Search: “safety”

Revisiting Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learning

arXiv:2510.10959v3 Announce Type: replace-cross Abstract: Reasoning ability has become a defining capability of Large Language Models (LLMs), with Reinforcement Learning with Verifiable Rewards (RLVR)

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Safety

Reward Weighted Classifier-Free Guidance as Policy Improvement in Autoregressive Models

DGX agent

arXiv:2604.15577v1 Announce Type: cross Abstract: Consider an auto-regressive model that produces outputs x (e.g., answers to questions, molecules) each of which can be summarized by an attribute vect

safetyarxiv-cs-ai

20 Apr 2026

Robust Multispectral Semantic Segmentation under Missing or Full Modalities via Structured Latent Projection

arXiv:2604.15856v1 Announce Type: cross Abstract: Multimodal remote sensing data provide complementary information for semantic segmentation, but in real-world deployments, some modalities may be unav

Robust Synchronisation for Federated Learning in The Face of Correlated Device Failure

arXiv:2604.16090v1 Announce Type: cross Abstract: Probabilistic Synchronous Parallel (PSP) is a technique in distributed learning systems to reduce synchronization bottlenecks by sampling a subset of

Sample Complexity Bounds for Stochastic Shortest Path with a Generative Model

arXiv:2604.16111v1 Announce Type: new Abstract: We study the sample complexity of learning an epsilon-optimal policy in the Stochastic Shortest Path (SSP) problem. We first derive sample complexity bo

Scalable Multi-Task Learning through Spiking Neural Networks with Adaptive Task-Switching Policy for Intelligent Autonomous Agents

arXiv:2504.13541v5 Announce Type: replace-cross Abstract: Training resource-constrained autonomous agents on multiple tasks simultaneously is crucial for adapting to diverse real-world environments. R

Scalable Unseen Objects 6-DoF Absolute Pose Estimation with Robotic Integration

arXiv:2503.05578v4 Announce Type: replace Abstract: Pose estimation-guided unseen object 6-DoF robotic manipulation is a key task in robotics. However, the scalability of current pose estimation metho

Self-Distillation as a Performance Recovery Mechanism for LLMs: Counteracting Compression and Catastrophic Forgetting

arXiv:2604.15794v1 Announce Type: cross Abstract: Large Language Models (LLMs) have achieved remarkable success, underpinning diverse AI applications. However, they often suffer from performance degra

Seriously. If you don’t understand the below, you really shouldn’t speculate about AI. It’s absolutely fundamental.

Seriously. If you don’t understand the below, you really shouldn’t speculate about AI. It’s absolutely fundamental. literally my basic model since 1998. crazy that some people still haven’t figured th

safetygary-marcus--x

SIMMER: Cross-Modal Food Image--Recipe Retrieval via MLLM-Based Embedding

arXiv:2604.15628v1 Announce Type: cross Abstract: Cross-modal retrieval between food images and recipe texts is an important task with applications in nutritional management, dietary logging, and cook

Skill-RAG: Failure-State-Aware Retrieval Augmentation via Hidden-State Probing and Skill Routing

arXiv:2604.15771v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) has emerged as a foundational paradigm for grounding large language models in external knowledge. While adaptive re

Subliminal Transfer of Unsafe Behaviors in AI Agent Distillation

arXiv:2604.15559v1 Announce Type: new Abstract: Recent work on subliminal learning demonstrates that language models can transmit semantic traits through data that is semantically unrelated to those t

'Taking Stock at FAccT': Using Participatory Design to Co-Create a Vision for the Fairness, Accountability and Transparency Community

arXiv:2604.16224v1 Announce Type: cross Abstract: As a relatively new forum, ACM FAccT has become a key space for activists and scholars to critically examine emerging AI and ML technologies. It bring

Targeted Exploration via Unified Entropy Control for Reinforcement Learning

arXiv:2604.14646v2 Announce Type: replace Abstract: Recent advances in reinforcement learning (RL) have improved the reasoning capabilities of large language models (LLMs) and vision-language models (

Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models

arXiv:2604.15383v1 Announce Type: cross Abstract: Large audio-language models (LALMs) generalize across speech, sound, and music, but unified decoders can exhibit a temporal smoothing bias: transient

The Harder Path: Last Iterate Convergence for Uncoupled Learning in Zero-Sum Games with Bandit Feedback

arXiv:2604.16087v1 Announce Type: new Abstract: We study the problem of learning in zero-sum matrix games with repeated play and bandit feedback. Specifically, we focus on developing uncoupled algorit

