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.

Grasp Any Region: Towards Precise, Contextual Pixel Understanding for Multimodal LLMs

ScaleNet: Scaling up Pretrained Neural Networks with Incremental Parameters

Text or Pixels? It Takes Half: On the Token Efficiency of Visual Text Inputs in Multimodal LLMs

The Emergence of Complex Behavior in Large-Scale Ecological Environments

PLAGUE: Plug-and-play framework for Lifelong Adaptive Generation of Multi-turn Exploits

Multi-Agent Design Assistant for the Simulation of Inertial Fusion Energy

MUG-V 10B: High-efficiency Training Pipeline for Large Video Generation Models

Context-Aware Pseudo-Label Scoring for Zero-Shot Video Summarization

SparseWorld: A Flexible, Adaptive, and Efficient 4D Occupancy World Model Powered by Sparse and Dynamic Queries

Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey

Agentic Inequality

Natural Language Processing for Cardiology: A Narrative Review

Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs

VERA-MH Concept Paper

The Coverage Principle: How Pre-Training Enables Post-Training

LeapFactual: Reliable Visual Counterfactual Explanation Using Conditional Flow Matching

Mathematics with large language models as provers and verifiers

Coordinated Strategies in Realistic Air Combat by Hierarchical Multi-Agent Reinforcement Learning

Learning What Matters: Steering Diffusion via Spectrally Anisotropic Forward Noise

dInfer: An Efficient Inference Framework for Diffusion Language Models

gLSTM: Mitigating Over-Squashing by Increasing Storage Capacity

BlockGPT: Spatio-Temporal Modelling of Rainfall via Frame-Level Autoregression

Improving Metacognition and Uncertainty Communication in Language Models

ALHD: A Large-Scale and Multigenre Benchmark Dataset for Arabic LLM-Generated Text Detection

ACT: Agentic Classification Tree

Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment

SPiDR: A Simple Approach for Zero-Shot Safety in Sim-to-Real Transfer

Training-Free Label Space Alignment for Universal Domain Adaptation

Conformal Prediction for Time-series Forecasting with Change Points

Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning

Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General LLM Reasoning

Who's Asking? Investigating Bias Through the Lens of Disability Framed Queries in LLMs

MeanAudio: Fast and Faithful Text-to-Audio Generation with Mean Flows

Explaining Time Series Classifiers with PHAR: Rule Extraction and Fusion from Post-hoc Attributions

RoboMemory: A Brain-inspired Multi-memory Agentic Framework for Interactive Environmental Learning in Physical Embodied Systems

Where are we with calibration under dataset shift in image classification?

On the Effectiveness of Methods and Metrics for Explainable AI in Remote Sensing Image Scene Classification

PRING: Rethinking Protein-Protein Interaction Prediction from Pairs to Graphs

PULSE: Practical Evaluation Scenarios for Large Multimodal Model Unlearning

Pay Attention to Small Weights

Understanding Reasoning in Thinking Language Models via Steering Vectors

Chiron-o1: Igniting Multimodal Large Language Models towards Generalizable Medical Reasoning via Mentor-Intern Collaborative Search

With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You

Improved Exploration in GFlownets via Enhanced Epistemic Neural Networks

One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution

Flexible-length Text Infilling for Discrete Diffusion Models

GUARD: Guided Unlearning and Retention via Data Attribution for Large Language Models

TimeWak: Temporal Chained-Hashing Watermark for Time Series Data

Horizon Reduction Makes RL Scalable

QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training

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

Unlearned but Not Forgotten: Data Extraction after Exact Unlearning in LLM

MLR-Bench: Evaluating AI Agents on Open-Ended Machine Learning Research

High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction

Evaluating NLP Embedding Models for Handling Science-Specific Symbolic Expressions in Student Texts

An Analysis of Concept Bottleneck Models: Measuring, Understanding, and Mitigating the Impact of Noisy Annotations

Improving planning and MBRL with temporally-extended actions

CtrlDiff: Boosting Large Diffusion Language Models with Dynamic Block Prediction and Controllable Generation

Concept-Guided Interpretability via Neural Chunking

A Multi-Task Foundation Model for Wireless Channel Representation Using Contrastive and Masked Autoencoder Learning

Memorization-Compression Cycles Improve Generalization

LongCodeBench: Evaluating Coding LLMs at 1M Context Windows

Quantum Natural Language Processing: A Comprehensive Review of Models, Methods, and Applications

Merging Embedded Topics with Optimal Transport for Online Topic Modeling on Data Streams

NAACL2025 Tutorial: Adaptation of Large Language Models

GeoBenchX: Benchmarking LLMs in Agent Solving Multistep Geospatial Tasks

Provably Efficient Reward Transfer in Reinforcement Learning with Discrete Markov Decision Processes

TinySQL: A Progressive Text-to-SQL Dataset for Mechanistic Interpretability Research

