Search: “safety”

A Continuous-Time Reinforcement Learning Framework for Fine-Tuning Discrete Diffusion Models

arXiv:2607.14522v2 Announce Type: replace Abstract: We formulate reinforcement learning (RL) in continuous time with discrete state spaces and possibly arbitrary action spaces via a stochastic control

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

A Journey to the Edge of Stability

DGX agent

arXiv:2609.32290v1 Announce Type: new Abstract: It has recently been found that deep learning often occurs at the 'edge of stability (EoS),' where the maximum Hessian eigenvalue of the model is stabil

safetyarxiv-cs-lg

29 Sept 2026

ABC-Align: Prediction-Powered Alignment with Adaptive Bias Control

arXiv:2609.34374v1 Announce Type: new Abstract: Language model post-training is often bottlenecked by the need for human-collected preference data, which is expensive and difficult to scale. Reinforce

ABFR-KAN: Kolmogorov-Arnold Networks for Functional Brain Analysis

arXiv:2601.00416v2 Announce Type: replace Abstract: Functional connectivity (FC) analysis, a valuable tool for computer-aided brain disorder diagnosis, traditionally relies on atlas-based parcellation

Achieve What You Imagined: Learning to Align Actions with Visual Plans

arXiv:2609.33832v1 Announce Type: cross Abstract: World-action models can jointly predict future visual observations and robot actions. However, discrepancies may exist between their visual prediction

Action-Driven Processes for Continuous-Time Control

arXiv:2510.26672v2 Announce Type: replace-cross Abstract: At the heart of reinforcement learning are actions -- decisions made in response to observations of the environment. Actions are equally funda

Action Shaping: Policies Absorb What They Can Express

arXiv:2609.32752v1 Announce Type: cross Abstract: Reward shaping has a theorem: a potential-based term can be removed without changing the optimal policy. The same practice on the action channel, an o

ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies

arXiv:2609.33256v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models map visual observations and language instructions directly to robot actions, but they do not explicitly represent th

Active Data Acquisition with Side Information via Discrete Diffusion Priors

arXiv:2609.32252v1 Announce Type: cross Abstract: Acquiring data is costly: higher measurement fidelity costs power and storage and risks collecting irrelevant content, while aggressive cost reduction

Adam or Gauss-Newton? A Comparative Study In Terms of Basis Alignment and SGD Noise

arXiv:2510.13680v2 Announce Type: replace Abstract: Preconditioned methods are central to deep learning optimization. Predominant approaches include computationally light diagonal preconditioners such

Adapting Nonstationary Multi-output Gaussian Processes to Bayesian Optimization

arXiv:2609.32464v1 Announce Type: new Abstract: Multi-objective Bayesian optimization (MOBO) commonly relies on independent Gaussian processes (GPs) with stationary kernels, limiting its ability to re

Adjoint Guidance Flow: Amortized Critic Guidance for VLA Policies

arXiv:2609.34944v1 Announce Type: cross Abstract: Flow-based Vision-Language-Action (VLA) policies are typically trained by behavior cloning and thus do not explicitly optimize long-term task return.

Adversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking

arXiv:2602.01750v2 Announce Type: replace-cross Abstract: Reinforcement Learning from Human Feedback (RLHF) remains vulnerable to reward hacking, where models exploit spurious correlations in learned

Agent Priors-guided Policy Learning

arXiv:2609.35690v1 Announce Type: new Abstract: Robots that learn from a few demonstrations often require two forms of generalization. Compositional generalization recombines skills to solve new tasks

Agentic Multi-Turn Reasoning: A Fairness Approach

arXiv:2609.33323v1 Announce Type: cross Abstract: Recent advances in Large Language Models (LLMs) have enabled agentic systems capable of solving complex tasks through multi-turn planning, tool use, v

AGRO-SUVIDE: Agentic Robotics for Surgical Viscoelastic Debridement

arXiv:2609.34823v1 Announce Type: new Abstract: Augmented dexterity has the potential to reduce the fatigue experienced by surgeons during repetitive surgical tasks. In this paper, we propose the firs

