Daily Arxiv

전 세계에서 발간되는 인공지능 관련 논문을 정리하는 페이지 입니다.
본 페이지는 Google Gemini를 활용해 요약 정리하며, 비영리로 운영 됩니다.
논문에 대한 저작권은 저자 및 해당 기관에 있으며, 공유 시 출처만 명기하면 됩니다.

Inductive Moment Matching

LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL

Ideas in Inference-time Scaling can Benefit Generative Pre-training Algorithms

FaceID-6M: A Large-Scale, Open-Source FaceID Customization Dataset

Interactive Medical Image Analysis with Concept-based Similarity Reasoning

Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models

LightMotion: A Light and Tuning-free Method for Simulating Camera Motion in Video Generation

A Transformer Model for Predicting Chemical Reaction Products from Generic Templates

CBW: Towards Dataset Ownership Verification for Speaker Verification via Clustering-based Backdoor Watermarking

The Lazy Student's Dream: ChatGPT Passing an Engineering Course on Its Own

VideoPainter: Any-length Video Inpainting and Editing with Plug-and-Play Context Control

Multi-Task Reinforcement Learning Enables Parameter Scaling

Balcony: A Lightweight Approach to Dynamic Inference of Generative Language Models

Prediction of Frozen Region Growth in Kidney Cryoablation Intervention Using a 3D Flow-Matching Model

Call for Rigor in Reporting Quality of Instruction Tuning Data

KunlunBaize: LLM with Multi-Scale Convolution and Multi-Token Prediction Under TransformerX Framework

(How) Do Language Models Track State?

IterPref: Focal Preference Learning for Code Generation via Iterative Debugging

PaCA: Partial Connection Adaptation for Efficient Fine-Tuning

Forgotten Polygons: Multimodal Large Language Models are Shape-Blind

Q-PETR: Quant-aware Position Embedding Transformation for Multi-View 3D Object Detection

FR-Spec: Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling

Helix-mRNA: A Hybrid Foundation Model For Full Sequence mRNA Therapeutics

IMLE Policy: Fast and Sample Efficient Visuomotor Policy Learning via Implicit Maximum Likelihood Estimation

Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Models

Equivariant Masked Position Prediction for Efficient Molecular Representation

IRepair: An Intent-Aware Approach to Repair Data-Driven Errors in Large Language Models

Automated Consistency Analysis of LLMs

ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates

Rationalization Models for Text-to-SQL

Reflection-Window Decoding: Text Generation with Selective Refinement

Agentic Bug Reproduction for Effective Automated Program Repair at Google

Data Duplication: A Novel Multi-Purpose Attack Paradigm in Machine Unlearning

MambaQuant: Quantizing the Mamba Family with Variance Aligned Rotation Methods

KAA: Kolmogorov-Arnold Attention for Enhancing Attentive Graph Neural Networks

TexAVi: Generating Stereoscopic VR Video Clips from Text Descriptions

Multi-P$^2$A: A Multi-perspective Benchmark on Privacy Assessment for Large Vision-Language Models

Faster Vision Mamba is Rebuilt in Minutes via Merged Token Re-training

SweetTok: Semantic-Aware Spatial-Temporal Tokenizer for Compact Video Discretization

DMin: Scalable Training Data Influence Estimation for Diffusion Models

MAGIC: Mastering Physical Adversarial Generation in Context through Collaborative LLM Agents

LayoutVLM: Differentiable Optimization of 3D Layout via Vision-Language Models

RL-MILP Solver: A Reinforcement Learning Approach for Solving Mixed-Integer Linear Programs with Graph Neural Networks

Proto Successor Measure: Representing the Behavior Space of an RL Agent

KinMo: Kinematic-aware Human Motion Understanding and Generation

Towards Million-Scale Adversarial Robustness Evaluation With Stronger Individual Attacks

OminiControl: Minimal and Universal Control for Diffusion Transformer

MTA: Multimodal Task Alignment for BEV Perception and Captioning

PyGen: A Collaborative Human-AI Approach to Python Package Creation

Fair Summarization: Bridging Quality and Diversity in Extractive Summaries

AtlasSeg: Atlas Prior Guided Dual-U-Net for Cortical Segmentation in Fetal Brain MRI

V-LoRA: An Efficient and Flexible System Boosts Vision Applications with LoRA LMM

Fourier Head: Helping Large Language Models Learn Complex Probability Distributions

Conditional diffusions for neural posterior estimation

LLM-HDR: Bridging LLM-based Perception and Self-Supervision for Unpaired LDR-to-HDR Image Reconstruction

Chemistry-Inspired Diffusion with Non-Differentiable Guidance

Taylor Unswift: Secured Weight Release for Large Language Models via Taylor Expansion

6DGS: Enhanced Direction-Aware Gaussian Splatting for Volumetric Rendering

CAX: Cellular Automata Accelerated in JAX

Emotion-Aware Embedding Fusion in LLMs (Flan-T5, LLAMA 2, DeepSeek-R1, and ChatGPT 4) for Intelligent Response Generation

What Information Contributes to Log-based Anomaly Detection? Insights from a Configurable Transformer-Based Approach

Meta-RTL: Reinforcement-Based Meta-Transfer Learning for Low-Resource Commonsense Reasoning

LEMMo-Plan: LLM-Enhanced Learning from Multi-Modal Demonstration for Planning Sequential Contact-Rich Manipulation Tasks

