[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.

Photonic Fabric Platform for AI Accelerators

Achieving Robust Channel Estimation Neural Networks by Designed Training Data

Can Mental Imagery Improve the Thinking Capabilities of AI Systems?

Characterizing State Space Model (SSM) and SSM-Transformer Hybrid Language Model Performance with Long Context Length

PGT-I: Scaling Spatiotemporal GNNs with Memory-Efficient Distributed Training

Robust 3D-Masked Part-level Editing in 3D Gaussian Splatting with Regularized Score Distillation Sampling

A Lightweight and Robust Framework for Real-Time Colorectal Polyp Detection Using LOF-Based Preprocessing and YOLO-v11n

HMID-Net: An Exploration of Masked Image Modeling and Knowledge Distillation in Hyperbolic Space

Synchronizing Task Behavior: Aligning Multiple Tasks during Test-Time Training

Resolving Token-Space Gradient Conflicts: Token Space Manipulation for Transformer-Based Multi-Task Learning

Fast Bilateral Teleoperation and Imitation Learning Using Sensorless Force Control via Accurate Dynamics Model

VisualSpeaker: Visually-Guided 3D Avatar Lip Synthesis

Reviving Cultural Heritage: A Novel Approach for Comprehensive Historical Document Restoration

Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning

Learning Software Bug Reports: A Systematic Literature Review

Rethinking Data Protection in the (Generative) Artificial Intelligence Era

Frequency-Aligned Knowledge Distillation for Lightweight Spatiotemporal Forecasting

TopoStreamer: Temporal Lane Segment Topology Reasoning in Autonomous Driving

"Before, I Asked My Mom, Now I Ask ChatGPT": Visual Privacy Management with Generative AI for Blind and Low-Vision People

QLPro: Automated Code Vulnerability Discovery via LLM and Static Code Analysis Integration

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

Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models

Bridging the Digital Divide: Small Language Models as a Pathway for Physics and Photonics Education in Underdeveloped Regions

DaMO: A Data-Efficient Multimodal Orchestrator for Temporal Reasoning with Video LLMs

Dynamic Context Tuning for Retrieval-Augmented Generation: Enhancing Multi-Turn Planning and Tool Adaptation

Specification and Evaluation of Multi-Agent LLM Systems -- Prototype and Cybersecurity Applications

PhysioWave: A Multi-Scale Wavelet-Transformer for Physiological Signal Representation

Draft-based Approximate Inference for LLMs

Label-semantics Aware Generative Approach for Domain-Agnostic Multilabel Classification

SemiOccam: A Robust Semi-Supervised Image Recognition Network Using Sparse Labels

Adversarial bandit optimization for approximately linear functions

Know Or Not: a library for evaluating out-of-knowledge base robustness

Leveraging Vision-Language Models for Visual Grounding and Analysis of Automotive UI

DualReal: Adaptive Joint Training for Lossless Identity-Motion Fusion in Video Customization

CoordField: Coordination Field for Agentic UAV Task Allocation In Low-altitude Urban Scenarios

Return Capping: Sample-Efficient CVaR Policy Gradient Optimisation

AnyTSR: Any-Scale Thermal Super-Resolution for UAV

Enhanced Pruning Strategy for Multi-Component Neural Architectures Using Component-Aware Graph Analysis

Executable Functional Abstractions: Inferring Generative Programs for Advanced Math Problems

Measuring Leakage in Concept-Based Methods: An Information Theoretic Approach

APIGen-MT: Agentic Pipeline for Multi-Turn Data Generation via Simulated Agent-Human Interplay

The Dual-Route Model of Induction

Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models

SWI: Speaking with Intent in Large Language Models

A Study of LLMs' Preferences for Libraries and Programming Languages

TruthLens: Explainable DeepFake Detection for Face Manipulated and Fully Synthetic Data

Sampling Decisions

Federated Continual Instruction Tuning

Fine-Tuning Diffusion Generative Models via Rich Preference Optimization

BriLLM: Brain-inspired Large Language Model

Studying Classifier(-Free) Guidance From a Classifier-Centric Perspective

RealGeneral: Unifying Visual Generation via Temporal In-Context Learning with Video Models

Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

PLADIS: Pushing the Limits of Attention in Diffusion Models at Inference Time by Leveraging Sparsity

DynamicID: Zero-Shot Multi-ID Image Personalization with Flexible Facial Editability

Symbolic Mixture-of-Experts: Adaptive Skill-based Routing for Heterogeneous Reasoning

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

Too Much to Trust? Measuring the Security and Cognitive Impacts of Explainability in AI-Driven SOCs

Attend or Perish: Benchmarking Attention in Algorithmic Reasoning

Can Optical Denoising Clean Sonar Images? A Benchmark and Fusion Approach

Brain Foundation Models: A Survey on Advancements in Neural Signal Processing and Brain Discovery

Winning Big with Small Models: Knowledge Distillation vs. Self-Training for Reducing Hallucination in Product QA Agents

Detecting Benchmark Contamination Through Watermarking

MEMERAG: A Multilingual End-to-End Meta-Evaluation Benchmark for Retrieval Augmented Generation

Steering into New Embedding Spaces: Analyzing Cross-Lingual Alignment Induced by Model Interventions in Multilingual Language Models

Analyze the Neurons, not the Embeddings: Understanding When and Where LLM Representations Align with Humans

MKE-Coder: Multi-Axial Knowledge with Evidence Verification in ICD Coding for Chinese EMRs

An Overall Real-Time Mechanism for Classification and Quality Evaluation of Rice

Layerwise Recall and the Geometry of Interwoven Knowledge in LLMs

Learning in Strategic Queuing Systems with Small Buffers

BARNN: A Bayesian Autoregressive and Recurrent Neural Network

HEPPO-GAE: Hardware-Efficient Proximal Policy Optimization with Generalized Advantage Estimation

CGP-Tuning: Structure-Aware Soft Prompt Tuning for Code Vulnerability Detection

A recent evaluation on the performance of LLMs on radiation oncology physics using questions of randomly shuffled options

A Survey on Large Language Model-Based Social Agents in Game-Theoretic Scenarios

PEMF-VTO: Point-Enhanced Video Virtual Try-on via Mask-free Paradigm

Understanding the Design Decisions of Retrieval-Augmented Generation Systems

DOGR: Towards Versatile Visual Document Grounding and Referring

Ev2R: Evaluating Evidence Retrieval in Automated Fact-Checking

DualSwinUnet++: An Enhanced Swin-Unet Architecture With Dual Decoders For PTMC Segmentation

PerspectiveNet: Multi-View Perception for Dynamic Scene Understanding

AlphaDPO: Adaptive Reward Margin for Direct Preference Optimization

Continual Learning with Neuromorphic Computing: Foundations, Methods, and Emerging Applications

FlexiTex: Enhancing Texture Generation via Visual Guidance

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

The unknotting number, hard unknot diagrams, and reinforcement learning

Hierarchical Reinforcement Learning for Temporal Abstraction of Listwise Recommendation

Enhancing Natural Language Inference Performance with Knowledge Graph for COVID-19 Automated Fact-Checking in Indonesian Language

CVPT: Cross Visual Prompt Tuning

Proficient Graph Neural Network Design by Accumulating Knowledge on Large Language Models

Stimulating Imagination: Towards General-purpose "Something Something Placement"

Why Does New Knowledge Create Messy Ripple Effects in LLMs?