AlignBeat: A Latent Variable Model for Multi-Class Beat Tracking from Partially Labeled Data

arXiv:2510.14391v3 Announce Type: replace-cross Abstract: Recent neural beat trackers predict beats and downbeats with two independent frame-wise binary classification heads, a convention adopted beca

An analysis of Mirror-Descent Soft Actor-Critic

arXiv:2609.35466v1 Announce Type: new Abstract: Soft Actor-Critic (SAC) is widely used for entropy-regularised reinforcement learning with continuous action spaces, and practical implementations perfo

An Empirical Study on What Matters for Viewpoint-Generalizable Policies in Visual Imitation Learning

arXiv:2609.32762v1 Announce Type: cross Abstract: Visual imitation learning is a promising approach to training robot manipulation policies capable of completing a wide variety of tasks. However, poli

An Imperfect Verifier is Good Enough: Learning with Noisy Rewards

arXiv:2604.07666v2 Announce Type: replace Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is widely used for post-training Large Language Models, but practical verifiers can make error

An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning

arXiv:2609.35505v1 Announce Type: new Abstract: We study on-policy distillation (OPD) through the lens of reinforcement learning, establishing a connection between the reverse-KL objective in OPD and

Attribution Without a Second Pass: Inline Per-Sample Gradient Provenance at ~1% Overhead

arXiv:2609.32380v1 Announce Type: new Abstract: Data attribution methods used in practice (TRAK, LoGRA, EK-FAC) are post-hoc: after training they make a second pass over the training set to recompute

Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification

arXiv:2609.33662v1 Announce Type: cross Abstract: Imperfect verifiers can assign a harmful update direction even when clipping and regularization bound its magnitude. We introduce Audit-First VAPO, wh

Benign Overfitting for General Norms and Distributions

arXiv:2609.33675v1 Announce Type: new Abstract: Understanding why predictors can generalize despite interpolating noisy training data is a central puzzle in machine learning. Most work on such 'benign

Beyond Conservatism: Recoverability-Conditioned Exploration for Model-Based Imitation Learning

arXiv:2609.33336v1 Announce Type: new Abstract: Model-based imitation learning (MBIL) improves real-environment interaction efficiency by optimizing policies on imagined rollouts from a learned world

Beyond Reconstruction Loss in Post-Training Quantization: Balanced Fitting for Large Vision-Language Models

arXiv:2609.34765v1 Announce Type: new Abstract: Post-training quantization (PTQ) enables efficient deployment of large vision-language models (LVLMs), but is typically calibrated on a small set while

Beyond Saying Less: Fine-Grained Alignment for Informative and Faithful Vision-Language Models

arXiv:2609.35294v1 Announce Type: new Abstract: Object hallucination remains a major challenge for large vision-language models. While off-policy preference optimization proves to be an effective solu

Beyond Site Agreement: Re-estimation for Brain Network Generalization

arXiv:2609.34611v1 Announce Type: new Abstract: Cross-site out-of-distribution (OOD) generalization in resting-state functional magnetic resonance imaging (rs-fMRI) often relies on learning task-discr

Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation

arXiv:2609.35347v1 Announce Type: new Abstract: Reinforcement learning can turn one language model into several specialists, each excellent at a single skill such as mathematics, coding or following i

Beyond Timestamps: Decision-Aligned On-Policy Distillation for Long-Horizon Agents

arXiv:2609.33391v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) often relies on sparse outcome rewards, providing coarse supervision for long-horizon agents. On-p

BiasReducer: Adaptive Bias Mitigation for Reward Models

arXiv:2609.32720v1 Announce Type: new Abstract: Reward models score responses from large language models (LLMs) and guide LLM training toward human preferences. However, reward models can favor superf

Black-Box Auditing of Epistemic Reliability in Multi-Agent Debate Distillation

arXiv:2609.32361v1 Announce Type: new Abstract: Debate distillation adapts weaker verifiers using multi-agent debate transcripts to improve their judgement in subsequent debates, but gains on monitore

Brain-Conditioned Action Policies for Neural Motor Decoding

arXiv:2609.34561v1 Announce Type: new Abstract: Motor brain-computer interfaces (BCIs) aim to decode motor intention, enabling people with paralysis to control external devices. Neural motor decoding

