safety

Position: Universal Aesthetic Alignment Narrows Artistic Expression

arXiv:2512.11883v3 Announce Type: replace-cross Abstract: Over-aligning image generation models to a generalized aesthetic preference conflicts with user intent, particularly when 'anti-aesthetic' out

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Post-ADC Inference: Valid Inference After Active Data Collection

DGX agent

arXiv:2605.11511v1 Announce Type: cross Abstract: The validity of statistical inference depends critically on how data are collected. When data gathered through active data collection (ADC) are reused

safetyarxiv-cs-lg

13 May 2026

Predictive Maps of Multi-Agent Reasoning: A Successor-Representation Spectrum for LLM Communication Topologies

arXiv:2605.11453v1 Announce Type: cross Abstract: Practitioners deploying multi-agent large language model (LLM) systems must currently choose between communication topologies such as chain, star, mes

Pretraining Exposure Explains Popularity Judgments in Large Language Models

arXiv:2605.12382v1 Announce Type: new Abstract: Large language models (LLMs) exhibit systematic preferences for well-known entities, a phenomenon often attributed to popularity bias. However, the exte

Primal-Dual Policy Optimization for Linear CMDPs with Adversarial Losses

arXiv:2605.11535v1 Announce Type: new Abstract: Existing work on linear constrained Markov decision processes (CMDPs) has primarily focused on stochastic settings, where the losses and costs are eithe

Primal Generation, Dual Judgment: Self-Training from Test-Time Scaling

arXiv:2605.11299v1 Announce Type: cross Abstract: Code generation is typically trained in the primal space of programs: a model produces a candidate solution and receives sparse execution feedback, of

PriorZero: Bridging Language Priors and World Models for Decision Making

arXiv:2605.12289v1 Announce Type: new Abstract: Leveraging the rich world knowledge of Large Language Models (LLMs) to enhance Reinforcement Learning (RL) agents offers a promising path toward general

Probabilistic Modeling of Latent Agentic Substructures in Deep Neural Networks

arXiv:2509.06701v2 Announce Type: replace Abstract: We develop a theory of intelligent agency grounded in probabilistic modeling for neural models. Agents are represented as outcome distributions with

probably correct, from @polynoamial: “with today’s AI models, intelligence is a function of inference compute.” but what about tomorrow’s mo…

probably correct, from @polynoamial: “with today’s AI models, intelligence is a function of inference compute.” but what about tomorrow’s models? never forget that humans are remarkably intelligent (t

safetygary-marcus--x

Prototype Fusion: A Training-Free Multi-Layer Approach to OOD Detection

arXiv:2603.23677v2 Announce Type: replace Abstract: Deep learning models are increasingly deployed in safety-critical applications, where reliable out-of-distribution (OOD) detection is essential to e

Question Difficulty Estimation for Large Language Models via Answer Plausibility Scoring

arXiv:2605.12398v1 Announce Type: new Abstract: Estimating question difficulty is a critical component in evaluating and improving large language models (LLMs) for question answering (QA). Existing ap

Quotient-Categorical Representations for Bellman-Compatible Average-Reward Distributional Reinforcement Learning

arXiv:2605.11289v1 Announce Type: new Abstract: Average-reward reinforcement learning requires estimating the gain and the bias, which is defined only up to an additive constant. This makes direct dis

RACC: Representation-Aware Coverage Criteria for LLM Safety Testing

arXiv:2602.02280v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) face severe safety risks from jailbreak attacks, yet current safety testing largely relies on static datasets and

Rainbow Deep Q-Learning with Kinematics-Aware Design for Cooperative Delta and 3-RRS Parallel Robot Insertion

arXiv:2605.11697v1 Announce Type: new Abstract: This paper presents a kinematics-aware deep reinforcement learning framework based on Rainbow Deep Q-Networks (DQN) for cooperative peg-in-hole manipula

RankQ: Offline-to-Online Reinforcement Learning via Self-Supervised Action Ranking

arXiv:2605.11151v1 Announce Type: cross Abstract: Offline-to-online reinforcement learning (RL) improves sample efficiency by leveraging pre-collected datasets prior to online interaction. A key chall

