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.

DNN-Based Precoding in RIS-Aided mmWave MIMO Systems With Practical Phase Shift

Holistic Tokenizer for Autoregressive Image Generation

MAGIC: Mask-Guided Diffusion Inpainting with Multi-Level Perturbations and Context-Aware Alignment for Few-Shot Anomaly Generation

Order Acquisition Under Competitive Pressure: A Rapidly Adaptive Reinforcement Learning Approach for Ride-Hailing Subsidy Strategies

Towards a Playground to Democratize Experimentation and Benchmarking of AI Agents for Network Troubleshooting

AC-DiT: Adaptive Coordination Diffusion Transformer for Mobile Manipulation

GPU-based complete search for nonlinear minimization subject to bounds

Text Detoxification: Data Efficiency, Semantic Preservation and Model Generalization

On-Policy Optimization of ANFIS Policies Using Proximal Policy Optimization

Robotic Manipulation by Imitating Generated Videos Without Physical Demonstrations

LD-RPS: Zero-Shot Unified Image Restoration via Latent Diffusion Recurrent Posterior Sampling

Horus: A Protocol for Trustless Delegation Under Uncertainty

Visual Anagrams Reveal Hidden Differences in Holistic Shape Processing Across Vision Models

Diversity Conscious Refined Random Forest

GLU Attention Improve Transformer

Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections

CRISP-SAM2: SAM2 with Cross-Modal Interaction and Semantic Prompting for Multi-Organ Segmentation

Treatment, evidence, imitation, and chat

EFRame: Deeper Reasoning via Exploration-Filter-Replay Reinforcement Learning Framework

Towards Understanding the Cognitive Habits of Large Reasoning Models

SEAL: Vision-Language Model-Based Safe End-to-End Cooperative Autonomous Driving with Adaptive Long-Tail Modeling

Inside you are many wolves: Using cognitive models to interpret value trade-offs in LLMs

AI and Agile Software Development: From Frustration to Success -- XP2025 Workshop Summary

Position: Machine Learning Conferences Should Establish a "Refutations and Critiques" Track

Why Do Open-Source LLMs Struggle with Data Analysis? A Systematic Empirical Study

NOVA: Navigation via Object-Centric Visual Autonomy for High-Speed Target Tracking in Unstructured GPS-Denied Environments

Comparative Evaluation of ChatGPT and DeepSeek Across Key NLP Tasks: Strengths, Weaknesses, and Domain-Specific Performance

RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models

Human2LocoMan: Learning Versatile Quadrupedal Manipulation with Human Pretraining

Casper: Inferring Diverse Intents for Assistive Teleoperation with Vision Language Models

Discrete Diffusion in Large Language and Multimodal Models: A Survey

Robust Molecular Property Prediction via Densifying Scarce Labeled Data

LearnAlign: Reasoning Data Selection for Reinforcement Learning in Large Language Models Based on Improved Gradient Alignment

AutoMind: Adaptive Knowledgeable Agent for Automated Data Science

HPCTransCompile: An AI Compiler Generated Dataset for High-Performance CUDA Transpilation and LLM Preliminary Exploration

Towards Practical Alzheimer's Disease Diagnosis: A Lightweight and Interpretable Spiking Neural Model

The Geometries of Truth Are Orthogonal Across Tasks

TrajFlow: Multi-modal Motion Prediction via Flow Matching

Graph-Assisted Stitching for Offline Hierarchical Reinforcement Learning

Making a Pipeline Production-Ready: Challenges and Lessons Learned in the Healthcare Domain

FinBERT2: A Specialized Bidirectional Encoder for Bridging the Gap in Finance-Specific Deployment of Large Language Models

Recommender systems, stigmergy, and the tyranny of popularity

HAVIR: HierArchical Vision to Image Reconstruction using CLIP-Guided Versatile Diffusion

SLAC: Simulation-Pretrained Latent Action Space for Whole-Body Real-World RL

RewardAnything: Generalizable Principle-Following Reward Models

BiMa: Towards Biases Mitigation for Text-Video Retrieval via Scene Element Guidance

A Quantum Information Theoretic Approach to Tractable Probabilistic Models

Playing with Transformer at 30+ FPS via Next-Frame Diffusion

SC-LoRA: Balancing Efficient Fine-tuning and Knowledge Preservation via Subspace-Constrained LoRA

Deep Retrieval at CheckThat! 2025: Identifying Scientific Papers from Implicit Social Media Mentions via Hybrid Retrieval and Re-Ranking

Extended Inductive Reasoning for Personalized Preference Inference from Behavioral Signals

Better Safe Than Sorry? Overreaction Problem of Vision Language Models in Visual Emergency Recognition

Adaptive Inference-Time Scaling via Cyclic Diffusion Search

Domain Adaptation of VLM for Soccer Video Understanding

Towards Universal Semantics With Large Language Models

ReviewInstruct: A Review-Driven Multi-Turn Conversations Generation Method for Large Language Models

An Exploration of Default Images in Text-to-Image Generation

Aggregating Concepts of Fairness and Accuracy in Prediction Algorithms

PRUNE: A Patching Based Repair Framework for Certifiable Unlearning of Neural Networks

Enhancing Satellite Object Localization with Dilated Convolutions and Attention-aided Spatial Pooling

