[공지사항]을 빙자한 안부와 근황

Show more

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.

SeC: Advancing Complex Video Object Segmentation via Progressive Concept Construction

GUI-G$^2$: Gaussian Reward Modeling for GUI Grounding

Romance, Relief, and Regret: Teen Narratives of Chatbot Overreliance

Supernova: Achieving More with Less in Transformer Architectures

GR-3 Technical Report

EndoControlMag: Robust Endoscopic Vascular Motion Magnification with Periodic Reference Resetting and Hierarchical Tissue-aware Dual-Mask Contro

AI-Enhanced Precision in Sport Taekwondo: Increasing Fairness, Speed, and Trust in Competition (FST.ai)

Promptomatix: An Automatic Prompt Optimization Framework for Large Language Models

NeuroHD-RA: Neural-distilled Hyperdimensional Model with Rhythm Alignment

IPPRO: Importance-based Pruning with PRojective Offset for Magnitude-indifferent Structural Pruning

Lessons from the TREC Plain Language Adaptation of Biomedical Abstracts (PLABA) track

Seed-X: Building Strong Multilingual Translation LLM with 7B Parameters

Physical models realizing the transformer architecture of large language models

GitChameleon 2.0: Evaluating AI Code Generation Against Python Library Version Incompatibilities

Multimodal Coordinated Online Behavior: Trade-offs and Strategies

A Survey of Deep Learning for Geometry Problem Solving

Prompt4Trust: A Reinforcement Learning Prompt Augmentation Framework for Clinically-Aligned Confidence Calibration in Multimodal Large Language Models

Pre-Training LLMs on a budget: A comparison of three optimizers

OPC: One-Point-Contraction Unlearning Toward Deep Feature Forgetting

Adaptive Gaussian Mixture Models-based Anomaly Detection for under-constrained Cable-Driven Parallel Robots

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving

Omni-Router: Sharing Routing Decisions in Sparse Mixture-of-Experts for Speech Recognition

INTER: Mitigating Hallucination in Large Vision-Language Models by Interaction Guidance Sampling

The Joys of Categorical Conformal Prediction

Towards Accurate and Efficient 3D Object Detection for Autonomous Driving: A Mixture of Experts Computing System on Edge

Capacity Planning and Scheduling for Jobs with Uncertainty in Resource Usage and Duration

FedWSQ: Efficient Federated Learning with Weight Standardization and Distribution-Aware Non-Uniform Quantization

Neural Approaches for Multi-Objective Routing on Multigraphs

Diffusion-Based Electrocardiography Noise Quantification via Anomaly Detection

CogStream: Context-guided Streaming Video Question Answering

Can LLMs Generate Reliable Test Case Generators? A Study on Competition-Level Programming Problems

Towards provable probabilistic safety for scalable embodied AI systems

SenWiCh: Sense-Annotation of Low-Resource Languages for WiC using Hybrid Methods

Human Empathy as Encoder: AI-Assisted Depression Assessment in Special Education

Multimodal Forecasting of Sparse Intraoperative Hypotension Events Powered by Language Model

Autocomp: LLM-Driven Code Optimization for Tensor Accelerators

ViP$^2$-CLIP: Visual-Perception Prompting with Unified Alignment for Zero-Shot Anomaly Detection

ReMi: A Random Recurrent Neural Network Approach to Music Production

Aitomia: Your Intelligent Assistant for AI-Driven Atomistic and Quantum Chemical Simulations

A Goal-Oriented Reinforcement Learning-Based Path Planning Algorithm for Modular Self-Reconfigurable Satellites

A Method for the Architecture of a Medical Vertical Large Language Model Based on Deepseek R1

Balancing Robustness and Efficiency in Embedded DNNs Through Activation Function Selection

Antithetic Sampling for Top-k Shapley Identification

GeoFlow-SLAM: A Robust Tightly-Coupled RGBD-Inertial and Legged Odometry Fusion SLAM for Dynamic Legged Robotics

SciFi-Benchmark: Leveraging Science Fiction To Improve Robot Behavior

ETCH: Generalizing Body Fitting to Clothed Humans via Equivariant Tightness

Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$P Parametrization

Curating Demonstrations using Online Experience

OMNISEC: LLM-Driven Provenance-based Intrusion Detection via Retrieval-Augmented Behavior Prompting

PRISM: High-Resolution & Precise Counterfactual Medical Image Generation using Language-guided Stable Diffusion

Reasoning Does Not Necessarily Improve Role-Playing Ability

BioMaze: Benchmarking and Enhancing Large Language Models for Biological Pathway Reasoning

Revealing Bias Formation in Deep Neural Networks Through the Geometric Mechanisms of Human Visual Decoupling

Conformal Predictions for Human Action Recognition with Vision-Language Models

Cross-Encoder Rediscovers a Semantic Variant of BM25

DisCoPatch: Taming Adversarially-driven Batch Statistics for Improved Out-of-Distribution Detection

RadAlign: Advancing Radiology Report Generation with Vision-Language Concept Alignment

R-Bot: An LLM-based Query Rewrite System

Towards a Universal 3D Medical Multi-modality Generalization via Learning Personalized Invariant Representation

Atomic Calibration of LLMs in Long-Form Generations

LibEER: A Comprehensive Benchmark and Algorithm Library for EEG-based Emotion Recognition

Aligning AI with Public Values: Deliberation and Decision-Making for Governing Multimodal LLMs in Political Video Analysis

VitaGlyph: Vitalizing Artistic Typography with Flexible Dual-branch Diffusion Models

From homeostasis to resource sharing: Biologically and economically aligned multi-objective multi-agent AI safety benchmarks

V-RoAst: Visual Road Assessment. Can VLM be a Road Safety Assessor Using the iRAP Standard?