Real-Scale Island Area and Coastline Estimation using Only its Place Name or Coordinates

arXiv:2605.11267v1 Announce Type: new Abstract: Accurate measurement of island area and coastline length is crucial for coastal zone monitoring and oceanographic analysis. However, traditional measure

Red-Teaming Text-to-Image Models via In-Context Experience Replay and Semantic-Preserving Prompt Rewriting

arXiv:2411.16769v3 Announce Type: replace-cross Abstract: Understanding the capabilities of text-to-image (T2I) models in harmful content generation is essential to safety and compliance. However, hum

REFNet++: Multi-Task Efficient Fusion of Camera and Radar Sensor Data in Bird's-Eye Polar View

arXiv:2605.11824v1 Announce Type: new Abstract: A realistic view of the vehicle's surroundings is generally offered by camera sensors, which is crucial for environmental perception. Affordable radar s

Rethink the Role of Neural Decoders in Quantum Error Correction

arXiv:2605.12046v1 Announce Type: cross Abstract: Quantum error correction (QEC) is essential for enabling quantum advantages, with decoding as a central algorithmic primitive. Owing to its importance

Rethinking external validation for the target population: Capturing patient-level similarity with a generative model

arXiv:2605.11284v1 Announce Type: cross Abstract: Background: External validation is essential for assessing the transportability of predictive models. However, its interpretation is often confounded

RIO: Flexible Real-Time Robot I/O for Cross-Embodiment Robot Learning

arXiv:2605.11564v1 Announce Type: new Abstract: Despite recent efforts to collect multi-task, multi-embodiment datasets, to design recipes for training Vision-Language-Action models (VLAs), and to sho

Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter

arXiv:2605.11685v1 Announce Type: new Abstract: Large language model (LLM) unlearning aims to remove specific data influences from pre-trained model without costly retraining, addressing privacy, copy

Robust Multi-Agent Path Finding under Observation Attacks: A Principled Adversarial-Plus-Smoothing Training Recipe

arXiv:2605.11469v1 Announce Type: new Abstract: Decentralized multi-agent path finding (MAPF) routes a team of agents on a shared grid, each acting from its own local view. The standard solution train

Robust Policy Optimization to Prevent Catastrophic Forgetting

arXiv:2602.08813v2 Announce Type: replace Abstract: Large language models are commonly trained through multi-stage post-training: first via RLHF, then fine-tuned for other downstream objectives. Yet e

Robustness Certificates for Neural Networks against Adversarial Attacks

arXiv:2512.20865v2 Announce Type: replace Abstract: The increasing use of machine learning in safety-critical domains amplifies the risk of adversarial threats, especially data poisoning attacks that

Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection

arXiv:2602.07892v2 Announce Type: replace-cross Abstract: Safety post-training can improve the harmfulness and policy compliance of Large Language Models (LLMs), but it may also reduce general utility

Safety-Oriented Evaluation of Language Understanding Systems for Air Traffic Control

arXiv:2605.11769v1 Announce Type: new Abstract: Air Traffic Control (ATC) is a safety-critical domain in which incorrect interpretation of instructions may lead to severe operational consequences. Whi

SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning

arXiv:2512.00775v2 Announce Type: replace Abstract: Linear Temporal Logic (LTL) provides a rigorous framework for specifying long-horizon robotic tasks, yet existing approaches face a trade-off: model

Sequential Off-Policy Learning with Logarithmic Smoothing

arXiv:2506.10664v2 Announce Type: replace-cross Abstract: Off-policy learning enables training policies from logged interaction data. Most prior work considers the batch setting, where a policy is lea

SEVO: Semantic-Enhanced Virtual Observation for Robust VLA Manipulation via Active Illumination and Data-Centric Collection

arXiv:2605.11114v1 Announce Type: new Abstract: Vision-Language-Action (VLA) and imitation-learning policies trained via community toolchains on low-cost hardware frequently fail when deployed outside

SI-Diff: A Framework for Learning Search and High-Precision Insertion with a Force-Domain Diffusion Policy

arXiv:2605.12247v1 Announce Type: new Abstract: Contact-rich assembly is fundamental in robotics but poses significant challenges due to uncertainties in relative poses, such as misalignments and smal