Explainable Coarse-to-Fine Ancient Manuscript Duplicates Discovery

Rewriting Pre-Training Data Boosts LLM Performance in Math and Code

The Role of Open-Source LLMs in Shaping the Future of GeoAI

EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models

AI for the Open-World: the Learning Principles

Towards Explainable Fusion and Balanced Learning in Multimodal Sentiment Analysis

TerraMind: Large-Scale Generative Multimodality for Earth Observation

NativQA Framework: Enabling LLMs with Native, Local, and Everyday Knowledge

Breach in the Shield: Unveiling the Vulnerabilities of Large Language Models

Quantifying Robustness: A Benchmarking Framework for Deep Learning Forecasting in Cyber-Physical Systems

Enhancing Personalized Multi-Turn Dialogue with Curiosity Reward

EAP4EMSIG -- Enhancing Event-Driven Microscopy for Microfluidic Single-Cell Analysis

UNITYAI-GUARD: Pioneering Toxicity Detection Across Low-Resource Indian Languages

CMD-HAR: Cross-Modal Disentanglement for Wearable Human Activity Recognition

Construction Identification and Disambiguation Using BERT: A Case Study of NPN

PVChat: Personalized Video Chat with One-Shot Learning

Neural Discrete Token Representation Learning for Extreme Token Reduction in Video Large Language Models

MASS: Mathematical Data Selection via Skill Graphs for Pretraining Large Language Models

Long Context Modeling with Ranked Memory-Augmented Retrieval

Probing Latent Subspaces in LLM for AI Security: Identifying and Manipulating Adversarial States

Language Models can Self-Improve at State-Value Estimation for Better Search

Offline RLAIF: Piloting VLM Feedback for RL via SFO

Integrating Biological and Machine Intelligence: Attention Mechanisms in Brain-Computer Interfaces

Graded Neural Networks

Do LLMs Understand the Safety of Their Inputs? Training-Free Moderation via Latent Prototypes

FairFare: A Tool for Crowdsourcing Rideshare Data to Empower Labor Organizers

Federated Continual Learning: Concepts, Challenges, and Solutions

UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation

Code Simulation as a Proxy for High-order Tasks in Large Language Models

An Efficient Local Search Approach for Polarized Community Discovery in Signed Networks

Learning Traffic Anomalies from Generative Models on Real-Time Observations

Explainable AI for Mental Health Emergency Returns: Integrating LLMs with Predictive Modeling

In-Context Meta LoRA Generation

Rethinking Table Instruction Tuning

Autonomous Microscopy Experiments through Large Language Model Agents

Static Segmentation by Tracking: A Label-Efficient Approach for Fine-Grained Specimen Image Segmentation

Random weights of DNNs and emergence of fixed points

On the Expressiveness and Length Generalization of Selective State-Space Models on Regular Languages

Mask Approximation Net: A Novel Diffusion Model Approach for Remote Sensing Change Captioning

Enhancing Long Video Generation Consistency without Tuning

The Leaderboard Illusion

Created by

Haebom

저자

Shivalika Singh, Yiyang Nan, Alex Wang, Daniel D'Souza, Sayash Kapoor, Ahmet Ustun, Sanmi Koyejo, Yuntian Deng, Shayne Longpre, Noah A. Smith, Beyza Ermis, Marzieh Fadaee, Sara Hooker

개요

본 논문은 Chatbot Arena의 순위 시스템이 내재적인 문제점으로 인해 왜곡된 경쟁 환경을 조성하고 있음을 밝힙니다. 특히 일부 제공업체들이 비공개 테스트를 통해 여러 변종을 테스트하고, 원하는 점수만 공개하는 등의 행위로 인해 편향된 결과가 생성된다는 것을 지적합니다. Meta의 Llama-4 출시 전 27개의 비공개 LLM 변종 테스트 사례를 예시로 들며, 독점 모델이 공개 모델보다 더 많은 배틀을 치르고 제거될 확률이 낮다는 점을 강조합니다. 이러한 불균형적인 데이터 접근으로 Google과 OpenAI와 같은 대기업이 Arena 데이터의 상당 부분을 독점하고 있으며, 이는 Arena 특화 과적합을 야기하고 실제 모델 성능과의 괴리를 심화시킨다고 주장합니다. 결론적으로 Chatbot Arena의 평가 프레임워크 개혁을 위한 실행 가능한 권장 사항을 제시합니다.

시사점, 한계점

•

시사점:

◦

Chatbot Arena의 순위 시스템의 편향성과 불투명성을 폭로함으로써 AI 모델 평가의 공정성 문제를 제기합니다.

◦

비공개 테스트 및 선택적 점수 공개가 AI 모델 순위에 미치는 영향을 정량적으로 분석합니다.

◦

데이터 접근의 불균형이 모델 성능 평가에 미치는 영향을 보여줍니다.

◦

Chatbot Arena의 평가 프레임워크 개선을 위한 구체적인 권고안을 제시합니다.

•

한계점:

◦

본 논문에서 제시된 권고안의 실효성 및 구현 가능성에 대한 추가적인 검토가 필요합니다.

◦

다른 AI 모델 평가 플랫폼에도 유사한 문제가 존재할 가능성을 고려해야 합니다.

◦

분석에 사용된 데이터의 한계 및 추정치의 신뢰도에 대한 추가적인 설명이 필요합니다.

Made with Slashpage