arXiv subjects

Zijian Zhang

Publications and source records attributed to Zijian Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2026

InfantAgent-Next: A Multimodal Generalist Agent for Automated Computer Interaction

This paper introduces \textsc{InfantAgent-Next}, a generalist agent capable of interacting with computers in a multimodal manner, encompassing text, images, audio, and video. Unlike existing approaches that either build intricate workflows around a single large model or only provide workflow modularity, our agent integrates tool-based and pure vision agents within a highly modular architecture, enabling different models to collaboratively solve decoupled tasks in a step-by-step manner. Our generality is demonstrated by our ability to evaluate not only pure vision-based real-world benchmarks (i.e., OSWorld), but also more general or tool-intensive benchmarks (e.g., GAIA and SWE-Bench). Specifically, we achieve $\mathbf{7.27\%}$ accuracy on OSWorld, higher than Claude-Computer-Use. Codes and evaluation scripts are open-sourced at https://github.com/bin123apple/InfantAgent.

cs.AI↗

arXiv2026

QCalEval: Benchmarking Vision-Language Models for Quantum Calibration Plot Understanding

Quantum computing calibration depends on interpreting experimental data, and calibration plots provide the most universal human-readable representation for this task, yet no systematic evaluation exists of how well vision-language models (VLMs) interpret them. We introduce QCalEval, the first VLM benchmark for quantum calibration plots: 243 samples across 87 scenario types from 22 experiment families, spanning superconducting qubits and neutral atoms, evaluated on six question types in both zero-shot and in-context learning settings. The best general-purpose zero-shot model reaches a mean score of 72.3, and many open-weight models degrade under multi-image in-context learning, whereas frontier closed models improve substantially. A supervised fine-tuning ablation at the 9-billion-parameter scale shows that SFT improves zero-shot performance but cannot close the multimodal in-context learning gap. As a reference case study, we release NVIDIA Ising Calibration 1, an open-weight model based on Qwen3.5-35B-A3B that reaches 74.7 zero-shot average score.

quant-ph↗

arXiv2026

Diverse Emission Patterns from Precessing Super-Eddington Disks Formed in Tidal Disruption Events

A tidal disruption event (TDE) occurs when a star passes within the tidal radius of a supermassive black hole (SMBH). In TDEs it is expected that the orbital angular momentum of the disrupted star is generally misaligned with the SMBH spin axis, which should result in a misaligned super-Eddington disk precessing around the SMBH spin axis due to the Lense-Thirring effect. In this paper, we investigate the distinct observational signatures produced from such TDE disks, by performing radiative transfer calculations upon previous super-Eddington disk simulations. We demonstrate that the precession of the disk and wind drive time-dependent obscuration and reprocessing of X-ray radiation. Depending on the orientation of the viewing angle of the observer and the tilt angle of the disk, four main types of variability are induced: 1) The smooth-TDEs: The emissions from these TDEs show no fluctuations; 2) The dimmer: The main emission type (X-ray or optical) stays the same, with small to moderate modulations of brightness; 3) The blinker: X-ray and optical emissions take turns to dominate in one cycle of precession, with dramatic changes in the X-ray fluxes. 4) The siren: X-ray and optical emissions take over each other twice per cycle, possibly with two different peak X-ray fluxes within one cycle. In all three scenarios, we observe an inverse correlation between X-ray and optical emissions. Our model provides a physical framework for interpreting TDE multi-wavelength variability through disk precession dynamics and gives an alternative interpretation to the interesting case of J045650.3-203750 which was suggested to be a repeated partial TDE previously.

astro-ph.HE↗

arXiv2026

Rényi Pufferfish Privacy with Gaussian-based Priors: From Single Gaussian to Mixture Model

Rényi Pufferfish Privacy (RPP) provides a Rényi divergence-based privacy framework for correlated data, but existing $\infty$-Wasserstein mechanisms are often conservative and sacrifice data utility. We study Gaussian mechanisms for RPP under Gaussian and Gaussian-mixture priors. For single Gaussian priors, we derive the exact Rényi divergence after Gaussian perturbation, obtain a relaxed closed-form sufficient condition for $(α,ε)$-RPP, and characterize the monotonicity of the calibrated noise with respect to the privacy budget $ε$ and the Rényi order $α$. To handle more general non-Gaussian and multimodal priors, we approximate secret-conditioned outputs with Gaussian mixture models and introduce an optimal-transport-based sufficient condition for RPP. Experiments on three UCI datasets with statistical (\textsc{RAW}, \textsc{MEAN}) and model-output (\textsc{BNN}, \textsc{GP}) queries show that our prior-aware mechanisms consistently require less noise than a recent RPP additive-noise baseline, achieving an average noise reduction of 48.9\%. These results show that our mechanisms can substantially improve the privacy-utility trade-off under RPP.