Training with Differential Privacy: A Gradient-Preserving Noise Reduction Approach with Provable Security

ASMA: An Adaptive Safety Margin Algorithm for Vision-Language Drone Navigation via Scene-Aware Control Barrier Functions

Surgical SAM 2: Real-time Segment Anything in Surgical Video by Efficient Frame Pruning

Inference-Time Selective Debiasing to Enhance Fairness in Text Classification Models

Detect, Investigate, Judge and Determine: A Knowledge-guided Framework for Few-shot Fake News Detection

How Data Inter-connectivity Shapes LLMs Unlearning: A Structural Unlearning Perspective

Value Improved Actor Critic Algorithms

Synthesizing Programmatic Reinforcement Learning Policies with Large Language Model Guided Search

Curriculum Direct Preference Optimization for Diffusion and Consistency Models

Is the House Ready For Sleeptime? Generating and Evaluating Situational Queries for Embodied Question Answering

RealmDreamer: Text-Driven 3D Scene Generation with Inpainting and Depth Diffusion

Adversarial Guided Diffusion Models for Adversarial Purification

M-HOF-Opt: Multi-Objective Hierarchical Output Feedback Optimization via Multiplier Induced Loss Landscape Scheduling

The VampPrior Mixture Model

Regularization by Texts for Latent Diffusion Inverse Solvers

Deep Tensor Network

Identifying the Truth of Global Model: A Generic Solution to Defend Against Byzantine and Backdoor Attacks in Federated Learning (full version)

Generalizable Imitation Learning Through Pre-Trained Representations

Hypergraph Structure Inference From Data Under Smoothness Prior

HowkGPT: Investigating the Detection of ChatGPT-generated University Student Homework through Context-Aware Perplexity Analysis

SketchOGD: Memory-Efficient Continual Learning

ChatGPT-4 in the Turing Test: A Critical Analysis

Toward an Evaluation Science for Generative AI Systems

WritingBench: A Comprehensive Benchmark for Generative Writing

ToolFuzz -- Automated Agent Tool Testing

Building Interval Type-2 Fuzzy Membership Function: A Deck of Cards based Co-constructive Approach

Optimus-2: Multimodal Minecraft Agent with Goal-Observation-Action Conditioned Policy

Learning to Plan with Personalized Preferences

Exponential Speedups by Rerooting Levin Tree Search

A Unified Framework for Motion Reasoning and Generation in Human Interaction

Agent-Oriented Planning in Multi-Agent Systems

Dynamic Analysis and Adaptive Discriminator for Fake News Detection

A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models

AI Data Readiness Inspector (AIDRIN) for Quantitative Assessment of Data Readiness for AI

Nondeterministic Causal Models

X-SHIELD: Regularization for eXplainable Artificial Intelligence

Categorical semantics of compositional reinforcement learning

Optimus-2: Multimodal Minecraft Agent with Goal-Observation-Action Conditioned Policy

Created by

Haebom

저자

Zaijing Li, Yuquan Xie, Rui Shao, Gongwei Chen, Dongmei Jiang, Liqiang Nie

개요

Optimus-2는 다양한 오픈월드 작업에서 인간 행동 패턴을 모방할 수 있는 에이전트를 구축하기 위한 연구의 결과물입니다. 고차원 계획을 위한 다중 모달 대규모 언어 모델(MLLM)과 저차원 제어를 위한 목표-관찰-행동 조건 정책(GOAP)을 통합하여 다양한 작업에 걸쳐 행동 패턴을 효과적으로 학습할 수 있도록 합니다. GOAP는 각 시간 단계에서 관찰과 행동 간의 인과 관계를 모델링하고, 이력 관찰-행동 시퀀스와 동적으로 상호 작용하여 고정 길이 행동 토큰으로 통합하는 행동 안내 행동 인코더와 행동 토큰을 열린 끝 언어 지침과 정렬하여 자기 회귀적으로 행동을 예측하는 MLLM으로 구성됩니다. 또한, 8가지 원자적 작업에 걸쳐 25,000개의 비디오를 포함하고 약 3000만 개의 목표-관찰-행동 쌍을 제공하는 고품질 Minecraft 목표-관찰-행동(MGOA) 데이터 세트를 도입했습니다. 자동화된 구성 방법과 MGOA 데이터 세트는 Minecraft 에이전트를 훈련하기 위한 커뮤니티의 노력에 기여할 수 있습니다. 광범위한 실험 결과는 Optimus-2가 Minecraft에서 원자적 작업, 장기 작업 및 개방형 지침 작업에서 우수한 성능을 보임을 보여줍니다.

시사점, 한계점

•

시사점:

◦

다중 모달 대규모 언어 모델과 목표-관찰-행동 조건 정책을 결합하여 복잡한 오픈월드 작업에서 인간과 유사한 행동을 보이는 에이전트를 구축하는 새로운 방법 제시.

◦

고품질 Minecraft 목표-관찰-행동(MGOA) 데이터셋 공개를 통한 Minecraft 에이전트 연구 발전에 기여.

◦

원자적 작업, 장기 작업, 개방형 지침 작업에서 우수한 성능을 입증.

•

한계점:

◦

MGOA 데이터셋의 규모 및 다양성에 대한 추가적인 개선 여지 존재.

◦

실제 세계의 복잡성을 완전히 반영하지 못할 가능성.

◦

에이전트의 일반화 능력에 대한 추가적인 연구 필요.

Made with SlashPage