Add to Chrome
Add to Firefox
Add to EdgeSafety Anchor: Defending Harmful Fine-tuning via Geometric Bottlenecks
May 07, 2026
The safety alignment of Large Language Models (LLMs) remains vulnerable to Harmful Fine-tuning (HFT). While existing defenses impose constraints on parameters, gradients, or internal representations, we observe that they can be effectively circumvented under persistent HFT. Our analysis traces this failure to the inherent redundancy of the high-dimensional parameter space: attackers exploit optimization trajectories that are orthogonal to defense constraints to restore harmful capabilities while deceptively adhering to safety restrictions. To address this, we propose Safety Bottleneck Regularization (SBR). SBR shifts the defensive focus from the redundant parameter space to the unembedding layer, which serves as a geometric bottleneck. By anchoring the final hidden states of harmful queries to those of the safety-aligned model, SBR enables the model to maintain safe responses even under persistent HFT. Extensive experiments confirm SBR's effectiveness, demonstrating that utilizing just a single safety anchor is sufficient to reduce the Harmful Score to $<$10 while preserving competitive performance on benign downstream tasks.
PerFlow: Physics-Embedded Rectified Flow for Efficient Reconstruction and Uncertainty Quantification of Spatiotemporal Dynamics
May 05, 2026Reconstructing PDE-governed fields from sparse and irregular measurements is challenging due to their ill-posed nature. Deterministic surrogates are trained on dense fields that struggle with limited measurements and uncertainty quantification. Generative models, by learning distributions over spatiotemporal fields, can better handle sparsity and uncertainty. However, existing generative approaches enforce data consistency and PDE constraints simultaneously via sampling-time gradient guidance, resulting in slow and unstable inference. To this end, we propose PerFlow, a Physics-embedded rectified Flow for efficient sparse reconstruction and uncertainty quantification of spatiotemporal dynamics. PerFlow decouples observation conditioning from physics enforcement, performing guidance-free conditioning by feeding observations into rectified-flow dynamics while embedding hard physics via a constraint-preserving projection (e.g., incompressibility or conservation). Theoretically, we establish invariance guarantees to ensure that trajectories remain on the physics-consistent manifold throughout sampling. Experiments on various PDE systems demonstrate competitive reconstruction accuracy with sound physics consistency, while enabling efficient conditional sampling (e.g., 50 steps) and up to 320 faster inference than 2000-step guided diffusion baselines.
BATON: A Multimodal Benchmark for Bidirectional Automation Transition Observation in Naturalistic Driving
Apr 08, 2026Existing driving automation (DA) systems on production vehicles rely on human drivers to decide when to engage DA while requiring them to remain continuously attentive and ready to intervene. This design demands substantial situational judgment and imposes significant cognitive load, leading to steep learning curves, suboptimal user experience, and safety risks from both over-reliance and delayed takeover. Predicting when drivers hand over control to DA and when they take it back is therefore critical for designing proactive, context-aware HMI, yet existing datasets rarely capture the multimodal context, including road scene, driver state, vehicle dynamics, and route environment. To fill this gap, we introduce BATON, a large-scale naturalistic dataset capturing real-world DA usage across 127 drivers, and 136.6 hours of driving. The dataset synchronizes front-view video, in-cabin video, decoded CAN bus signals, radar-based lead-vehicle interaction, and GPS-derived route context, forming a closed-loop multimodal record around each control transition. We define three benchmark tasks: driving action understanding, handover prediction, and takeover prediction, and evaluate baselines spanning sequence models, classical classifiers, and zero-shot VLMs. Results show that visual input alone is insufficient for reliable transition prediction: front-view video captures road context but not driver state, while in-cabin video reflects driver readiness but not the external scene. Incorporating CAN and route-context signals substantially improves performance over video-only settings, indicating strong complementarity across modalities. We further find takeover events develop more gradually and benefit from longer prediction horizons, whereas handover events depend more on immediate contextual cues, revealing an asymmetry with direct implications for HMI design in assisted driving systems.
