new

Get trending papers in your email inbox once a day!

Get trending papers in your email inbox!

Daily Papers

byAK and the research community

May 13

Submitted by

Paranioar

SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture

sensenova

Submitted by

Hush-cd

MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud Agents

MemTensor

Submitted by

taesiri

δ-mem: Efficient Online Memory for Large Language Models

mindlab-research

Submitted by

gaotang

RubricEM: Meta-RL with Rubric-guided Policy Decomposition beyond Verifiable Rewards

google

Submitted by

patricebechard

Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics

ServiceNow-AI

1

Submitted by

sinwang

World Action Models: The Next Frontier in Embodied AI

OpenMOSS-Team

Submitted by

huangrh9

AlphaGRPO: Unlocking Self-Reflective Multimodal Generation in UMMs via Decompositional Verifiable Reward

·
5 authors

Submitted by

DogNeverSleep

Beyond the Last Layer: Multi-Layer Representation Fusion for Visual Tokenization

·
7 authors

Submitted by

Foreshhh

ToolCUA: Towards Optimal GUI-Tool Path Orchestration for Computer Use Agents

AlibabaTongyiLab

Submitted by

bloc97

Efficient Pre-Training with Token Superposition

NousResearch

Submitted by

zhen-nan

L2P: Unlocking Latent Potential for Pixel Generation

·
10 authors

Submitted by

Yhmeng1106

CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives

antgroup

Submitted by

DhavalPatel

MCP-Cosmos: World Model-Augmented Agents for Complex Task Execution in MCP Environments

ibm-research

Submitted by

Hyeoni

Teaching Language Models to Think in Code

Korea University

1

Submitted by

JoeYing

Towards On-Policy Data Evolution for Visual-Native Multimodal Deep Search Agents

The Hong Kong University of Science and Technology

Submitted by

weiqingXiao

Relit-LiVE: Relight Video by Jointly Learning Environment Video

nanjinguniv

Nanjing University

Submitted by

LIQIIIII

On-Policy Self-Evolution via Failure Trajectories for Agentic Safety Alignment

·
3 authors

Submitted by

Jongwondd

Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States

SeoulNatlUniv

Seoul National University

Submitted by

JonasGeiping

Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputs

ELLIS-Institute-Tuebingen

ELLIS Institute Tübingen

Submitted by

guanzhong2

Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction

jingdong1

Submitted by

Kun-Xiang

SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning

·
18 authors

Submitted by

Sicong

World Model for Robot Learning: A Comprehensive Survey

·
18 authors

Submitted by

taesiri

From Web to Pixels: Bringing Agentic Search into Visual Perception

·
6 authors

Submitted by

rishabh2k1

Learning, Fast and Slow: Towards LLMs That Adapt Continually

UCBerkeley

University of California, Berkeley

Submitted by

milkkarten

Continual Harness: Online Adaptation for Self-Improving Foundation Agents

princetonu

Princeton University

Submitted by

Thrillcrazyer

LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models

BAELABPNU

BIGDATA ANALYTICS ENGINEERING LAB, Pusan National University, Busan, Korea

Submitted by

FSCCS

PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks

Submitted by

dmux

Debiased Model-based Representations for Sample-efficient Continuous Control

Tencent-Hunyuan

Tencent Hunyuan

Submitted by

xuyd16

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training

princetonu

Princeton University

Submitted by

henry-yeh

Agent-ValueBench: A Comprehensive Benchmark for Evaluating Agent Values

PKU1898

Peking University

Submitted by

andreagurioli1995

Do not copy and paste! Rewriting strategies for code retrieval

·
3 authors

Submitted by

Frinkleko

One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue

Graph-COM

Graph Computation and Machine Learning (GCOM) Group

Submitted by

Gigglingface

MEME: Multi-entity & Evolving Memory Evaluation

kaist-ai

Submitted by

WenDingY

Beyond Reasoning: Reinforcement Learning Unlocks Parametric Knowledge in LLMs

·
8 authors

Submitted by

rntc

A Causal Language Modeling Detour Improves Encoder Continued Pretraining

almanach

ALMAnaCH (Inria)

