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.

OmniEVA: Embodied Versatile Planner via Task-Adaptive 3D-Grounded and Embodiment-aware Reasoning

Open-sci-ref-0.01: open and reproducible reference baselines for language model and dataset comparison

Similarity-based Outlier Detection for Noisy Object Re-Identification Using Beta Mixtures

MachineLearningLM: Scaling Many-shot In-context Learning via Continued Pretraining

Musculoskeletal simulation of limb movement biomechanics in Drosophila melanogaster

Between a Rock and a Hard Place: Exploiting Ethical Reasoning to Jailbreak LLMs

Cut Costs, Not Accuracy: LLM-Powered Data Processing with Guarantees

The Architecture of AI Transformation: Four Strategic Patterns and an Emerging Frontier

IS${}^3$ : Generic Impulsive--Stationary Sound Separation in Acoustic Scenes using Deep Filtering

Beyond Ensembles: Simulating All-Atom Protein Dynamics in a Learned Latent Space

HiddenObject: Modality-Agnostic Fusion for Multimodal Hidden Object Detection

Input-Time Scaling

Counterfactual Probabilistic Diffusion with Expert Models

Feedback-Driven Tool-Use Improvements in Large Language Models via Automated Build Environments

Bridging the Gap: A Framework for Real-World Video Deepfake Detection via Social Network Compression Emulation

Imposing AI: Deceptive design patterns against sustainability

LMAR: Language Model Augmented Retriever for Domain-specific Knowledge Indexing

Data-Driven Discovery of Mobility Periodicity for Understanding Urban Systems

Privacy Risks of LLM-Empowered Recommender Systems: An Inversion Attack Perspective

Web3 x AI Agents: Landscape, Integrations, and Foundational Challenges

Atherosclerosis through Hierarchical Explainable Neural Network Analysis

Survivability of Backdoor Attacks on Unconstrained Face Recognition Systems

Geological Everything Model 3D: A Promptable Foundation Model for Unified and Zero-Shot Subsurface Understanding

HiLight: A Hierarchical Reinforcement Learning Framework with Global Adversarial Guidance for Large-Scale Traffic Signal Control

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

OMGM: Orchestrate Multiple Granularities and Modalities for Efficient Multimodal Retrieval

The Precautionary Principle and the Innovation Principle: Incompatible Guides for AI Innovation Governance?

A Framework for Testing and Adapting REST APIs as LLM Tools

D\'ej\`a Vu: Multilingual LLM Evaluation through the Lens of Machine Translation Evaluation

JARVIS-VLA: Post-Training Large-Scale Vision Language Models to Play Visual Games with Keyboards and Mouse

Prompt Programming: A Platform for Dialogue-based Computational Problem Solving with Generative AI Models

Auxiliary Discrminator Sequence Generative Adversarial Networks (ADSeqGAN) for Few Sample Molecule Generation

Neural Force Field: Few-shot Learning of Generalized Physical Reasoning

Towards Developing Socially Compliant Automated Vehicles: Advances, Expert Insights, and A Conceptual Framework

A Novel Approach to Balance Convenience and Nutrition in Meals With Long-Term Group Recommendations and Reasoning on Multimodal Recipes and its Implementation in BEACON

Can LLM Prompting Serve as a Proxy for Static Analysis in Vulnerability Detection

Tokens, the oft-overlooked appetizer: Large language models, the distributional hypothesis, and meaning

Polish-English medical knowledge transfer: A new benchmark and results

InterFormer: Effective Heterogeneous Interaction Learning for Click-Through Rate Prediction

AdvI2I: Adversarial Image Attack on Image-to-Image Diffusion models

A Survey on Group Fairness in Federated Learning: Challenges, Taxonomy of Solutions and Directions for Future Research

A Conflicts-free, Speed-lossless KAN-based Reinforcement Learning Decision System for Interactive Driving in Roundabouts

The Overcooked Generalisation Challenge: Evaluating Cooperation with Novel Partners in Unknown Environments Using Unsupervised Environment Design

