a target spatially-variant, dense, complex kernel using a set of sparse kernel samples. Our approach features (i) a decomposition that enables differentiable optimization of sparse kernels, (ii) a dedicated initialization strategy for non-convex shapes to avoid poor local minima, and (iii) a kernel-space interpolation scheme that extends single-kernel filtering to spatially varying filtering without retraining and additional runtime overhead. Experiments on Gaussian and non-convex kernels show that our method achieves higher fidelity than simulated annealing and significantly lower cost than low-rank decompositions. Our approach provides a practical solution for mobile imaging and real-time rendering, while remaining fully differentiable for integration into broader learning pipelines.

cs.GR↗
arXiv2026

Frontier-Eng: Benchmarking Self-Evolving Agents on Real-World Engineering Tasks with Generative Optimization

Current LLM agent benchmarks, which predominantly focus on binary pass/fail tasks such as code generation or search-based question answering, often neglect the value of real-world engineering that is often captured through the iterative optimization of feasible designs. To this end, we introduce Frontier-Eng, a human-verified benchmark for generative optimization -- an iterative propose-execute-evaluate loop in which an agent generates candidate artifacts, receives executable verifier feedback, and revises them under a fixed interaction budget -- spanning $47$ tasks across five broad engineering categories. Unlike previous suites, Frontier-Eng tasks are grounded in industrial-grade simulators and verifiers that provide continuous reward signals and enforce hard feasibility constraints under constrained budgets. We evaluate eight frontier language models using representative search frameworks, finding that while GPT 5.4 achieves the most robust performance, the benchmark remains challenging for all models. Our analysis suggests a dual power-law decay in improvement frequency ($\sim$ 1/iteration) and magnitude ($\sim$ 1/improvement count). We further show that although width improves parallelism and diversity, depth remains crucial for hard-won improvements under a fixed budget. Frontier-Eng establishes a new standard for assessing the capacity of AI agents to integrate domain knowledge with executable feedback to solve complex, open-ended engineering problems.

cs.AI↗
arXiv2026

Ultra-high-energy $γ$-ray imprints from PeV particles accelerated by supernova remnants

The quest for the origin of cosmic ray (CRs) is a fundamental issue in astrophysics. Shocks of supernova remnants (SNRs) have been considered as the dominant contributors to Galactic CRs below the spectral knee near $\sim 3$ petaelectronvolt (PeV). Whether SNRs are efficient accelerators of particles beyond PeV energies has long been debated. Here we report observations of very-high-energy $γ$-ray emission up to hundreds of TeV from two middle age shell-type SNRs, G150.3$+$4.5 and $γ$-Cygni, with the Large High Altitude Air Shower Observatory (LHAASO). Two (or three) distinct morphological/spectral components with convex spectral shapes are observed in both sources, with the low-energy one being more extended than the high-energy one. %Although it is possible that these high-energy components may be driven by powerful pulsars, The likely association of the high-energy component with molecular clouds at similar distances, and the weakness/absence of pulsar wind nebulae (PWNe) inside these SNRs clearly indicate for the first time that the highest energy emission is produced by collision of hadronic CRs up to PeV energies with the clouds. These results are compatible with the classic model prediction that PeV particles accelerated near the end of the free expansion phase of SNR evolution can illuminate nearby molecular clouds (MCs) to produce strong $γ$-ray emission.

astro-ph.HE↗
arXiv2026

Cygnus X-3: A variable petaelectronvolt gamma-ray source

We report the discovery of variable $γ$-rays up to petaelectronvolt from Cygnus X-3, an iconic X-ray binary. The $γ$-ray signal was detected with a statistical significance of approximately 10 $σ$ by the Large High Altitude Air Shower Observatory (LHAASO). Its intrinsic spectral energy distribution (SED), extending from 0.06 to 3.7 PeV, shows a pronounced rise toward 1 PeV after accounting for absorption by the cosmic microwave background radiation. We find variability on month-long timescales at a significance of $8.6 σ$, coinciding with a high state of the GeV gamma-ray flux detected by the Fermi-LAT. This,together with a 3.2$σ$ evidence for orbital modulation, suggests that the PeV $γ$-rays originate within, or in close proximity to, the binary system itself. The observed energy spectrum and temporal modulation can be naturally explained by $γ$-ray production through photomeson processes in the innermost region of the relativistic jet, where protons need to be accelerated to tens of PeV energies.

