乐于分享
好东西不私藏
当前位置：夜雨聆风 > 技术教程 > 软件教程 > AI前沿论文TOP10(2026年4月22日)

AI前沿论文TOP10(2026年4月22日)

当前时间： 2026-04-22 17:07:04 更新时间： 2026-04-22 分类：软件教程 评论(0)

AI前沿论文TOP10(2026年4月22日)

AI 前沿论文 TOP10（2026年4月22日）

一、Large Language Model

1. Your Language Model Secretly Contains Personality Subnetworks

  • 作者： Y. Chen et al.
  • 发表： arXiv 2026-02-24 (ICLR 2026)
  • 中文摘要： 人类根据社交情境切换不同人格，LLM 同样展现出灵活的人格适应能力。以往方法通常依赖外部知识（提示、RAG 或微调）来调整行为。本文首次揭示 LLM 参数空间中天然存在人格专门化子网络。通过小规模校准数据集识别不同人格对应的激活特征，开发一种无需训练的掩码策略分离轻量人格子网络，并提出对比剪枝策略增强对立人格的二分类场景中的分离效果。实验表明，该方法无需外部知识即可实现更强的人格对齐，同时更高效，证明多样化人类行为已嵌入模型参数空间，为可控可解释的 LLM 个性化提供了新视角。
  • 英文摘要： Humans shift between different personas depending on social context. Large Language Models (LLMs) demonstrate a similar flexibility in adopting different personas and behaviors. Existing approaches, however, typically adapt such behavior through external knowledge such as prompting, retrieval-augmented generation (RAG), or fine-tuning. We ask: do LLMs really need external context or parameters to adapt to different behaviors, or do they already have such knowledge embedded in their parameters? In this work, we show that LLMs already contain persona-specialized subnetworks in their parameter space. Using small calibration datasets, we identify distinct activation signatures associated with different personas. Guided by these statistics, we develop a masking strategy that isolates lightweight persona subnetworks. Building on the findings, we further discuss: how can we discover opposing subnetwork from the model that lead to binary-opposing personas, such as introvert-extrovert? To further enhance separation in binary opposition scenarios, we introduce a contrastive pruning strategy that identifies parameters responsible for the statistical divergence between opposing personas. Our method is entirely training-free and relies solely on the language model’s existing parameter space. Across diverse evaluation settings, the resulting subnetworks exhibit significantly stronger persona alignment than baselines that require external knowledge while being more efficient. Our findings suggest that diverse human-like behaviors are not merely induced in LLMs, but are already embedded in their parameter space, pointing toward a new perspective on controllable and interpretable personalization in large language models.
  • 链接： https://arxiv.org/abs/2602.07164

2. Continual Learning in Large Language Models: Methods, Challenges, and Opportunities

  • 作者： Zhongwu Sun et al.
  • 发表： arXiv 2026-03-14
  • 中文摘要： 持续学习旨在使模型不断适应新数据同时保留已有知识，是 LLM 走向终身智能的关键能力。本文系统梳理了 LLM 持续学习的最新进展，从方法、挑战到未来机遇进行了全面综述。内容涵盖持续学习的核心问题（灾难性遗忘、知识迁移等）、主流技术路线（参数隔离、正则化、记忆回放等）及其在 LLM 时代的演变，并讨论了数据稀缺、评估标准缺失、计算成本等现实挑战，为该领域后续研究提供了清晰的研究图谱和发展方向。
  • 英文摘要： (Continual learning aims to enable models to continuously adapt to new data while retaining previously acquired knowledge, representing a key capability toward lifelong intelligence for LLMs. This paper systematically reviews the latest advances in continual learning for LLMs, covering methods, challenges, and future opportunities. The survey addresses core problems including catastrophic forgetting and knowledge transfer, reviews mainstream technical approaches such as parameter isolation, regularization, and memory replay, and discusses practical challenges such as data scarcity, lack of evaluation benchmarks, and computational costs.)
  • 链接： https://arxiv.org/abs/2603.12658