Submitted by

zsqzz

The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes

UIUC-CS

University of Illinois at Urbana-Champaign

Submitted by

DarkBluee

A Single Layer to Explain Them All:Understanding Massive Activations in Large Language Models

RutgersU

Rutgers University

1

Submitted by

wy1iu

Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation

CUHK-CSE

The Chinese University of Hong Kong

Submitted by

xiaowu0162

LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues

uclanlp

Submitted by

wufeim

LychSim: A Controllable and Interactive Simulation Framework for Vision Research

JohnsHopkins

Johns Hopkins University

Submitted by

jindongwang

UniPath: Adaptive Coordination of Understanding and Generation for Unified Multimodal Reasoning

williammary

The College of William & Mary

Submitted by

taesiri

VidSplat: Gaussian Splatting Reconstruction with Geometry-Guided Video Diffusion Priors

·
8 authors

Submitted by

taesiri

Reward Hacking in Rubric-Based Reinforcement Learning

·
6 authors

Submitted by

taesiri

Images in Sentences: Scaling Interleaved Instructions for Unified Visual Generation

·
5 authors

Submitted by

taicheng

AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive

notredame

University of Notre Dame

Submitted by

utopiar

MoCam: Unified Novel View Synthesis via Structured Denoising Dynamics

Orange-Team

Submitted by

alphadl

IndustryBench: Probing the Industrial Knowledge Boundaries of LLMs

alibaba-multimodal-industrial-ai

1688 multimodal & industrial AI

Submitted by

ZBox008003

Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts

RutgersU

Rutgers University

Submitted by

Lxyhaha

TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent Systems

·
7 authors

Submitted by

xwen99

AdaPreLoRA: Adafactor Preconditioned Low-Rank Adaptation

·
3 authors

Submitted by

pariard

Solve the Loop: Attractor Models for Language and Reasoning

UniversityofSouthernCalifornia

University of Southern California

Submitted by

nverma

ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging

·
10 authors

Submitted by

monurcan

WildRelight: A Real-World Benchmark and Physics-Guided Adaptation for Single-Image Relighting

dtudk

Technical University of Denmark

Submitted by

SteveZeyuZhang

Lite3R: A Model-Agnostic Framework for Efficient Feed-Forward 3D Reconstruction

PekingUniversity

Peking University

Submitted by

JarvisPei

FocuSFT: Bilevel Optimization for Dilution-Aware Long-Context Fine-Tuning

·
6 authors

Submitted by

stefan-it

GLiNER-Relex: A Unified Framework for Joint Named Entity Recognition and Relation Extraction

·
4 authors

Submitted by

liangqiy

PAAC: Privacy-Aware Agentic Device-Cloud Collaboration

·
4 authors

1

Submitted by

liangqiy

Large Language Models over Networks: Collaborative Intelligence under Resource Constraints

·
5 authors

1

Submitted by

niwase

Reliable Chain-of-Thought via Prefix Consistency

MBZUAI

Mohamed Bin Zayed University of Artificial Intelligence

Submitted by

mdswyz

Implicit Preference Alignment for Human Image Animation

Tencent-Hunyuan

Tencent Hunyuan

Submitted by

mdswyz

FaithfulFaces: Pose-Faithful Facial Identity Preservation for Text-to-Video Generation

Tencent-Hunyuan

Tencent Hunyuan

1

Submitted by

joykirat

Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty

·
8 authors

Submitted by

ravfogs

Geometric Factual Recall in Transformers

·
4 authors

Submitted by

Yiwei-Ou

Urban-ImageNet: A Large-Scale Multi-Modal Dataset and Evaluation Framework for Urban Space Perception

auckland

University of Auckland