astro-ph.HE↗
arXiv2026

Precise Measurement of the Cosmic Ray Helium Spectrum above 0.1 PeV

We report a measurement of the cosmic ray helium energy spectrum in the energy interval 0.16 -- 13~PeV, derived by subtracting the proton spectrum from the light component~(proton and helium) spectrum obtained with observations made by the Large High Altitude Air Shower Observatory~(LHAASO) under a consistent energy scale. The helium spectrum shows a significant hardening centered at $E \simeq$ 1.1~PeV, followed by a softening at $\sim$ 7 PeV, indicating the appearance of a helium `knee'. Comparing the proton and helium spectra in the LHAASO energy range reveals some remarkable facts. In the lower part of this range, in contrast to the behavior at lower energies, the helium spectrum is significantly softer than the proton spectrum. This results in protons overtaking helium nuclei and becoming the largest cosmic ray component at $E \simeq$ 0.7 PeV. A second crossing of the two spectra is observed at $E \simeq$ 5 PeV, above the proton knee, when helium nuclei overtake protons to become the largest cosmic ray component again. These results have important implications for our understanding of the Galactic cosmic ray sources.

astro-ph.HE↗
arXiv2026

First detection of ultra-high energy emission from gamma-ray binary LS I +61 303

We report the first detection of gamma-ray emission up to ultra-high-energy (UHE; $>$100 TeV) emission from the prototypical gamma-ray binary system LS I +61 303 using data from the Large High Altitude Air Shower Observatory (LHAASO). It is detected with significances of 9.2$σ$ in WCDA (1.4--30.5 TeV) and 6.2$σ$ in KM2A (25--267 TeV); in KM2A alone we identify 16 photon-like events above 100 TeV against an estimated 5.1 background events, corresponding to a 3.8$σ$ detection. These results provide compelling evidence of extreme particle acceleration in LS I +61 303. Furthermore, we observe orbital modulation at 3.9$σ$ confidence level, between 25 and 100 TeV, with a hint that the orbital modulation is energy-dependent. These features can be understood in a composite scenario in which leptonic and hadronic processes jointly contribute.

astro-ph.HE↗
arXiv2026

Evidence of cosmic-ray acceleration up to sub-PeV energies in the supernova remnant IC 443

Supernova remnants (SNRs) have been considered as the primary contributors to cosmic rays (CRs) in our Galaxy. However, the maximum energy of particles that can be accelerated by shocks of SNRs is uncertain, and SNRs' contribution to CRs around PeV energies is unclear. In this study, we present observations of high-energy $γ$-ray emission from the SNR IC 443 using the Large High Altitude Air Shower Observatory (LHAASO). The morphological analysis reveals a pointlike source whose location and spectrum are consistent with those of the Fermi-LAT-detected compact source with $π^0$-decay signature, and a more extended source that is associated with a newly discovered Fermi source. The spectrum of the point source can be described by a power-law function with an index of $\sim3.0$, extending beyond $\sim 30$ TeV without apparent cutoff. Assuming a hadronic origin of the $γ$-ray emission, the $95\%$ lower limit of accelerated protons reaches about 300 TeV. The extended source might be associated with IC 443, SNR G189.6+3.3 or the putative pulsar wind nebula CXOU J061705.3+222127, and can be explained by either a hadronic or a leptonic model with particles reaching hundreds of TeV. These LHAASO results provide compelling evidence that sub-PeV CRs can be accelerated efficiently by shocks of SNRs.