FLAIN: Mitigating Backdoor Attacks in Federated Learning via Flipping Weight Updates of Low-Activation Input Neurons

FBSDiff: Plug-and-Play Frequency Band Substitution of Diffusion Features for Highly Controllable Text-Driven Image Translation

Analysis of the 2024 BraTS Meningioma Radiotherapy Planning Automated Segmentation Challenge

Unisolver: PDE-Conditional Transformers Are Universal PDE Solvers

LangBiTe: A Platform for Testing Bias in Large Language Models

Practical Insights into Knowledge Distillation for Pre-Trained Models

Risks of AI Scientists: Prioritizing Safeguarding Over Autonomy

Energy-Efficient and Real-Time Sensing for Federated Continual Learning via Sample-Driven Control

Gemini 2.5 Pro Capable of Winning Gold at IMO 2025

Hierarchical Budget Policy Optimization for Adaptive Reasoning

BioGraphFusion: Graph Knowledge Embedding for Biological Completion and Reasoning

Routine: A Structural Planning Framework for LLM Agent System in Enterprise

CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning

Assessing Adaptive World Models in Machines with Novel Games

A Multi-granularity Concept Sparse Activation and Hierarchical Knowledge Graph Fusion Framework for Rare Disease Diagnosis

An Integrated Framework of Prompt Engineering and Multidimensional Knowledge Graphs for Legal Dispute Analysis

Hierarchical Reasoning Model

The Unified Cognitive Consciousness Theory for Language Models: Anchoring Semantics, Thresholds of Activation, and Emergent Reasoning

DCG-SQL: Enhancing In-Context Learning for Text-to-SQL with Deep Contextual Schema Link Graph

InternAgent: When Agent Becomes the Scientist -- Building Closed-Loop System from Hypothesis to Verification

R2D2: Remembering, Replaying and Dynamic Decision Making with a Reflective Agentic Memory

Efficient Strategy Learning by Decoupling Searching and Pathfinding for Object Navigation

Alto: Orchestrating Distributed Compound AI Systems with Nested Ancestry

Toward A Causal Framework for Modeling Perception

ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning

MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning

Rethinking LLM-Based RTL Code Optimization Via Timing Logic Metamorphosis

Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty

Decoding Translation-Related Functional Sequences in 5'UTRs Using Interpretable Deep Learning Models

Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning

Never Come Up Empty: Adaptive HyDE Retrieval for Improving LLM Developer Support

AI-enhanced conversational agents for personalized asthma support Factors for engagement, value and efficacy

RAVine: Reality-Aligned Evaluation for Agentic Search

Experience is the Best Teacher: Grounding VLMs for Robotics through Self-Generated Memory

Gemini 2.5 Pro Capable of Winning Gold at IMO 2025

Created by

Haebom

저자

Yichen Huang, Lin F. Yang

개요

본 논문은 거대 언어 모델(LLM)인 Google의 Gemini 2.5 Pro를 사용하여 2025년 국제 수학 올림피아드(IMO) 문제 6개 중 5개를 정답을 도출했다는 연구 결과를 제시합니다. IMO 문제는 심오한 통찰력, 창의력, 형식적 추론을 요구하는 독특하고 어려운 문제로, 기존 LLM들은 풀이에 어려움을 겪는 것으로 알려져 있습니다. 본 연구는 데이터 오염을 피하기 위해 최신 IMO 문제를 사용하였고, 신중한 프롬프트 설계와 자체 검증 파이프라인을 통해 높은 정확도를 달성했습니다. 이는 복잡한 추론 작업에 강력한 LLM의 잠재력을 최대한 활용하기 위한 최적 전략 개발의 중요성을 강조합니다.

시사점, 한계점

•

시사점:

◦

강력한 LLM이 복잡한 수학적 추론 문제 해결에 상당한 잠재력을 가지고 있음을 보여줍니다.

◦

최적의 프롬프트 설계 및 자체 검증 파이프라인이 LLM의 성능 향상에 중요한 역할을 함을 시사합니다.

◦

LLM을 활용한 수학 문제 해결 전략 연구의 필요성을 강조합니다.

•

한계점:

◦

6개 문제 중 1개 문제는 정답을 도출하지 못했습니다. (자세한 내용은 논문 참고)

◦

사용된 LLM의 성능이 특정 문제 유형에 편향될 가능성이 있습니다.

◦

다른 LLM이나 다른 유형의 수학 문제에 대한 일반화 가능성에 대한 추가 연구가 필요합니다.

Made with Slashpage