PTFA: An LLM-based Agent that Facilitates Online Consensus Building through Parallel Thinking

EgoBlind: Towards Egocentric Visual Assistance for the Blind

Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs

FairGen: Controlling Sensitive Attributes for Fair Generations in Diffusion Models via Adaptive Latent Guidance

Towards Enhanced Image Generation Via Multi-modal Chain of Thought in Unified Generative Models

Probing Perceptual Constancy in Large Vision-Language Models

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

ModServe: Modality- and Stage-Aware Resource Disaggregation for Scalable Multimodal Model Serving

Graph Representation Learning with Diffusion Generative Models

Demystifying Domain-adaptive Post-training for Financial LLMs

Explainable fault and severity classification for rolling element bearings using Kolmogorov-Arnold networks

Open-World Drone Active Tracking with Goal-Centered Rewards

Fast MRI for All: Bridging Access Gaps by Training without Raw Data

Benchmarking Large Language Models with Integer Sequence Generation Tasks

Model-based Large Language Model Customization as Service

Learning Linear Attention in Polynomial Time

Unveiling Transformer Perception by Exploring Input Manifolds

PARCO: Parallel AutoRegressive Models for Multi-Agent Combinatorial Optimization

Estimating Long-term Heterogeneous Dose-response Curve: Generalization Bound Leveraging Optimal Transport Weights

LICO: Large Language Models for In-Context Molecular Optimization

Towards Context-Aware Domain Generalization: Understanding the Benefits and Limits of Marginal Transfer Learning

Embedding in Recommender Systems: A Survey

FP-IRL: Fokker-Planck Inverse Reinforcement Learning -- A Physics-Constrained Approach to Markov Decision Processes

FST.ai 2.0: An Explainable AI Ecosystem for Fair, Fast, and Inclusive Decision-Making in Olympic and Paralympic Taekwondo

A Brain Cell Type Resource Created by Large Language Models and a Multi-Agent AI System for Collaborative Community Annotation

MedRule-KG: A Knowledge-Graph--Steered Scaffold for Mathematical Reasoning with a Lightweight Verifier

The Right to Be Remembered: Preserving Maximally Truthful Digital Memory in the Age of AI

Invoice Information Extraction: Methods and Performance Evaluation

RoboGPT-R1: Enhancing Robot Planning with Reinforcement Learning

ARM-FM: Automated Reward Machines via Foundation Models for Compositional Reinforcement Learning

Base Models Know How to Reason, Thinking Models Learn When

The Open Syndrome Definition

GeoBenchX: Benchmarking LLMs in Agent Solving Multistep Geospatial Tasks

Created by

Haebom

저자

Varvara Krechetova, Denis Kochedykov

개요

본 논문은 상업용 GIS 실무자에게 관련된 다단계 공간 작업을 수행하는 대규모 언어 모델(LLM)의 도구 호출 능력을 평가하기 위한 벤치마크를 제시합니다. 23개의 공간 기능을 갖춘 단순한 도구 호출 에이전트를 사용하여 8개의 상업용 LLM(Claude Sonnet 3.5 및 4, Claude Haiku 3.5, Gemini 2.0 Flash, Gemini 2.5 Pro Preview, GPT-4o, GPT-4.1 및 o4-mini)을 평가합니다. 벤치마크는 난이도가 증가하는 4가지 범주의 작업으로 구성되며, 거부 정확도를 테스트하기 위해 해결 가능한 작업과 의도적으로 해결할 수 없는 작업이 모두 포함됩니다. LLM-as-Judge 평가 프레임워크를 개발하여 에이전트 솔루션을 참조 솔루션과 비교합니다.

시사점, 한계점

•

o4-mini와 Claude 3.5 Sonnet이 전반적으로 가장 우수한 성능을 보임.

•

OpenAI의 GPT-4.1, GPT-4o, Google의 Gemini 2.5 Pro Preview도 근접한 성능을 보였지만, 후자는 해결 불가능한 작업 식별에 더 효율적임.

•

Claude Sonnet 4는 작업을 거부하기보다 어떤 해결책이라도 제공하려는 경향으로 인해 정확도가 낮음.

•

Anthropic 모델은 다른 경쟁 모델보다 더 많은 토큰을 사용함.

•

일반적인 오류로는 기하학적 관계 오해, 오래된 지식 의존, 비효율적인 데이터 조작 등이 있음.

•

결과적으로 벤치마크 세트, 평가 프레임워크 및 데이터 생성 파이프라인은 오픈 소스 리소스로 공개됨(https://github.com/Solirinai/GeoBenchX).

GitHub - Solirinai/GeoBenchX: LLM-agents benchmark set of geospatial tasks requiring multi-step tool use; and LLM-as-Judge based evaluation framework.

LLM-agents benchmark set of geospatial tasks requiring multi-step tool use; and LLM-as-Judge based evaluation framework. - Solirinai/GeoBenchX

Made with Slashpage