AI研究論文每日精選

Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning

SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics

REASONING GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewards

Taming LLMs by Scaling Learning Rates with Gradient Grouping

Temporal In-Context Fine-Tuning for Versatile Control of Video Diffusion Models

SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning

ShapeLLM-Omni: A Native Multimodal LLM for 3D Generation and Understanding

ARIA: Training Language Agents with Intention-Driven Reward Aggregation

LoHoVLA: A Unified Vision-Language-Action Model for Long-Horizon Embodied Tasks

Jigsaw-R1: A Study of Rule-based Visual Reinforcement Learning with Jigsaw Puzzles

Learning Video Generation for Robotic Manipulation with Collaborative Trajectory Control

EarthMind: Towards Multi-Granular and Multi-Sensor Earth Observation with Large Multimodal Models

AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning

Unified Scaling Laws for Compressed Representations

MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning

Incentivizing Reasoning for Advanced Instruction-Following of Large Language Models

IVY-FAKE: A Unified Explainable Framework and Benchmark for Image and Video AIGC Detection

From Token to Action: State Machine Reasoning to Mitigate Overthinking in Information Retrieval

Reasoning Like an Economist: Post-Training on Economic Problems Induces Strategic Generalization in LLMs

Cora: Correspondence-aware image editing using few step diffusion

WebChoreArena: Evaluating Web Browsing Agents on Realistic Tedious Web Tasks

VisualSphinx: Large-Scale Synthetic Vision Logic Puzzles for RL

OWSM v4: Improving Open Whisper-Style Speech Models via Data Scaling and Cleaning

Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors

Stress-testing Machine Generated Text Detection: Shifting Language Models Writing Style to Fool Detectors

CodeV-R1: Reasoning-Enhanced Verilog Generation

DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors

Normalized Attention Guidance: Universal Negative Guidance for Diffusion Model

Esoteric Language Models

zip2zip: Inference-Time Adaptive Vocabularies for Language Models via Token Compression

Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents

WHEN TO ACT, WHEN TO WAIT: Modeling Structural Trajectories for Intent Triggerability in Task-Oriented Dialogue

Cascading Adversarial Bias from Injection to Distillation in Language Models

VAU-R1: Advancing Video Anomaly Understanding via Reinforcement Fine-Tuning

SATA-BENCH: Select All That Apply Benchmark for Multiple Choice Questions

Pro3D-Editor : A Progressive-Views Perspective for Consistent and Precise 3D Editing

Stepsize anything: A unified learning rate schedule for budgeted-iteration training

From Guidelines to Practice: A New Paradigm for Arabic Language Model Evaluation

From Guidelines to Practice: A New Paradigm for Arabic Language Model Evaluation

LLM in the Loop: Creating the PARADEHATE Dataset for Hate Speech Detoxification

RARE: Retrieval-Aware Robustness Evaluation for Retrieval-Augmented Generation Systems

ComposeAnything: Composite Object Priors for Text-to-Image Generation

OmniResponse: Online Multimodal Conversational Response Generation in Dyadic Interactions

How Programming Concepts and Neurons Are Shared in Code Language Models

SealQA: Raising the Bar for Reasoning in Search-Augmented Language Models

Pitfalls in Evaluating Language Model Forecasters

SenseFlow: Scaling Distribution Matching for Flow-based Text-to-Image Distillation

MaskSearch: A Universal Pre-Training Framework to Enhance Agentic Search Capability

Think Again! The Effect of Test-Time Compute on Preferences, Opinions, and Beliefs of Large Language Models

Aligning VLM Assistants with Personalized Situated Cognition

LIFT the Veil for the Truth: Principal Weights Emerge after Rank Reduction for Reasoning-Focused Supervised Fine-Tuning

CityLens: Benchmarking Large Language-Vision Models for Urban Socioeconomic Sensing

Massively Multilingual Adaptation of Large Language Models Using Bilingual Translation Data

MagiCodec: Simple Masked Gaussian-Injected Codec for High-Fidelity Reconstruction and Generation

Neuro2Semantic: A Transfer Learning Framework for Semantic Reconstruction of Continuous Language from Human Intracranial EEG

BinauralFlow: A Causal and Streamable Approach for High-Quality Binaural Speech Synthesis with Flow Matching Models

R1-Code-Interpreter: Training LLMs to Reason with Code via Supervised and Reinforcement Learning

Frankentext: Stitching random text fragments into long-form narratives

Plan and Budget: Effective and Efficient Test-Time Scaling on Large Language Model Reasoning

Pixels Versus Priors: Controlling Knowledge Priors in Vision-Language Models through Visual Counterfacts

MIKU-PAL: An Automated and Standardized Multi-Modal Method for Speech Paralinguistic and Affect Labeling

Shuffle PatchMix Augmentation with Confidence-Margin Weighted Pseudo-Labels for Enhanced Source-Free Domain Adaptation

Synthesis of discrete-continuous quantum circuits with multimodal diffusion models