cs.CR↗

arXiv2026

Multi-user Pufferfish Privacy

This paper studies how to achieve individual indistinguishability by pufferfish privacy in aggregated query to a multi-user system. It is assumed that each user reports realization of a random variable. We study how to calibrate Laplace noise, added to the query answer, to attain pufferfish privacy when user changes his/her reported data value, leaves the system and is replaced by another use with different randomness. Sufficient conditions are derived for all scenarios for attaining statistical indistinguishability on four sets of secret pairs. They are derived using the existing Kantorovich method (Wasserstain metric of order $1$). These results can be applied to attain indistinguishability when a certain class of users is added or removed from a tabular data. It is revealed that attaining indifference in individual's data is conditioned on the statistics of this user only. For binary (Bernoulli distributed) random variables, the derived sufficient conditions can be further relaxed to reduce the noise and improve data utility.

cs.CR↗

arXiv2026

El Agente Forjador: Task-Driven Agent Generation for Quantum Simulation

AI for science promises to accelerate the discovery process. The advent of large language models (LLMs) and agentic workflows enables the expediting of a growing range of scientific tasks. However, most of the current generation of agentic systems depend on static, hand-curated toolsets that hinder adaptation to new domains and evolving libraries. We present El Agente Forjador, a multi-agent framework in which universal coding agents autonomously forge, validate, and reuse computational tools through a four-stage workflow of tool analysis, tool generation, task execution, and iterative solution evaluation. Evaluated across 24 tasks spanning quantum chemistry and quantum dynamics on five coding agent setups, we compare three operating modes: zero-shot generation of tools per task, reuse of a curriculum-built toolset, and direct problem-solving with the coding agents as the baseline. We find that our tool generation and reuse framework consistently improves accuracy over the baseline. We also show that reusing a toolset built by a stronger coding agent can reduce API cost and substantially raises the solution quality for weaker coding agents. Case studies further demonstrate that tools forged for different domains can be combined to solve hybrid tasks. Taken together, these results show that LLM-based agents can use their scientific knowledge and coding capabilities to autonomously build reusable scientific tools, pointing toward a paradigm in which agent capabilities are defined by the tasks they are designed to solve rather than by explicitly engineered implementations.

cs.AI↗

arXiv2026

Conformal Margin Risk Minimization: An Envelope Framework for Robust Learning under Label Noise

Most methods for learning with noisy labels require privileged knowledge such as noise transition matrices, clean subsets or pretrained feature extractors, resources typically unavailable when robustness is most needed. We propose Conformal Margin Risk Minimization (CMRM), a plug-and-play envelope framework that improves any classification loss under label noise by adding a single quantile-calibrated regularization term, with no privileged knowledge or training pipeline modification. CMRM measures the confidence margin between the observed label and competing labels, and thresholds it with a conformal quantile estimated per batch to focus training on high-margin samples while suppressing likely mislabeled ones. We derive a learning bound for CMRM under arbitrary label noise requiring only mild regularity of the margin distribution. Across five base methods and six benchmarks with synthetic and real-world noise, CMRM consistently improves accuracy (up to +3.39%), reduces conformal prediction set size (up to -20.44%) and does not hurt under 0% noise, showing that CMRM captures a method-agnostic uncertainty signal that existing mechanisms did not exploit.

cs.LG↗

arXiv2026

ModalImmune: Immunity Driven Unlearning via Self Destructive Training

Multimodal systems are vulnerable to partial or complete loss of input channels at deployment, which undermines reliability in real-world settings. This paper presents ModalImmune, a training framework that enforces modality immunity by intentionally and controllably collapsing selected modality information during training so the model learns joint representations that are robust to destructive modality influence. The framework combines a spectrum-adaptive collapse regularizer, an information-gain guided controller for targeted interventions, curvature-aware gradient masking to stabilize destructive updates, and a certified Neumann-truncated hyper-gradient procedure for automatic meta-parameter adaptation. Empirical evaluation on standard multimodal benchmarks demonstrates that ModalImmune improves resilience to modality removal and corruption while retaining convergence stability and reconstruction capacity.