astro-ph.HE↗
arXiv2026

OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs

Recent advances in multimodal large language models (MLLMs) have demonstrated substantial potential in video understanding. However, existing benchmarks fail to comprehensively evaluate synergistic reasoning capabilities across audio and visual modalities, often neglecting either one of the modalities or integrating them in a logically inconsistent manner. To bridge this gap, we introduce OmniVideoBench, a large-scale and rigorously designed benchmark dedicated to assessing synergistic audio-visual understanding, with a strong emphasis on modality complementarity and logical consistency. Specifically, OmniVideoBench comprises 1000 high-quality question-answer(QA) pairs, each annotated with step-by-step reasoning traces, derived from 628 diverse videos ranging from several seconds to 30 minutes, and manually verified to guarantee complete correctness and uniqueness. Moreover, OmniVideoBench encompasses 13 carefully designed question types, covering temporal reasoning, spatial localization, counting, causal inference, summarization, and beyond, thereby capturing the essential challenges of video understanding. Evaluation of multiple MLLMs on OmniVideoBench reveals a pronounced gap between model performance and human reasoning, with open-source models lagging significantly behind their closed-source counterparts, underscoring the inherent difficulty of genuine audio-visual reasoning. We will release OmniVideoBench to foster the development of MLLMs with stronger and more generalizable reasoning capabilities.

cs.AI↗
arXiv2026

LHAASO observation of Mrk 421 during 2021 March - 2024 March: a comprehensive VHE catalog of multi-timescale outbursts and its time average behavior

The Large High Altitude Air Shower Observatory (LHAASO) monitors sources within its field of view for up to 7 hours daily, achieving a duty cycle exceeding 98% and an annual point-source sensitivity of 1.5% Crab Units (CU) in the very high energy (VHE) band. This unbiased sky-survey mode facilitates systematic monitoring and investigation of outburst phenomena. In this paper, we present results from an unprecedented three-year monitoring campaign (March 2021--March 2024) of Mrk421 using LHAASO, spanning energies from 0.4 TeV to 20 TeV. We find that the blazar stayed in a quiescent state in 2021 and became active starting in 2022 with a total of 23 VHE outburst events identified, where the highest observed daily significance reaches $20\,σ$ with a flux equivalent to approximately 3.3~CU. LHAASO's continuous monitoring suggests the flaring occupancy of Mrk~421 to be around 14%. During long-term monitoring, multiwavelength (MWL) variability and correlation analyses are conducted using complementary data from Fermi-LAT, MAXI-GSC, Swift-XRT, and ZTF. A significant correlation ($>3\,σ$) is observed between X-ray and VHE bands with no detectable time lag, while the correlation between GeV and TeV bands is weaker. The flux distribution of the TeV emission during the quiescent state is different from that in the active state, implying the existence of two modes of energy dissipation in the blazar jet. Using simultaneous MWL data, we also analyzed both the long-term and outburst-period SEDs, and discussed the possible origin of the outburst events.

astro-ph.HE↗
arXiv2026

Stringent Tests of Lorentz Invariance Violation from LHAASO Observations of GRB 221009A

On October 9, 2022, the Large High Altitude Air Shower Observatory (LHAASO) reported the observation of the very early TeV afterglow of the brightest-of-all-time GRB 221009A, recording the highest photon statistics in the TeV band ever from a gamma-ray burst. We use this unique observation to place stringent constraints on an energy dependence of the speed of light in vacuum, a manifestation of Lorentz invariance violation (LIV) predicted by some quantum gravity (QG) theories. Our results show that the 95% confidence level lower limits on the QG energy scales are $E_{\mathrm{QG},1}>10$ times of the Planck energy $E_\mathrm{Pl}$ for the linear, and $E_{\mathrm{QG},2}>6\times10^{-8}E_\mathrm{Pl}$ for the quadratic LIV effects, respectively. Our limits on the quadratic LIV case improve previous best bounds by factors of 5--7.

astro-ph.HE↗
arXiv2026

On very badly approximable numbers

We prove a refined version of Markov's theorem in Diophantine approximation. More precisely, we characterize completely the set of irrationals $x$ such that $\left|x-\frac{p}{q}\right|<\frac{1}{3q^2}$ has only finitely many rational solutions: their continued fraction is eventually a balanced sequence through a simple coding. As consequence, we show that all such numbers are either quadratic surds or transcendental numbers. In particular, for any algebraic real number $x$ of degree at least $3$ there are infinitely rational numbers $\frac{p}{q}$ such that $\left|x-\frac{p}{q}\right|<\frac{1}{3q^2}$.