二、Multimodal / Vision Language

3. GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning

  • 作者： Jiajin Liu et al.
  • 发表： arXiv 2026-03-09 (CVPR 2026)
  • 中文摘要： 视觉-语言模型（VLM）在多模态对齐与理解方面表现卓越，但其对结构化数据（多模态实体通过显式关系图连接）进行推理的潜力尚待充分探索。本文提出 GraphVLM，首个系统性评估 VLM 多模态图学习（MMGL）能力的基准。GraphVLM 研究了三种互补范式：VLM-as-Encoder（通过多模态特征融合增强图神经网络）、VLM-as-Aligner（在潜在空间或语言空间桥接模态以支持结构化推理）、VLM-as-Predictor（直接用 VLM 作为多模态骨干进行图学习）。在六个跨领域数据集上的广泛实验表明，VLM-as-Predictor 模式取得最显著且一致的性能提升，揭示了 VLM 作为多模态图学习新基础的巨大潜力。
  • 英文摘要： Vision-Language Models (VLMs) have demonstrated remarkable capabilities in aligning and understanding multimodal signals, yet their potential to reason over structured data, where multimodal entities are connected through explicit relational graphs, remains largely underexplored. Unlocking this capability is crucial for real-world applications such as social networks, recommendation systems, and scientific discovery, where multimodal information is inherently structured. To bridge this gap, we present GraphVLM, a systematic benchmark designed to evaluate and harness the capabilities of VLMs for multimodal graph learning (MMGL). GraphVLM investigates three complementary paradigms for integrating VLMs with graph reasoning: (1) VLM-as-Encoder, which enriches graph neural networks through multimodal feature fusion; (2) VLM-as-Aligner, which bridges modalities in latent or linguistic space to facilitate LLM-based structured reasoning; and (3) VLM-as-Predictor, which directly employs VLMs as multimodal backbones for graph learning tasks. Extensive experiments across six datasets from diverse domains demonstrate that VLMs enhance multimodal graph learning via all three roles. Among these paradigms, VLM-as-Predictor achieves the most substantial and consistent performance gains, revealing the untapped potential of vision-language models as a new foundation for multimodal graph learning.
  • 链接： https://arxiv.org/abs/2603.13370

4. Incentivizing DeepResearch Capability in Multimodal Large Language Models

  • 作者： Wenxuan Huang, Wanli Ouyang et al.
  • 发表： arXiv 2026-01-09
  • 中文摘要： 多模态大语言模型（MLLM）在广泛视觉任务中成效显著，但受限于内部世界知识的容量，现有方法通过”推理-然后-调用工具”模式访问视觉和文本搜索引擎以获取外部事实。然而这些方法通常将多模态搜索定义在过于简单的场景中，假设单一全图或实体级图像查询加少量文本查询足以提取关键证据，这在现实场景（大量视觉噪声）中并不成立。此外，现有方法在推理深度和搜索广度上存在局限，难以解决需要聚合多样化视觉证据的复杂问题。本文提出新方法，通过增强 MLLM 的深度研究能力，使其能够在复杂视觉问答场景中进行更深入、更广泛的多模态信息检索与推理。
  • 英文摘要： Multimodal large language models (MLLMs) have achieved remarkable success across a broad range of vision tasks. However, constrained by the capacity of their internal world knowledge, prior work has proposed augmenting MLLMs by reasoning-then-tool-call for visual and textual search engines to obtain substantial gains on tasks requiring extensive factual information. However, these approaches typically define multimodal search in a naive setting, assuming that a single full-level or entity-level image query and few text query suffices to retrieve the key evidence needed to answer the question, which is unrealistic in real-world scenarios with substantial visual noise. Moreover, they are often limited in the reasoning depth and search breadth, making it difficult to solve complex questions that require aggregating evidence from diverse visual evidence.
  • 链接： https://arxiv.org/abs/2601.22060