Modelling the 5G Energy Consumption using Real-world Data: Energy Fingerprint is All You Need

Slaves to the Law of Large Numbers: An Asymptotic Equipartition Property for Perplexity in Generative Language Models

Interpretable Data-driven Anomaly Detection in Industrial Processes with ExIFFI

Analyzing the Impact of Adversarial Examples on Explainable Machine Learning

TSGCNeXt: Dynamic-Static Multi-Graph Convolution for Efficient Skeleton-Based Action Recognition with Long-term Learning Potential

Sufficient Invariant Learning for Distribution Shift

TORSO: Template-Oriented Reasoning Towards General Tasks

ForTIFAI: Fending Off Recursive Training Induced Failure for AI Models

Oyster-I: Beyond Refusal -- Constructive Safety Alignment for Responsible Language Models

Building Self-Evolving Agents via Experience-Driven Lifelong Learning: A Framework and Benchmark

QuantX: A Framework for Hardware-Aware Quantization of Generative AI Workloads

Learning to Plan with Personalized Preferences

Multi-Turn Human-LLM Interaction Through the Lens of a Two-Way Intelligibility Protocol

Spatio-Temporal Graphical Counterfactuals: An Overview

Standards in the Preparation of Biomedical Research Metadata: A Bridge2AI Perspective

Is In-Context Learning Learning?

Multimodal SAM-adapter for Semantic Segmentation

Diversified recommendations of cultural activities with personalized determinantal point processes

Improving Audio Event Recognition with Consistency Regularization

Data distribution impacts the performance and generalisability of contrastive learning-based foundation models of electrocardiograms

Towards Understanding Visual Grounding in Visual Language Models

GLAM: Geometry-Guided Local Alignment for Multi-View VLP in Mammography

I-Segmenter: Integer-Only Vision Transformer for Efficient Semantic Segmentation

Generalizing Beyond Suboptimality: Offline Reinforcement Learning Learns Effective Scheduling through Random Data

We Need a New Ethics for a World of AI Agents

SignClip: Leveraging Mouthing Cues for Sign Language Translation by Multimodal Contrastive Fusion

Openness in AI and downstream governance: A global value chain approach

SI-FACT: Mitigating Knowledge Conflict via Self-Improving Faithfulness-Aware Contrastive Tuning

Benchmark of stylistic variation in LLM-generated texts

BenchECG and xECG: a benchmark and baseline for ECG foundation models

Efficient Learning-Based Control of a Legged Robot in Lunar Gravity

Population-Aligned Persona Generation for LLM-based Social Simulation

Realism Control One-step Diffusion for Real-World Image Super-Resolution

Generating Energy-Efficient Code via Large-Language Models -- Where are we now?

Established Psychometric vs. Ecologically Valid Questionnaires: Rethinking Psychological Assessments in Large Language Models

Predictive Spike Timing Enables Distributed Shortest Path Computation in Spiking Neural Networks

TwinTac: A Wide-Range, Highly Sensitive Tactile Sensor with Real-to-Sim Digital Twin Sensor Model

Multimodal Mathematical Reasoning Embedded in Aerial Vehicle Imagery: Benchmarking, Analysis, and Exploration

Reinforcement learning for spin torque oscillator tasks

Exploring Expert Specialization through Unsupervised Training in Sparse Mixture of Experts

Intrinsic Dimension Estimating Autoencoder (IDEA) Using CancelOut Layer and a Projected Loss

Unsupervised Hallucination Detection by Inspecting Reasoning Processes

Drone-Based Multispectral Imaging and Deep Learning for Timely Detection of Branched Broomrape in Tomato Farms

Securing LLM-Generated Embedded Firmware through AI Agent-Driven Validation and Patching

Large Language Models Meet Legal Artificial Intelligence: A Survey

Limited Reference, Reliable Generation: A Two-Component Framework for Tabular Data Generation in Low-Data Regimes