ATG-MoE: Autoregressive trajectory generation with mixture-of-experts for assembly skill learning
Mar 19, 2026Flexible manufacturing requires robot systems that can adapt to constantly changing tasks, objects, and environments. However, traditional robot programming is labor-intensive and inflexible, while existing learning-based assembly methods often suffer from weak positional generalization, complex multi-stage designs, and limited multi-skill integration capability. To address these issues, this paper proposes ATG-MoE, an end-to-end autoregressive trajectory generation method with mixture of experts for assembly skill learning from demonstration. The proposed method establishes a closed-loop mapping from multi-modal inputs, including RGB-D observations, natural language instructions, and robot proprioception to manipulation trajectories. It integrates multi-modal feature fusion for scene and task understanding, autoregressive sequence modeling for temporally coherent trajectory generation, and a mixture-of-experts architecture for unified multi-skill learning. In contrast to conventional methods that separate visual perception and control or train different skills independently, ATG-MoE directly incorporates visual information into trajectory generation and supports efficient multi-skill integration within a single model. We train and evaluate the proposed method on eight representative assembly skills from a pressure-reducing valve assembly task. Experimental results show that ATG-MoE achieves strong overall performance in simulation, with an average grasp success rate of 96.3% and an average overall success rate of 91.8%, while also demonstrating strong generalization and effective multi-skill integration. Real-world experiments further verify its practicality for multi-skill industrial assembly. The project page can be found at https://hwh23.github.io/ATG-MoE
Recover to Predict: Progressive Retrospective Learning for Variable-Length Trajectory Prediction
Mar 11, 2026Trajectory prediction is critical for autonomous driving, enabling safe and efficient planning in dense, dynamic traffic. Most existing methods optimize prediction accuracy under fixed-length observations. However, real-world driving often yields variable-length, incomplete observations, posing a challenge to these methods. A common strategy is to directly map features from incomplete observations to those from complete ones. This one-shot mapping, however, struggles to learn accurate representations for short trajectories due to significant information gaps. To address this issue, we propose a Progressive Retrospective Framework (PRF), which gradually aligns features from incomplete observations with those from complete ones via a cascade of retrospective units. Each unit consists of a Retrospective Distillation Module (RDM) and a Retrospective Prediction Module (RPM), where RDM distills features and RPM recovers previous timesteps using the distilled features. Moreover, we propose a Rolling-Start Training Strategy (RSTS) that enhances data efficiency during PRF training. PRF is plug-and-play with existing methods. Extensive experiments on datasets Argoverse 2 and Argoverse 1 demonstrate the effectiveness of PRF. Code is available at https://github.com/zhouhao94/PRF.
CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation
Feb 27, 2026GPU kernel optimization is fundamental to modern deep learning but remains a highly specialized task requiring deep hardware expertise. Despite strong performance in general programming, large language models (LLMs) remain uncompetitive with compiler-based systems such as torch.compile for CUDA kernel generation. Existing CUDA code generation approaches either rely on training-free refinement or fine-tune models within fixed multi-turn execution-feedback loops, but both paradigms fail to fundamentally improve the model's intrinsic CUDA optimization ability, resulting in limited performance gains. We present CUDA Agent, a large-scale agentic reinforcement learning system that develops CUDA kernel expertise through three components: a scalable data synthesis pipeline, a skill-augmented CUDA development environment with automated verification and profiling to provide reliable reward signals, and reinforcement learning algorithmic techniques enabling stable training. CUDA Agent achieves state-of-the-art results on KernelBench, delivering 100\%, 100\%, and 92\% faster rate over torch.compile on KernelBench Level-1, Level-2, and Level-3 splits, outperforming the strongest proprietary models such as Claude Opus 4.5 and Gemini 3 Pro by about 40\% on the hardest Level-3 setting.
Large Language Model-Assisted UAV Operations and Communications: A Multifaceted Survey and Tutorial
Feb 23, 2026Uncrewed Aerial Vehicles (UAVs) are widely deployed across diverse applications due to their mobility and agility. Recent advances in Large Language Models (LLMs) offer a transformative opportunity to enhance UAV intelligence beyond conventional optimization-based and learning-based approaches. By integrating LLMs into UAV systems, advanced environmental understanding, swarm coordination, mobility optimization, and high-level task reasoning can be achieved, thereby allowing more adaptive and context-aware aerial operations. This survey systematically explores the intersection of LLMs and UAV technologies and proposes a unified framework that consolidates existing architectures, methodologies, and applications for UAVs. We first present a structured taxonomy of LLM adaptation techniques for UAVs, including pretraining, fine-tuning, Retrieval-Augmented Generation (RAG), and prompt engineering, along with key reasoning capabilities such as Chain-of-Thought (CoT) and In-Context Learning (ICL). We then examine LLM-assisted UAV communications and operations, covering navigation, mission planning, swarm control, safety, autonomy, and network management. After that, the survey further discusses Multimodal LLMs (MLLMs) for human-swarm interaction, perception-driven navigation, and collaborative control. Finally, we address ethical considerations, including bias, transparency, accountability, and Human-in-the-Loop (HITL) strategies, and outline future research directions. Overall, this work positions LLM-assisted UAVs as a foundation for intelligent and adaptive aerial systems.