三、Reinforcement Learning

5. LongRLVR: Long-Context Reinforcement Learning Requires Verifiable Context Rewards

  • 作者： Guanzheng Chen et al.
  • 发表： arXiv 2026-03-02 (ICLR 2026)
  • 中文摘要： RLVR（可验证奖励的强化学习）通过优化事实结果显著提升了 LLM 的推理能力，但在需要上下文锚定（从外部信息中查找和推理）长上下文场景中表现不佳。本文揭示了根本原因：仅基于最终答案的奖励过于稀疏，无法有效引导模型识别相关证据。理论上证明了仅结果奖励会导致上下文锚定过程的梯度消失，使学习不可行。为解决这一瓶颈，LongRLVR 引入了一种密集且可验证的上下文奖励来增强稀疏的答案奖励，该辅助信号直接激励模型选择正确的锚定信息，提供稳健的学习梯度。在 Qwen 和 LLaMA 模型上的长上下文基准测试中，LongRLVR 一致且显著地超越了标准 RLVR，例如将 14B 模型在 RULER-QA 上的分数从 73.17 提升至 88.90，LongBench v2 从 39.8 提升至 46.5。
  • 英文摘要： Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs) by optimizing them against factual outcomes. However, this paradigm falters in long-context scenarios, as its reliance on internal parametric knowledge is ill-suited for tasks requiring contextual grounding–the ability to find and reason over externally provided information. We identify a key reason for this failure: a reward based solely on the final answer is too sparse to effectively guide the model for identifying relevant evidence. We formally prove that the outcome-only reward leads to significant vanishing gradients for the context grounding process, rendering learning intractable. To overcome this bottleneck, we introduce LongRLVR to augment the sparse answer reward with a dense and verifiable context reward. This auxiliary signal directly incentivizes the model for selecting the correct grounding information, providing a robust learning gradient that solves the underlying optimization challenge. We validate our method on challenging long-context benchmarks using Qwen and LLaMA models. LongRLVR consistently and significantly outperforms the standard RLVR across all models and benchmarks, e.g., boosting a 14B model’s scores on RULER-QA from 73.17 to 88.90 and on LongBench v2 from 39.8 to 46.5. Our work demonstrates that explicitly rewarding the grounding process is a critical and effective strategy for unlocking the full reasoning potential of LLMs in long-context applications.
  • 链接： https://arxiv.org/abs/2603.02146

6. Evolutionary Discovery of Reinforcement Learning Algorithms via Large Language Models

  • 作者： Alkis Sygkounas et al.
  • 发表： arXiv 2026-03-30 (GECCO 2026)
  • 中文摘要： 强化学习算法通常由人工设计且固定不变。本文提出一种进化框架，直接在可执行更新规则的空间中搜索完整的训练程序来自动发现 RL 算法。该方法基于 REvolve 进化系统，利用 LLM 作为生成变异算子，将搜索范围从奖励函数发现扩展到算法发现。为促进非标准学习规则的涌现，搜索排除了典型机制（如 Actor-Critic 结构、时序差分损失、值 bootstrapping）。由于 RL 算法对内部标量参数高度敏感，本文还引入了后进化精调阶段，由 LLM 为每个进化出的更新规则提出可行的超参数范围。在多个 Gymnasium 基准上通过完整训练运行进行端到端评估，发现的算法与 SAC、PPO、DQN、A2C 等成熟基线相比具有竞争力。
  • 英文摘要： Reinforcement learning algorithms are defined by their learning update rules, which are typically hand-designed and fixed. We present an evolutionary framework for discovering reinforcement learning algorithms by searching directly over executable update rules that implement complete training procedures. The approach builds on REvolve, an evolutionary system that uses large language models as generative variation operators, and extends it from reward-function discovery to algorithm discovery. To promote the emergence of nonstandard learning rules, the search excludes canonical mechanisms such as actor-critic structures, temporal-difference losses, and value bootstrapping. Because reinforcement learning algorithms are highly sensitive to internal scalar parameters, we introduce a post-evolution refinement stage in which a large language model proposes feasible hyperparameter ranges for each evolved update rule. Evaluated end-to-end by full training runs on multiple Gymnasium benchmarks, the discovered algorithms achieve competitive performance relative to established baselines, including SAC, PPO, DQN, and A2C.
  • 链接： https://arxiv.org/abs/2603.28416