Zero-Shot Referring Expression Comprehension via Visual-Language True/False Verification

Adaptive Token Merging for Efficient Transformer Semantic Communication at the Edge

SmartCoder-R1: Towards Secure and Explainable Smart Contract Generation with Security-Aware Group Relative Policy Optimization

WALL: A Web Application for Automated Quality Assurance using Large Language Models

An Autoencoder and Vision Transformer-based Interpretability Analysis of the Differences in Automated Staging of Second and Third Molars

Tackling One Health Risks: How Large Language Models are leveraged for Risk Negotiation and Consensus-building

Self-Augmented Robot Trajectory: Efficient Imitation Learning via Safe Self-augmentation with Demonstrator-annotated Precision

Automated Tuning for Diffusion Inverse Problem Solvers without Generative Prior Retraining

From Hugging Face to GitHub: Tracing License Drift in the Open-Source AI Ecosystem

Emulating Public Opinion: A Proof-of-Concept of AI-Generated Synthetic Survey Responses for the Chilean Case

Vibe Check: Understanding the Effects of LLM-Based Conversational Agents' Personality and Alignment on User Perceptions in Goal-Oriented Tasks

Slaves to the Law of Large Numbers: An Asymptotic Equipartition Property for Perplexity in Generative Language Models

Created by

Haebom

저자

Tyler Bell, Avinash Mudireddy, Ivan Johnson-Eversoll, Soura Dasgupta, Raghu Mudumbai

개요

본 논문은 언어 모델이 생성하는 긴 텍스트의 perplexity에 대한 새로운 점근적 비균등 분할(un-equipartition) 특성을 증명하고, 오픈소스 모델을 이용한 실험적 증거를 제시합니다. 구체적으로, 언어 모델이 생성하는 모든 큰 텍스트의 로그 perplexity는 토큰 분포의 평균 엔트로피에 점근적으로 수렴함을 보입니다. 이는 언어 모델이 생성하는 모든 긴 합성 텍스트가 속해야 하는 "전형적인 집합(typical set)"을 정의합니다. 논문은 문법적으로 정확한 텍스트만 포함하도록 "전형적인 집합"의 개념을 세분화하고, 매우 일반적인 문법 정의 하에서 이러한 세분화된 전형적인 집합이 모든 가능한 문법적으로 정확한 텍스트의 매우 작은 부분집합임을 보입니다. 즉, 언어 모델은 가능한 동작 및 출력 범위에 강하게 제약을 받습니다. 본 연구는 언어 모델 출력의 통계에 대한 정상성과 같은 단순화된 가정을 하지 않으므로, 근사 없이 실제 모델에 직접 적용 가능합니다. 합성 텍스트 탐지 및 훈련 데이터셋의 멤버십 추론과 같은 문제에 대한 전형적인 집합 개념의 가능한 응용에 대해 논의합니다.

시사점, 한계점

•

시사점:

◦

언어 모델의 출력이 평균 엔트로피에 수렴하는 점근적 비균등 분할 특성을 밝힘으로써, 언어 모델의 출력 가능성에 대한 새로운 이해를 제공합니다.

◦

언어 모델이 생성할 수 있는 텍스트의 범위가 제한적임을 밝힘으로써, 합성 텍스트 탐지 및 멤버십 추론 등의 응용 분야에 활용 가능성을 제시합니다.

◦

단순화된 가정 없이 실제 모델에 직접 적용 가능한 결과를 제시합니다.

•

한계점:

◦

본 연구의 실험적 증거는 오픈소스 모델에 국한되어 있으며, 다양한 모델과 데이터셋에 대한 추가적인 실험이 필요합니다.

◦

"문법"의 정의가 매우 일반적이기 때문에, 특정 언어나 응용 분야에 대한 구체적인 분석이 추가적으로 필요할 수 있습니다.

◦

전형적인 집합의 크기와 경계에 대한 더 정교한 분석이 필요할 수 있습니다.

Made with Slashpage