Simpson's Paradox in Behavioral Curves: How Aggregation Distorts Parametric Models of User Dynamics

arXiv:2605.11017v1 Announce Type: new Abstract: Behavioral curve modeling -- fitting parametric functions to engagement-versus-exposure data -- is standard practice in recommendation, advertising, and

Simulation Distillation: Pretraining World Models in Simulation for Rapid Real-World Adaptation

arXiv:2603.15759v2 Announce Type: replace-cross Abstract: Robot learning requires adaptation methods that improve reliably from limited, mixed-quality interaction data. This is especially challenging

Simulation-Ready Cluttered Scene Estimation via Physics-aware Joint Shape and Pose Optimization

arXiv:2602.20150v2 Announce Type: replace-cross Abstract: Estimating simulation-ready scenes from real-world observations is crucial for downstream planning and policy learning tasks. Regretfully, exi

SkillGraph: Skill-Augmented Reinforcement Learning for Agents via Evolving Skill Graphs

arXiv:2605.12039v1 Announce Type: new Abstract: Skill libraries enable large language model agents to reuse experience from past interactions, but most existing libraries store skills as isolated entr

Smart moves: Building resilient transportation systems with Google AI

What does transportation mean to you? For some, it’s making sure the train is on schedule so they can get to work on time. Maybe it’s making sure you have time connecting between flights. Maybe it’s a

safetygoogle-cloud-ai

Space Syntax-guided Post-training for Residential Floor Plan Generation

arXiv:2602.22507v2 Announce Type: replace-cross Abstract: Residential floor plan generation requires not only geometric fidelity but also spatial configurational logic: shared living spaces should be

Sparse Offline Reinforcement Learning with Corruption Robustness

arXiv:2512.24768v3 Announce Type: replace-cross Abstract: We investigate robustness to strong data corruption in offline sparse reinforcement learning (RL). In our setting, an adversary may arbitraril

Sparsity and Out-of-Distribution Generalization

arXiv:2603.07388v2 Announce Type: replace Abstract: Explaining out-of-distribution generalization has been a central problem in epistemology since Goodman's 'grue' puzzle in 1946. Today it's a central

Spectral-Adaptive Modulation Networks for Visual Perception

arXiv:2503.23947v2 Announce Type: replace Abstract: Recent studies have shown that 2D convolution and self-attention exhibit distinct spectral behaviors, and optimizing their spectral properties can e

Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training

arXiv:2605.11134v1 Announce Type: new Abstract: Preference learning methods such as Direct Preference Optimization (DPO) are known to induce reliance on spurious correlations, leading to sycophancy an

STRIDE: Training-Free Diversity Guidance via PCA-Directed Feature Perturbation in Single-Step Diffusion Models

arXiv:2605.11494v1 Announce Type: new Abstract: Distilled one-step (T=1) or few-step (Tleq4) diffusion models enable real-time image generation but often exhibit reduced sample diversity compared to t

SURGE: Surrogate Gradient Adaptation in Binary Neural Networks

arXiv:2605.10989v1 Announce Type: new Abstract: The training of Binary Neural Networks (BNNs) is fundamentally based on gradient approximation for non-differentiable binarization operations (e.g., sig

Tacmap: Bridging the Tactile Sim-to-Real Gap via Geometry-Consistent Penetration Depth Map

arXiv:2602.21625v2 Announce Type: replace Abstract: Vision-Based Tactile Sensors (VBTS) are essential for achieving dexterous robotic manipulation, yet the tactile sim-to-real gap remains a fundamenta

Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting

arXiv:2605.11538v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) has emerged as a promising approach for improving the reasoning capabilities of large language models. However

TAR: Text Semantic Assisted Cross-modal Image Registration Framework for Optical and SAR Images

arXiv:2605.12064v1 Announce Type: new Abstract: Existing deep learning-based methods can capture shared features from optical and synthetic aperture radar (SAR) images for spatial alignment. However,

Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling Failures

arXiv:2605.10991v1 Announce Type: new Abstract: Existing approaches to LLM personalization focus on constructing better personalized models or inputs, while treating inference as a single-shot process

The DAWN of World-Action Interactive Models

arXiv:2605.11550v1 Announce Type: new Abstract: A plausible scene evolution depends on the maneuver being considered, while a good maneuver depends on how the scene may evolve. Existing World Action M