Enhancing Large Language Models (LLMs) for Telecom using Dynamic Knowledge Graphs and Explainable Retrieval-Augmented Generation
Feb 19, 2026Large language models (LLMs) have shown strong potential across a variety of tasks, but their application in the telecom field remains challenging due to domain complexity, evolving standards, and specialized terminology. Therefore, general-domain LLMs may struggle to provide accurate and reliable outputs in this context, leading to increased hallucinations and reduced utility in telecom operations.To address these limitations, this work introduces KG-RAG-a novel framework that integrates knowledge graphs (KGs) with retrieval-augmented generation (RAG) to enhance LLMs for telecom-specific tasks. In particular, the KG provides a structured representation of domain knowledge derived from telecom standards and technical documents, while RAG enables dynamic retrieval of relevant facts to ground the model's outputs. Such a combination improves factual accuracy, reduces hallucination, and ensures compliance with telecom specifications.Experimental results across benchmark datasets demonstrate that KG-RAG outperforms both LLM-only and standard RAG baselines, e.g., KG-RAG achieves an average accuracy improvement of 14.3% over RAG and 21.6% over LLM-only models. These results highlight KG-RAG's effectiveness in producing accurate, reliable, and explainable outputs in complex telecom scenarios.
Quark Medical Alignment: A Holistic Multi-Dimensional Alignment and Collaborative Optimization Paradigm
Feb 12, 2026While reinforcement learning for large language model alignment has progressed rapidly in recent years, transferring these paradigms to high-stakes medical question answering reveals a fundamental paradigm mismatch. Reinforcement Learning from Human Feedback relies on preference annotations that are prohibitively expensive and often fail to reflect the absolute correctness of medical facts. Reinforcement Learning from Verifiable Rewards lacks effective automatic verifiers and struggles to handle complex clinical contexts. Meanwhile, medical alignment requires the simultaneous optimization of correctness, safety, and compliance, yet multi-objective heterogeneous reward signals are prone to scale mismatch and optimization conflicts.To address these challenges, we propose a robust medical alignment paradigm. We first construct a holistic multi-dimensional medical alignment matrix that decomposes alignment objectives into four categories: fundamental capabilities, expert knowledge, online feedback, and format specifications. Within each category, we establish a closed loop of where observable metrics inform attributable diagnosis, which in turn drives optimizable rewards, thereby providing fine-grained, high-resolution supervision signals for subsequent iterative optimization. To resolve gradient domination and optimization instability problem caused by heterogeneous signals, we further propose a unified optimization mechanism. This mechanism employs Reference-Frozen Normalization to align reward scales and implements a Tri-Factor Adaptive Dynamic Weighting strategy to achieve collaborative optimization that is weakness-oriented, risk-prioritized, and redundancy-reducing. Experimental results demonstrate the effectiveness of our proposed paradigm in real-world medical scenario evaluations, establishing a new paradigm for complex alignment in vertical domains.
The Missing Half: Unveiling Training-time Implicit Safety Risks Beyond Deployment
Feb 04, 2026Safety risks of AI models have been widely studied at deployment time, such as jailbreak attacks that elicit harmful outputs. In contrast, safety risks emerging during training remain largely unexplored. Beyond explicit reward hacking that directly manipulates explicit reward functions in reinforcement learning, we study implicit training-time safety risks: harmful behaviors driven by a model's internal incentives and contextual background information. For example, during code-based reinforcement learning, a model may covertly manipulate logged accuracy for self-preservation. We present the first systematic study of this problem, introducing a taxonomy with five risk levels, ten fine-grained risk categories, and three incentive types. Extensive experiments reveal the prevalence and severity of these risks: notably, Llama-3.1-8B-Instruct exhibits risky behaviors in 74.4% of training runs when provided only with background information. We further analyze factors influencing these behaviors and demonstrate that implicit training-time risks also arise in multi-agent training settings. Our results identify an overlooked yet urgent safety challenge in training.