cs.LG↗

arXiv2026

TR-ICRL: Test-Time Rethinking for In-Context Reinforcement Learning

In-Context Reinforcement Learning (ICRL) enables Large Language Models (LLMs) to learn online from external rewards directly within the context window. However, a central challenge in ICRL is reward estimation, as models typically lack access to ground-truths during inference. To address this limitation, we propose Test-Time Rethinking for In-Context Reinforcement Learning (TR-ICRL), a novel ICRL framework designed for both reasoning and knowledge-intensive tasks. TR-ICRL operates by first retrieving the most relevant instances from an unlabeled evaluation set for a given query. During each ICRL iteration, LLM generates a set of candidate answers for every retrieved instance. Next, a pseudo-label is derived from this set through majority voting. This label then serves as a proxy to give reward messages and generate formative feedbacks, guiding LLM through iterative refinement. In the end, this synthesized contextual information is integrated with the original query to form a comprehensive prompt, with the answer determining through a final round of majority voting. TR-ICRL is evaluated on mainstream reasoning and knowledge-intensive tasks, where it demonstrates significant performance gains. Remarkably, TR-ICRL improves Qwen2.5-7B by 21.23% on average on MedQA and even 137.59% on AIME2024. Extensive ablation studies and analyses further validate the effectiveness and robustness of our approach. Our code is available at https://github.com/pangpang-xuan/TR_ICRL.

cs.CL↗

arXiv2026

$AutoDrive\text{-}P^3$: Unified Chain of Perception-Prediction-Planning Thought via Reinforcement Fine-Tuning

Vision-language models (VLMs) are increasingly being adopted for end-to-end autonomous driving systems due to their exceptional performance in handling long-tail scenarios. However, current VLM-based approaches suffer from two major limitations: 1) Some VLMs directly output planning results without chain-of-thought (CoT) reasoning, bypassing crucial perception and prediction stages which creates a significant domain gap and compromises decision-making capability; 2) Other VLMs can generate outputs for perception, prediction, and planning tasks but employ a fragmented decision-making approach where these modules operate separately, leading to a significant lack of synergy that undermines true planning performance. To address these limitations, we propose ${AutoDrive\text{-}P^3}$, a novel framework that seamlessly integrates $\textbf{P}$erception, $\textbf{P}$rediction, and $\textbf{P}$lanning through structured reasoning. We introduce the ${P^3\text{-}CoT}$ dataset to facilitate coherent reasoning and propose ${P^3\text{-}GRPO}$, a hierarchical reinforcement learning algorithm that provides progressive supervision across all three tasks. Specifically, ${AutoDrive\text{-}P^3}$ progressively generates CoT reasoning and answers for perception, prediction, and planning, where perception provides essential information for subsequent prediction and planning, while both perception and prediction collectively contribute to the final planning decisions, enabling safer and more interpretable autonomous driving. Additionally, to balance inference efficiency with performance, we introduce dual thinking modes: detailed thinking and fast thinking. Extensive experiments on both open-loop (nuScenes) and closed-loop (NAVSIMv1/v2) benchmarks demonstrate that our approach achieves state-of-the-art performance in planning tasks. Code is available at https://github.com/haha-yuki-haha/AutoDrive-P3.

cs.RO↗

arXiv2026

LongCat-Next: Lexicalizing Modalities as Discrete Tokens

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

cs.CV↗

arXiv2026

Rydberg Atomic Receivers for Net-Zero 6G Wireless Communication and Sensing: Progress, Experiments, and Sustainable Prospects