math.NT↗
arXiv2026

On irrationals with Lagrange value exactly 3

For $c>0$, let $X_c$ denote the set of $x\in\mathbb{R}\backslash\mathbb{Q}$ such that $\left| x-\frac{p}{q} \right|<\frac{1}{cq^2}$ has only finitely many rational solutions $\frac{p}{q}$. It is a classical fact, known since the 1950s, that $X_c$ is uncountable for $c>3$ and countable for $c<3$. However, the cardinality of $X_3$ does not appear to be present in the literature. We prove that $X_3$ is uncountable. More generally, we show that for any $n\in\mathbb{N}\cup\{\infty\}$, the set of $x\in\mathbb{R}\backslash\mathbb{Q}$ with Lagrange value exactly $3$ and such that $\left| x-\frac{p}{q} \right|<\frac{1}{3q^2}$ has exactly $n$ rational solutions $\frac{p}{q}$ is also uncountable.

math.NT↗
arXiv2026

SafeDialBench: A Fine-Grained Safety Evaluation Benchmark for Large Language Models in Multi-Turn Dialogues with Diverse Jailbreak Attacks

With the rapid advancement of Large Language Models (LLMs), the safety of LLMs has been a critical concern requiring precise assessment. Current benchmarks primarily concentrate on single-turn dialogues or a single jailbreak attack method to assess the safety. Additionally, these benchmarks have not taken into account the LLM's capability of identifying and handling unsafe information in detail. To address these issues, we propose a fine-grained benchmark SafeDialBench for evaluating the safety of LLMs across various jailbreak attacks in multi-turn dialogues. Specifically, we design a two-tier hierarchical safety taxonomy that considers 6 safety dimensions and generates more than 4000 multi-turn dialogues in both Chinese and English under 22 dialogue scenarios. We employ 7 jailbreak attack strategies, such as reference attack and purpose reverse, to enhance the dataset quality for dialogue generation. Notably, we construct an innovative assessment framework of LLMs, measuring capabilities in detecting, and handling unsafe information and maintaining consistency when facing jailbreak attacks. Experimental results across 17 LLMs reveal that Yi-34B-Chat and GLM4-9B-Chat demonstrate superior safety performance, while Llama3.1-8B-Instruct and o3-mini exhibit safety vulnerabilities.

cs.CL↗
arXiv2026

RoboStriker: Hierarchical Decision-Making for Autonomous Humanoid Boxing

Achieving human-level competitive intelligence and physical agility in humanoid robots remains a major challenge, particularly in contact-rich and highly dynamic tasks such as boxing. While Multi-Agent Reinforcement Learning (MARL) offers a principled framework for strategic interaction, its direct application to humanoid control is hindered by high-dimensional contact dynamics and the absence of strong physical motion priors. We propose RoboStriker, a hierarchical three-stage framework that enables fully autonomous humanoid boxing by decoupling high-level strategic reasoning from low-level physical execution. The framework first learns a comprehensive repertoire of boxing skills by training a single-agent motion tracker on human motion capture data. These skills are subsequently distilled into a structured latent manifold, regularized by projecting the Gaussian-parameterized distribution onto a unit hypersphere. This topological constraint effectively confines exploration to the subspace of physically plausible motions. In the final stage, we introduce Latent-Space Neural Fictitious Self-Play (LS-NFSP), where competing agents learn competitive tactics by interacting within the latent action space rather than the raw motor space, significantly stabilizing multi-agent training. Experimental results demonstrate that RoboStriker achieves superior competitive performance in simulation and exhibits sim-to-real transfer. Our website is available at RoboStriker.

cs.RO↗
arXiv2026

Exploring Lorentz Invariance Violation from Ultra-high-energy Gamma Rays Observed by LHAASO

Recently the LHAASO Collaboration published the detection of 12 ultra-high-energy gamma-ray sources above 100 TeV, with the highest energy photon reaching 1.4 PeV. The first detection of PeV gamma rays from astrophysical sources may provide a very sensitive probe of the effect of the Lorentz invariance violation (LIV), which results in decay of high-energy gamma rays in the superluminal scenario and hence a sharp cutoff of the energy spectrum. Two highest energy sources are studied in this work. No signature of the existence of LIV is found in their energy spectra, and the lower limits on the LIV energy scale are derived. Our results show that the first-order LIV energy scale should be higher than about 10^5 times the Planck scale M_{pl} and that the second-order LIV scale is >10^{-3}M_{pl}. Both limits improve by at least one order of magnitude the previous results.

astro-ph.HE↗