A Mathematical Framework and a Suite of Learning Techniques for Neural-Symbolic Systems

How to Leverage Predictive Uncertainty Estimates for Reducing Catastrophic Forgetting in Online Continual Learning

Towards the Next Frontier in Speech Representation Learning Using Disentanglement

Hierarchical Prompting Taxonomy: A Universal Evaluation Framework for Large Language Models Aligned with Human Cognitive Principles

Which Experiences Are Influential for RL Agents? Efficiently Estimating The Influence of Experiences

Oversmoothing Alleviation in Graph Neural Networks: A Survey and Unified View

OCK: Unsupervised Dynamic Video Prediction with Object-Centric Kinematics

Benchmarking Mobile Device Control Agents across Diverse Configurations

BEnchmarking LLMs for Ophthalmology (BELO) for Ophthalmological Knowledge and Reasoning

Created by

Haebom

저자

Sahana Srinivasan, Xuguang Ai, Thaddaeus Wai Soon Lo, Aidan Gilson, Minjie Zou, Ke Zou, Hyunjae Kim, Mingjia Yang, Krithi Pushpanathan, Samantha Yew, Wan Ting Loke, Jocelyn Goh, Yibing Chen, Yiming Kong, Emily Yuelei Fu, Michelle Ongyong Hui, Kristen Nwanyanwu, Amisha Dave, Kelvin Zhenghao Li, Chen-Hsin Sun, Mark Chia, Gabriel Dawei Yang, Wendy Meihua Wong, David Ziyou Chen, Dianbo Liu, Maxwell Singer, Fares Antaki, Lucian V Del Priore, Jost Jonas, Ron Adelman, Qingyu Chen, Yih-Chung Tham

개요

안과 분야에서 대규모 언어 모델(LLM)을 평가하는 기존 벤치마크는 범위가 제한적이고 정확도에 지나치게 중점을 둡니다. 본 논문에서는 13명의 안과 전문의가 여러 차례 전문가 검토를 거쳐 개발된 표준화되고 종합적인 평가 벤치마크인 BELO(BEnchmarking LLMs for Ophthalmology)를 제시합니다. BELO는 안과 관련 임상 정확도와 추론 품질을 평가합니다. 키워드 매칭 및 미세 조정된 PubMedBERT 모델을 사용하여 다양한 의료 데이터 세트(BCSC, MedMCQA, MedQA, BioASQ 및 PubMedQA)에서 안과 관련 다지선다형 질문(MCQ)을 엄선했습니다. 데이터 세트는 여러 차례 전문가 검토를 거쳤으며, 중복 및 저품질 질문은 체계적으로 제거되었습니다. 10명의 안과 전문의가 각 MCQ 정답에 대한 설명을 개선했고, 3명의 수석 안과 전문의가 추가로 심사했습니다. BELO의 유용성을 보여주기 위해 정확도, 매크로 F1 및 5가지 텍스트 생성 지표(ROUGE-L, BERTScore, BARTScore, METEOR 및 AlignScore)를 사용하여 6개의 LLM(OpenAI o1, o3-mini, GPT-4o, DeepSeek-R1, Llama-3-8B 및 Gemini 1.5 Pro)을 평가했습니다. 또한, 2명의 안과 전문의가 무작위로 선택된 50개의 출력에 대해 정확성, 포괄성 및 완전성을 질적으로 검토하는 추가 평가를 수행했습니다. BELO는 5가지 출처(BCSC(260개), BioASQ(10개), MedMCQA(572개), MedQA(40개), PubMedQA(18개))에서 집계된 900개의 고품질 전문가 검토 질문으로 구성됩니다. 투명한 평가 및 보고를 장려하기 위해 공개 리더보드가 마련되었으며, BELO 데이터 세트는 향후 모델의 공정하고 재현 가능한 비교를 보장하기 위해 홀드아웃 평가 전용 벤치마크로 유지됩니다.

시사점, 한계점

•

시사점: 안과 분야 LLM 평가를 위한 표준화되고 종합적인 벤치마크 제공, 전문가 검토를 통한 고품질 데이터 세트 구축, 다양한 LLM의 성능 비교 및 투명한 평가 환경 조성, 향후 LLM 개발의 방향 제시.

•

한계점: 현재 벤치마크에 포함된 LLM의 종류가 제한적일 수 있음, 질적 평가 부분의 주관성 개입 가능성, 안과 영역 특수성으로 인한 일반화의 어려움, 데이터 세트의 지속적인 업데이트 및 확장 필요성.

Made with Slashpage