Against the backdrop of the global drive to advance the green transformation of the information and communications technology (ICT) industry and leverage technological innovation to facilitate the achievement of Net-Zero carbon goals, research into Rydberg atomic receivers (RAREs) is gaining significant interest. RAREs leverage the electron transition phenomenon for signal reception, offering significant advantages over conventional radio frequency receivers in terms of miniaturized antenna design, high sensitivity, robust interference resistance, and compact form factors, which positions them as a competitive alternative for meeting zero-carbon communication demands. This article systematically elaborates on the basic principle, state-of-the-art progress, and novel experiments of RAREs in quantum wireless communication and sensing. In this first-of-its-kind work, we experimentally verify the RARE-based orthogonal frequency division multiplexing transmission and reveal the potential of deep learning design in optimizing quantum wireless systems. Finally, we delve into the prospect of integrating RARE with existing cutting-edge application scenarios, while mapping out critical pathways for developing Rydberg-based wireless systems.

eess.SP↗

arXiv2026

COSMOS-3D: Dense Circumnuclear Gas across Black Hole Growth Phases at z ~ 3

We report the discovery of two broad-line X-ray AGNs (cid_414 and cid_947) at z ~ 3 identified in the JWST Cycle 3 COSMOS-3D program using NIRCam F444W grism spectroscopy. Both exhibit prominent HeI+Pa$γ$ emission and absorption, indicative of circumnuclear dense gas that is traced in these systems. Complementary UV and optical spectroscopy in the COSMOS field provides Ly$α$, Si IV, and C IV measurements. Both sources are detected in MIRI F1000W, and cid_414 is also detected in F2100W, indicating hot dust emission. The two AGNs show distinct black hole and obscuration properties. The source cid_414 displays little red dots (LRD)-like V-shape spectra energy distribution (SED) shape with a turnover near the Balmer 4000 Å break, and a narrow Ly$α$ line with $\log L_{\rm Lyα}=42.49\pm0.01~\mathrm{erg\ s^{-1}}$, with no additional metal lines detected. In contrast, cid_947 exhibits a higher HeI absorption column density, larger X-ray--inferred $N_{\rm H}$, lower intrinsic 2--10 keV luminosity, and strong blueshifted features in He I, Si IV, and C IV absorption with velocity offsets exceeding $5000~\mathrm{km\ s^{-1}}$. Photoionization modeling implies gas densities of $\sim10^{9-10} \mathrm{cm^{-3}}$ and sizes comparable to the broad-line region, consistent with dense gas envelopes predicted for LRDs. Together with previous detections of HeI absorption in compact little red dots, these results suggest that dense circumnuclear gas is likely prevalent at high redshift and may regulate obscuration and black hole--host co-evolution across AGN types.

astro-ph.GA↗

arXiv2026

COSMOS-3D: Black Hole Mass Estimators and Luminosity Functions of Paschen-line AGNs

Near-IR Paschen lines are potentially an excellent tracer of Type 1 AGNs that is hardly affected by dust extinction. JWST allows us, for the first time, to explore Paschen-line objects at redshift z>1. Here we present a study of 62 AGNs with broad Pa$α$ and Pa$β$ lines at 1<z<3 using data from the JWST COSMOS-3D program. These AGNs are efficiently selected and identified using NIRCam imaging and grism slitless spectroscopic data. We separate the AGN-host emission with image decomposition and quantify dust attenuation with multi-band data. We construct a calibration sample with optical spectroscopy and use single-epoch, Mg II-based black hole masses ($M_{\mathrm{BH}}$) as an anchor to derive new, Paschen-based $M_{\mathrm{BH}}$ estimators. We obtain three sets of $M_{\mathrm{BH}}$ estimators based on Paschen line luminosities and AGN continuum luminosities at 1 and 2 $μ$m, respectively. After dust corrections, they are well consistent with each other, and also broadly agree with previous results. With this AGN sample, we further construct the first Pa$α$ and Pa$β$ luminosity functions (LFs) of Type 1 AGNs. The derived LFs are 3-5 times higher than those of UV/optical-selected AGNs, indicating that Paschen-selected Type 1 AGNs are more complete. In addition, the intrinsic properties of our AGNs show no dependence on dust reddening, suggesting that the observed reddening is unrelated to the central engine and is thus likely caused by line-of-sight obscuration.