The Informational Cost of Agency: A Bounded Measure of Interaction Efficiency for Deployed Reinforcement Learning

arXiv:2603.01283v2 Announce Type: replace Abstract: Deployed RL agents operate in closed-loop systems where reliable performance depends on maintaining coherent coupling between observations, actions,

The Price of Paranoia: Robust Risk-Sensitive Cooperation in Non-Stationary Multi-Agent Reinforcement Learning

arXiv:2604.15695v1 Announce Type: cross Abstract: Cooperative equilibria are fragile. When agents learn alongside each other rather than in a fixed environment, the process of learning destabilizes th

Truncated Kernel Stochastic Gradient Descent with General Losses and Spherical Radial Basis Functions

arXiv:2510.04237v5 Announce Type: replace Abstract: In this paper, we propose a novel kernel stochastic gradient descent (SGD) algorithm for large-scale supervised learning with general losses. Compar

Unsupervised domain adaptation for radioisotope identification in gamma spectroscopy

arXiv:2603.05719v2 Announce Type: replace Abstract: Training machine learning models for radioisotope identification using gamma spectroscopy remains an elusive challenge for many practical applicatio

UsefulBench: Towards Decision-Useful Information as a Target for Information Retrieval

arXiv:2604.15827v1 Announce Type: cross Abstract: Conventional information retrieval is concerned with identifying the relevance of texts for a given query. Yet, the conventional definition of relevan

VADF: Vision-Adaptive Diffusion Policy Framework for Efficient Robotic Manipulation

arXiv:2604.15938v1 Announce Type: new Abstract: Diffusion policies are becoming mainstream in robotic manipulation but suffer from hard negative class imbalance due to uniform sampling and lack of sam

safetyarxiv-cs-ro

What Makes LLMs Effective Sequential Recommenders? A Study on Preference Intensity and Temporal Context

arXiv:2506.02261v3 Announce Type: replace-cross Abstract: What enables large language models (LLMs) to effectively model user preferences in sequential recommendation? Our investigation reveals that e

Whose Facts Win? LLM Source Preferences under Knowledge Conflicts

arXiv:2601.03746v3 Announce Type: replace Abstract: As large language models (LLMs) are more frequently used in retrieval-augmented generation pipelines, it is increasingly relevant to study their beh

Why Colors Make Clustering Harder:Global Integrality Gaps, the Price of Fairness, and Color-Coupled Algorithms in Chromatic Correlation Clustering

arXiv:2604.15738v1 Announce Type: new Abstract: Chromatic Correlation Clustering (CCC) extends Correlation Clustering by assigning semantic colors to edges and requiring each cluster to receive a sing

WildFeedback: Aligning LLMs With In-situ User Interactions And Feedback

arXiv:2408.15549v4 Announce Type: replace Abstract: As large language models (LLMs) continue to advance, aligning these models with human preferences has emerged as a critical challenge. Traditional a

good to know that a trillion dollars’ investment scaling has thoroughly solved the challenges of common sense.

Gary Marcus critiques the assumption that massive financial investment in AI scaling alone can solve fundamental challenges related to common sense reasoning in artificial intelligence systems. The po

safetygary-marcus--x

19 Apr 2026

easyaligner: Forced alignment with GPU acceleration and flexible text normalization (compatible with all w2v2 models on HF Hub) [P]

easyaligner is a forced alignment library designed to be performant and easy to use , leveraging GPU acceleration to align audio with text transcriptions. The tool supports flexible text normalization

safetyr-machinelearning

18 Apr 2026

EVERYONE who heard about @SeismicOrg (Seismic Foundation)'s report that AI 'salience' is low NEEDS to hear about this! New research from dat…

EVERYONE who heard about @SeismicOrg (Seismic Foundation)'s report that AI 'salience' is low NEEDS to hear about this! New research from data science wiz @davidshor shows AI is now ahead of ABORTION a

safetyconnor-leahy--x

18 Apr 2026

“my honest read is that a significant portion of this spending is driven by competitive fear rather than demonstrated returns. Nobody wants …

“my honest read is that a significant portion of this spending is driven by competitive fear rather than demonstrated returns. Nobody wants to be the company that didn't invest in AI when everyone els

safetygary-marcus--x

18 Apr 2026

Model Releases

3D Instruction Ambiguity Detection

model-releasesarxiv-cs-ai

arXiv:2601.05991v2 Announce Type: replace Abstract: In safety-critical domains, linguistic ambiguity can have severe consequences; a vague command like 'Pass me the vial' in a surgical setting could l