四、World Model / Video Generation

7. A Mechanistic View on Video Generation as World Models: State and Dynamics

  • 作者： Luozhou Wang, Ying-Cong Chen et al.
  • 发表： arXiv 2026-01-08
  • 中文摘要： 大规模视频生成模型展现出新兴的物理一致性，被认为是潜在的世界模型。然而，当代”无状态”视频架构与经典以状态为中心的世界模型理论之间仍存在显著差距。本文通过提出以两大支柱为中心的新分类法来弥合这一差距：状态构建（State Construction）和动力学建模（Dynamics Modeling）。状态构建分为隐式范式（上下文管理）和显式范式（潜在压缩），动力学建模通过知识集成和架构重构进行分析。此外，文章呼吁评估范式从视觉保真度转向功能基准测试，考查物理持续性和因果推理能力。最后识别了两个关键前沿：通过数据驱动记忆和压缩保真度增强持续性，以及通过潜在因子解耦和推理先验集成推进因果性。
  • 英文摘要： Large-scale video generation models have demonstrated emergent physical coherence, positioning them as potential world models. However, a gap remains between contemporary “stateless” video architectures and classic state-centric world model theories. This work bridges this gap by proposing a novel taxonomy centered on two pillars: State Construction and Dynamics Modeling. We categorize state construction into implicit paradigms (context management) and explicit paradigms (latent compression), while dynamics modeling is analyzed through knowledge integration and architectural reformulation. Furthermore, we advocate for a transition in evaluation from visual fidelity to functional benchmarks, testing physical persistence and causal reasoning. We conclude by identifying two critical frontiers: enhancing persistence via data-driven memory and compressed fidelity, and advancing causality through latent factor decoupling and reasoning-prior integration. By addressing these challenges, the field can evolve from generating visually plausible videos to building robust, general-purpose world models.
  • 链接： https://arxiv.org/abs/2601.17067

8. DreamWorld: Unified World Modeling in Video Generation

  • 作者： Shaofeng Zhang et al.
  • 发表： arXiv 2026-02-28
  • 中文摘要： 尽管视频生成取得了令人瞩目的进展，现有模型仍局限于表面 plausibility，缺乏对世界的连贯统一理解。以往方法通常只引入单一形式的世界相关知识或依赖刚性对齐策略来整合额外知识。然而，仅对齐单一世界知识不足以构成需要联合建模多个异质维度（如物理常识、3D 和时序一致性）的世界模型。本文提出 DreamWorld，通过联合世界建模范式将互补的世界知识整合到视频生成器中，联合预测视频像素和基础模型的特征以捕捉时序动态、空间几何和语义一致性。DreamWorld 提出一致性约束退火（CCA）在训练过程中渐进调节世界级约束，并提出多源内部引导（Multi-Source Inner-Guidance）在推理时强制执行学习到的世界先验。在 VBench 上相比 Wan2.1 提升 2.26 分。
  • 英文摘要： Despite impressive progress in video generation, existing models remain limited to surface-level plausibility, lacking a coherent and unified understanding of the world. Prior approaches typically incorporate only a single form of world-related knowledge or rely on rigid alignment strategies to introduce additional knowledge. However, aligning the single world knowledge is insufficient to constitute a world model that requires jointly modeling multiple heterogeneous dimensions (e.g., physical commonsense, 3D and temporal consistency). To address this limitation, we introduce DreamWorld, a unified framework that integrates complementary world knowledge into video generators via a Joint World Modeling Paradigm, jointly predicting video pixels and features from foundation models to capture temporal dynamics, spatial geometry, and semantic consistency. However, naively optimizing these heterogeneous objectives can lead to visual instability and temporal flickering. To mitigate this issue, we propose Consistent Constraint Annealing (CCA) to progressively regulate world-level constraints during training, and Multi-Source Inner-Guidance to enforce learned world priors at inference. Extensive evaluations show that DreamWorld improves world consistency, outperforming Wan2.1 by 2.26 points on VBench.
  • 链接： https://arxiv.org/abs/2603.00466