astro-ph.GA↗

arXiv2026

Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation

A recent cutting-edge topic in multimodal modeling is to unify visual comprehension and generation within a single model. However, the two tasks demand mismatched decoding regimes and visual representations, making it non-trivial to jointly optimize within a shared feature space. In this work, we present Cheers, a unified multimodal model that decouples patch-level details from semantic representations, thereby stabilizing semantics for multimodal understanding and improving fidelity for image generation via gated detail residuals. Cheers includes three key components: (i) a unified vision tokenizer that encodes and compresses image latent states into semantic tokens for efficient LLM conditioning, (ii) an LLM-based Transformer that unifies autoregressive decoding for text generation and diffusion decoding for image generation, and (iii) a cascaded flow matching head that decodes visual semantics first and then injects semantically gated detail residuals from the vision tokenizer to refine high-frequency content. Experiments on popular benchmarks demonstrate that Cheers matches or surpasses advanced UMMs in both visual understanding and generation. Cheers also achieves 4x token compression, enabling more efficient high-resolution image encoding and generation. Notably, Cheers outperforms the Tar-1.5B on the popular benchmarks GenEval and MMBench, while requiring only 20% of the training cost, indicating effective and efficient (i.e., 4x token compression) unified multimodal modeling. We will release all code and data for future research.

cs.CV↗

arXiv2026

UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations

Current unified multimodal models typically rely on discrete visual tokenizers to bridge the modality gap. However, discretization inevitably discards fine-grained semantic information, leading to suboptimal performance in visual understanding tasks. Conversely, directly modeling continuous semantic representations (e.g., CLIP, SigLIP) poses significant challenges in high-dimensional generative modeling, resulting in slow convergence and training instability. To resolve this dilemma, we introduce UniCom, a unified framework that harmonizes multimodal understanding and generation via compressed continuous representation. We empirically demonstrate that reducing channel dimension is significantly more effective than spatial downsampling for both reconstruction and generation. Accordingly, we design an attention-based semantic compressor to distill dense features into a compact unified representation. Furthermore, we validate that the transfusion architecture surpasses query-based designs in convergence and consistency. Experiments demonstrate that UniCom achieves state-of-the-art generation performance among unified models. Notably, by preserving rich semantic priors, it delivers exceptional controllability in image editing and maintains image consistency even without relying on VAE.

cs.CV↗

arXiv2026

El Agente Cuantico: Automating quantum simulations

Quantum simulation is central to understanding and designing quantum systems across physics and chemistry. Yet it has barriers to access from both computational complexity and computational perspectives, due to the exponential growth of Hilbert space and the complexity of modern software tools. Here we introduce{\cinzel El Agente Cuántico}, a multi-agent AI system that automates quantum-simulation workflows by translating natural-language scientific intent into executed and validated computations across heterogeneous quantum-software frameworks. By reasoning directly over library documentation and APIs, our agentic system dynamically assembles end-to-end simulations spanning state preparation, closed- and open-system dynamics, tensor-network methods, quantum control, quantum error correction, and quantum resource estimation. The developed system unifies traditionally distinct simulation paradigms behind a single natural-language interface. Beyond reducing technical barriers, this approach opens a path toward scalable, adaptive, and increasingly autonomous quantum simulation, enabling faster exploration of physical models, rapid hypothesis testing, and closer integration between theory, simulation, and emerging quantum hardware.

quant-ph↗

arXiv2026

D3MAS: Decompose, Deduce, and Distribute for Enhanced Knowledge Sharing in Multi-Agent Systems

Multi-agent systems powered by large language models exhibit strong capabilities in collaborative problem-solving. However, these systems suffer from substantial knowledge redundancy. Agents duplicate efforts in retrieval and reasoning processes. This inefficiency stems from a deeper issue: current architectures lack mechanisms to ensure agents share minimal sufficient information at each operational stage. Empirical analysis reveals an average knowledge duplication rate of 47.3\% across agent communications. We propose D3MAS (Decompose, Deduce, and Distribute), a hierarchical coordination framework addressing redundancy through structural design rather than explicit optimization. The framework organizes collaboration across three coordinated layers. Task decomposition filters irrelevant sub-problems early. Collaborative reasoning captures complementary inference paths across agents. Distributed memory provides access to non-redundant knowledge. These layers coordinate through structured message passing in a unified heterogeneous graph. This cross-layer alignment ensures information remains aligned with actual task needs. Experiments on four challenging datasets show that D3MAS consistently improves reasoning accuracy by 8.7\% to 15.6\% and reduces knowledge redundancy by 46\% on average.

cs.GR↗