Bridging Generative and Discriminative Noisy-Label Learning via Direction-Agnostic EM Formulation

arXiv:2308.01184v4 Announce Type: replace Abstract: Although noisy-label learning is often approached with discriminative methods for simplicity and speed, generative modeling offers a principled alte

Cache-Aware Conv3D Lowering Across Embedded World-Model Decoders

arXiv:2609.31938v1 Announce Type: new Abstract: Generative world models can provide visual rollouts for embodied planning, yet their feasibility on edge devices depends not only on the learned model b

Can Circuit Alignment Predict OOD Generalization?

arXiv:2609.31996v1 Announce Type: new Abstract: Can out-of-distribution (OOD) generalization be predicted from a trained model's weights alone, without any target-domain data? Existing representationa

CAPEX: Efficiently Distilling Foundation Model Behavior into Deployable Robot Policies through Experience-Adaptive Reasoning

arXiv:2609.33007v1 Announce Type: cross Abstract: Robot learning has largely relied on human-teleoperated demonstrations to acquire effective learnable behaviors. However, human-operated data collecti

CapField-OPD: Learning Continuous Capability Fields via Joint-Anchored Multi-Teacher On-Policy Distillation for Flow Models

arXiv:2609.34658v1 Announce Type: new Abstract: Reward-specialized post-training produces strong experts for flow-based generative models, while multi-teacher on-policy distillation (OPD) consolidates

CaptchaArena: A Large-Scale, Fine-Grained Dataset for Training Computer-Use Agents on Interactive CAPTCHAs

arXiv:2609.31957v1 Announce Type: new Abstract: Interactive CAPTCHAs remain challenging for computer-use agents, while existing datasets face trade-offs among type coverage, interaction fidelity, and

CARVE: Breaking Data Barriers in Chip Placement by Harnessing Reusable Expertise

arXiv:2609.33106v1 Announce Type: new Abstract: Pretrained macro-placement policies can reduce repeated optimization across circuits, but deployment often exposes them to unfamiliar designs when the o

Certified Multi-Source Integrity for Structured Agent Actions

arXiv:2609.34245v1 Announce Type: cross Abstract: LLM agents increasingly take privileged, often irreversible structured actions, such as paying an invoice. They assemble each action from action-criti

ChemOPD: Multi-Teacher On-Policy Distillation for Multi-Task Chemical Reasoning

arXiv:2609.33838v1 Announce Type: new Abstract: Large language models are increasingly expected to support diverse chemical reasoning capabilities within a unified model. One approach is to develop sp

CLAP: Closed-Loop Alignment with Pressure for Precise Suction Manipulation

arXiv:2609.32767v1 Announce Type: new Abstract: Stacking and palletising demand precise placement: error left in one layer is inherited by the next, and a flat pad offers no feature to funnel a wrong

Clarify the User or Verify the World? Uncertainty Routing for Proactive Agents

arXiv:2609.32255v1 Announce Type: cross Abstract: Tool-using LLM agents must decide not only whether additional information is needed, but also which source can resolve the uncertainty. Existing proac

Clinical Trajectory Alignment for Medical Vision-Language Pre-training

arXiv:2609.34439v1 Announce Type: new Abstract: Medical vision-language pre-training largely follows a visit-level image-report matching paradigm, aligning paired images and reports at individual visi

Clipped or Unclipped? Finite-Sample Trade-offs for Averaged SGD under Heavy-Tailed Noise

arXiv:2609.32962v1 Announce Type: new Abstract: Gradient clipping is widely used to stabilize training, but it need not improve the statistical accuracy of averaged SGD, even under heavy-tailed noise.

CoBrush: A Hierarchical Planning Framework for Human-Robot Co-Painting

arXiv:2609.35476v1 Announce Type: new Abstract: Embodied co-painting requires a robot to repeatedly update a shared physical canvas while human intent evolves over interaction. Existing reference-driv

Coding Agent Memory Post-training: Unlocking the Memory Potential of Pre-trained File Operations for Long-Horizon Tasks via Reinforcement Learning

arXiv:2609.34422v1 Announce Type: new Abstract: Language-model agents increasingly tackle long-horizon tasks whose interaction histories exceed the model's active context. Recent work has begun to use