五、AI Agent

9. SpecOps: A Fully Automated AI Agent Testing Framework in Real-World GUI Environments

  • 作者： Syed Yusuf Ahmed et al.
  • 发表： arXiv 2026-03-17 (ICSE 2026)
  • 中文摘要： 由大语言模型驱动的自主 AI Agent 正在被部署到现实应用中，可靠和稳健的行为至关重要。然而，现有 Agent 评估框架要么严重依赖人工，要么在模拟环境中运行，或缺乏对复杂多模态真实世界 Agent 的测试聚焦。本文提出 SpecOps，一个全新的全自动测试框架，专为真实 GUI 环境中的基于 GUI 的 AI Agent 设计。SpecOps 将测试过程分解为四个专门阶段——测试用例生成、环境设置、测试执行和验证——每个阶段由不同的基于 LLM 的专业 Agent 处理。该架构解决了端到端任务一致性、鲁棒错误处理和跨多样化 Agent 平台（CLI 工具、Web 应用、浏览器扩展）适应性等关键挑战。在五个真实 Agent 的综合评估中，SpecOps 在规划准确性、执行成功率和 Bug 检测有效性上均优于基线（包括 AutoGPT 等通用 Agent 系统），以不到 0.73 美元的成本和不到 8 分钟的运行时间识别了 164 个真实 Bug（F1 分数 0.89）。
  • 英文摘要： Autonomous AI agents powered by large language models (LLMs) are increasingly deployed in real-world applications, where reliable and robust behavior is critical. However, existing agent evaluation frameworks either rely heavily on manual efforts, operate within simulated environments, or lack focus on testing complex, multimodal, real-world agents. We introduce SpecOps, a novel, fully automated testing framework designed to evaluate GUI-based AI agents in real-world environments. SpecOps decomposes the testing process into four specialized phases – test case generation, environment setup, test execution, and validation – each handled by a distinct LLM-based specialist agent. This structured architecture addresses key challenges including end-to-end task coherence, robust error handling, and adaptability across diverse agent platforms including CLI tools, web apps, and browser extensions. In comprehensive evaluations across five diverse real-world agents, SpecOps outperforms baselines including general-purpose agentic systems such as AutoGPT and LLM-crafted automation scripts in planning accuracy, execution success, and bug detection effectiveness. SpecOps identifies 164 true bugs in the real-world agents with an F1 score of 0.89. With a cost of under 0.73 USD and a runtime of under eight minutes per test, it demonstrates its practical viability and superiority in automated, real-world agent testing.
  • 链接： https://arxiv.org/abs/2603.10268

10. The Attack and Defense Landscape of Agentic AI: A Comprehensive Survey

  • 作者： Juhee Kim et al.
  • 发表： arXiv 2026-03-11 (USENIX Security 2026)
  • 中文摘要： 将大语言模型与非 AI 系统组件相结合的 AI Agent 正在现实应用中快速兴起，提供了前所未有的自动化和灵活性。然而，这种前所未有的灵活性引入了与传统软件系统根本不同的复杂安全挑战。本文首次对 AI Agent 安全进行了系统全面的综述，包括 AI Agent 系统的设计空间、攻击面和防御机制分析。通过案例研究指出了现有 Agentic AI 系统安全保护的差距，并确定了该新兴领域的开放挑战。本文还首次提出了理解 AI Agent 安全风险和防御策略的系统化框架，为构建安全的 Agentic 系统和推进该关键领域的研究奠定了基础。
  • 英文摘要： AI agents that combine large language models with non-AI system components are rapidly emerging in real-world applications, offering unprecedented automation and flexibility. However, this unprecedented flexibility introduces complex security challenges fundamentally different from those in traditional software systems. This paper presents the first systematic and comprehensive survey of AI agent security, including an analysis of the design space, attack landscape, and defense mechanisms for secure AI agent systems. We further conduct case studies to point out existing gaps in securing agentic AI systems and identify open challenges in this emerging domain. Our work also introduces the first systematic framework for understanding the security risks and defense strategies of AI agents, serving as a foundation for building both secure agentic systems and advancing research in this critical area.
  • 链接： https://arxiv.org/abs/2603.11088

本报告由 Euler 基于网络搜索整理 | 2026年4月22日