A Mechanistic Account of Attention Sinks in GPT-2: One Circuit, Broader Implications for Mitigation

arXiv:2604.14722v1 Announce Type: new Abstract: Transformers commonly exhibit an attention sink: disproportionately high attention to the first position. We study this behavior in GPT-2-style models w

Abstract Sim2Real through Approximate Information States

arXiv:2604.15289v1 Announce Type: new Abstract: In recent years, reinforcement learning (RL) has shown remarkable success in robotics when a fast and accurate simulator is available for a given task.

safetyarxiv-cs-ro

AFFORD2ACT: Affordance-Guided Automatic Keypoint Selection for Generalizable and Lightweight Robotic Manipulation

arXiv:2510.01433v2 Announce Type: replace Abstract: Vision-based robot learning often relies on dense image or point-cloud inputs, which are computationally heavy and entangle irrelevant background fe

safetyarxiv-cs-ro

Beyond Importance Sampling: Rejection-Gated Policy Optimization

arXiv:2604.14895v1 Announce Type: new Abstract: We propose a new perspective on policy optimization: rather than reweighting all samples by their importance ratios, an optimizer should select which sa

Bias in Surface Electromyography Features across a Demographically Diverse Cohort

arXiv:2604.14460v1 Announce Type: cross Abstract: Neuromotor decoding from upper-limb electromyography (sEMG) can enhance human-machine interfaces and offer a more natural means of controlling prosthe

Bird-SR: Bidirectional Reward-Guided Diffusion for Real-World Image Super-Resolution

arXiv:2602.07069v2 Announce Type: replace Abstract: Powered by multimodal text-to-image priors, diffusion-based super-resolution excels at synthesizing intricate details; however, models trained on sy

BoundRL: Efficient Structured Text Segmentation through Reinforced Boundary Generation

arXiv:2510.20151v2 Announce Type: replace Abstract: Structured texts refer to texts containing structured elements beyond plain texts, such as code snippets and placeholders. Such structured texts inc

Calibration-Gated LLM Pseudo-Observations for Online Contextual Bandits

arXiv:2604.14961v1 Announce Type: new Abstract: Contextual bandit algorithms suffer from high regret during cold-start, when the learner has insufficient data to distinguish good arms from bad. We pro

Chain of Modality: From Static Fusion to Dynamic Orchestration in Omni-MLLMs

arXiv:2604.14520v1 Announce Type: new Abstract: Omni-modal Large Language Models (Omni-MLLMs) promise a unified integration of diverse sensory streams. However, recent evaluations reveal a critical pe

ClimateCause: Complex and Implicit Causal Structures in Climate Reports

arXiv:2604.14856v1 Announce Type: new Abstract: Understanding climate change requires reasoning over complex causal networks. Yet, existing causal discovery datasets predominantly capture explicit, di

ConfLayers: Adaptive Confidence-based Layer Skipping for Self-Speculative Decoding

arXiv:2604.14612v1 Announce Type: cross Abstract: Self-speculative decoding is an inference technique for large language models designed to speed up generation without sacrificing output quality. It c

Continuous-time reinforcement learning: ellipticity enables model-free value function approximation

arXiv:2602.06930v2 Announce Type: replace Abstract: We study off-policy reinforcement learning for controlling continuous-time Markov diffusion processes with discrete-time observations and actions. W

Controllable Video Object Insertion via Multiview Priors

arXiv:2604.14556v1 Announce Type: new Abstract: Video object insertion is a critical task for dynamically inserting new objects into existing environments. Previous video generation methods focus prim

Crowdsourcing of Real-world Image Annotation via Visual Properties

arXiv:2604.14449v1 Announce Type: new Abstract: Recent advances in data-centric artificial intelligence highlight inherent limitations in object recognition datasets. One of the primary issues stems f

CURA: Clinical Uncertainty Risk Alignment for Language Model-Based Risk Prediction

arXiv:2604.14651v1 Announce Type: new Abstract: Clinical language models (LMs) are increasingly applied to support clinical risk prediction from free-text notes, yet their uncertainty estimates often

Diagnosing and Improving Diffusion Models by Estimating the Optimal Loss Value

arXiv:2506.13763v2 Announce Type: replace-cross Abstract: Diffusion models have achieved remarkable success in generative modeling. Despite more stable training, the loss of diffusion models is not in

Direct Preference Optimization for Primitive-Enabled Hierarchical RL: A Bilevel Approach

arXiv:2411.00361v4 Announce Type: replace Abstract: Hierarchical reinforcement learning (HRL) enables agents to solve complex, long-horizon tasks by decomposing them into manageable sub-tasks. However