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.

Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction

ArchiLense: A Framework for Quantitative Analysis of Architectural Styles Based on Vision Large Language Models

MoE-MLoRA for Multi-Domain CTR Prediction: Efficient Adaptation with Expert Specialization

From Generation to Generalization: Emergent Few-Shot Learning in Video Diffusion Models

Lingshu: A Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning

Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test

TimeWak: Temporal Chained-Hashing Watermark for Time Series Data

How Malicious AI Swarms Can Threaten Democracy

Mutual-Taught for Co-adapting Policy and Reward Models

Efficient Robust Conformal Prediction via Lipschitz-Bounded Networks

Mechanistic Decomposition of Sentence Representations

TextAtari: 100K Frames Game Playing with Language Agents

Why Gradients Rapidly Increase Near the End of Training

Fighting Fire with Fire (F3): A Training-free and Efficient Visual Adversarial Example Purification Method in LVLMs

AnnaAgent: Dynamic Evolution Agent System with Multi-Session Memory for Realistic Seeker Simulation

Supervised Quantum Machine Learning: A Future Outlook from Qubits to Enterprise Applications

Can Slow-thinking LLMs Reason Over Time? Empirical Studies in Time Series Forecasting

ATI: Any Trajectory Instruction for Controllable Video Generation

Flexible Tool Selection through Low-dimensional Attribute Alignment of Vision and Language

Optimized Text Embedding Models and Benchmarks for Amharic Passage Retrieval

How Do Images Align and Complement LiDAR? Towards a Harmonized Multi-modal 3D Panoptic Segmentation

Dual Attention Residual U-Net for Accurate Brain Ultrasound Segmentation in IVH Detection

Mixture of Decoding: An Attention-Inspired Adaptive Decoding Strategy to Mitigate Hallucinations in Large Vision-Language Models

Delving into RL for Image Generation with CoT: A Study on DPO vs. GRPO

Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence

Zero-Shot Gaze-based Volumetric Medical Image Segmentation

DISCO Balances the Scales: Adaptive Domain- and Difficulty-Aware Reinforcement Learning on Imbalanced Data

DisCO: Reinforcing Large Reasoning Models with Discriminative Constrained Optimization

Fusing Bidirectional Chains of Thought and Reward Mechanisms A Method for Enhancing Question-Answering Capabilities of Large Language Models for Chinese Intangible Cultural Heritage

Efficient Fine-Tuning of Quantized Models via Adaptive Rank and Bitwidth

An Analysis of Hyper-Parameter Optimization Methods for Retrieval Augmented Generation

Value Portrait: Assessing Language Models' Values through Psychometrically and Ecologically Valid Items

Scalable Meta-Learning via Mixed-Mode Differentiation

Doxing via the Lens: Revealing Location-related Privacy Leakage on Multi-modal Large Reasoning Models

VIST-GPT: Ushering in the Era of Visual Storytelling with LLMs?

Dual-Individual Genetic Algorithm: A Dual-Individual Approach for Efficient Training of Multi-Layer Neural Networks

Activated LoRA: Fine-tuned LLMs for Intrinsics

On Large-scale Evaluation of Embedding Models for Knowledge Graph Completion

PatchTrAD: A Patch-Based Transformer focusing on Patch-Wise Reconstruction Error for Time Series Anomaly Detection

Optuna vs Code Llama: Are LLMs a New Paradigm for Hyperparameter Tuning?

Quamba2: A Robust and Scalable Post-training Quantization Framework for Selective State Space Models

Evolutionary Policy Optimization

Big Help or Big Brother? Auditing Tracking, Profiling, and Personalization in Generative AI Assistants

Everything Can Be Described in Words: A Simple Unified Multi-Modal Framework with Semantic and Temporal Alignment

Will Neural Scaling Laws Activate Jevons' Paradox in AI Labor Markets? A Time-Varying Elasticity of Substitution (VES) Analysis

Compositional Causal Reasoning Evaluation in Language Models

Towards Autonomous Reinforcement Learning for Real-World Robotic Manipulation with Large Language Models

LLaSE-G1: Incentivizing Generalization Capability for LLaMA-based Speech Enhancement

Self-Training Elicits Concise Reasoning in Large Language Models

Scalable Equilibrium Sampling with Sequential Boltzmann Generators

Self-Supervised Transformers as Iterative Solution Improvers for Constraint Satisfaction

Understand User Opinions of Large Language Models via LLM-Powered In-the-Moment User Experience Interviews

Retrieval-augmented systems can be dangerous medical communicators

On the Privacy Risks of Spiking Neural Networks: A Membership Inference Analysis

Towards Practical First-Order Model Counting

Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning

Comprehensive Review of Neural Differential Equations for Time Series Analysis

TRAVEL: Training-Free Retrieval and Alignment for Vision-and-Language Navigation

Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty

Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies

Zero-shot Meta-learning for Tabular Prediction Tasks with Adversarially Pre-trained Transformer

Calibrated Physics-Informed Uncertainty Quantification

In Praise of Stubbornness: An Empirical Case for Cognitive-Dissonance Aware Continual Update of Knowledge in LLMs

LLM Alignment as Retriever Optimization: An Information Retrieval Perspective

Generative Psycho-Lexical Approach for Constructing Value Systems in Large Language Models

Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defense

Digital Twin Synchronization: Bridging the Sim-RL Agent to a Real-Time Robotic Additive Manufacturing Control

RLHS: Mitigating Misalignment in RLHF with Hindsight Simulation

From Pixels to Predicates: Learning Symbolic World Models via Pretrained Vision-Language Models

A Culturally-Aware Benchmark for Person Re-Identification in Modest Attire

Double Landmines: Invisible Textual Backdoor Attacks based on Dual-Trigger

SnapGen-V: Generating a Five-Second Video within Five Seconds on a Mobile Device

JuStRank: Benchmarking LLM Judges for System Ranking

GRAM: Generalization in Deep RL with a Robust Adaptation Module

From Language Models over Tokens to Language Models over Characters

Tight Lower Bounds and Improved Convergence in Performative Prediction

Towards a Mechanistic Explanation of Diffusion Model Generalization

The BS-meter: A ChatGPT-Trained Instrument to Detect Sloppy Language-Games

Developer Perspectives on Licensing and Copyright Issues Arising from Generative AI for Software Development

DiffLM: Controllable Synthetic Data Generation via Diffusion Language Models

Length-Induced Embedding Collapse in PLM-based Models

Phonology-Guided Speech-to-Speech Translation for African Languages

xGen-MM-Vid (BLIP-3-Video): You Only Need 32 Tokens to Represent a Video Even in VLMs

FlickerFusion: Intra-trajectory Domain Generalizing Multi-Agent RL

Enhancing Retrieval-Augmented Audio Captioning with Generation-Assisted Multimodal Querying and Progressive Learning

Positional Attention: Expressivity and Learnability of Algorithmic Computation

The Causal Information Bottleneck and Optimal Causal Variable Abstractions

Speech to Reality: On-Demand Production using Natural Language, 3D Generative AI, and Discrete Robotic Assembly

The Impact of Large Language Models on Open-source Innovation: Evidence from GitHub Copilot

A Survey of the Self Supervised Learning Mechanisms for Vision Transformers

High-Throughput Phenotyping of Clinical Text Using Large Language Models

An Explainable Vision Transformer with Transfer Learning Combined with Support Vector Machine Based Efficient Drought Stress Identification

Chip Placement with Diffusion Models

Textual Unlearning Gives a False Sense of Unlearning

Scaling Laws in Linear Regression: Compute, Parameters, and Data

AI as Decision-Maker: Ethics and Risk Preferences of LLMs

Exploring the Escalation of Source Bias in User, Data, and Recommender System Feedback Loop

Structuring Concept Space with the Musical Circle of Fifths by Utilizing Music Grammar Based Activations

Innate-Values-driven Reinforcement Learning based Cooperative Multi-Agent Cognitive Modeling

Edge Computing based Human-Robot Cognitive Fusion: A Medical Case Study in the Autism Spectrum Disorder Therapy

Deep Learning for Retinal Degeneration Assessment: A Comprehensive Analysis of the MARIO AMD Progression Challenge

Created by

Haebom

저자

Rachid Zeghlache, Ikram Brahim, Pierre-Henri Conze, Mathieu Lamard, Mohammed El Amine Lazouni, Zineb Aziza Elaouaber, Leila Ryma Lazouni, Christopher Nielsen, Ahmad O. Ahsan, Matthias Wilms, Nils D. Forkert, Lovre Antonio Budimir, Ivana Matovinovic, Donik Vr\v{s}nak, Sven Lon\v{c}aric, Philippe Zhang, Weili Jiang, Yihao Li, Yiding Hao, Markus Frohmann, Patrick Binder, Marcel Huber, Taha Emre, Teresa Finisterra Araujo, Marzieh Oghbaie, Hrvoje Bogunovic, Amerens A. Bekkers, Nina M. van Liebergen, Hugo J. Kuijf, Abdul Qayyum, Moona Mazher, Steven A. Niederer, Alberto J. Beltran-Carrero, Juan J. Gomez-Valverde, Javier Torresano-Rodriquez, Alvaro Caballero-Sastre, Maria J. Ledesma Carbayo, Yosuke Yamagishi, Yi Ding, Robin Peretzke, Alexandra Ertl, Maximilian Fischer, Jessica Kachele, Sofiane Zehar, Karim Boukli Hacene, Thomas Monfort, Beatrice Cochener, Mostafa El Habib Daho, Anas-Alexis Benyoussef, Gwenole Quellec

개요

2024년 MICCAI에서 개최된 MARIO 챌린지는 광학 단층 촬영(OCT) 영상 분석을 통해 연령 관련 황반변성(AMD)의 자동화된 탐지 및 모니터링을 발전시키는 데 중점을 두었습니다. AMD 내 신생혈관 활동 변화 감지를 위한 알고리즘 성능 평가를 위해 고안된 이 챌린지는 독특한 다중 모달 데이터 세트를 통합했습니다. 프랑스 브레스트에서 얻은 주요 데이터 세트는 참가팀이 모델을 훈련하고 테스트하는 데 사용되었습니다. 최종 순위는 이 데이터 세트에 대한 성능을 기반으로 결정되었습니다. 알제리의 보조 데이터 세트는 챌린지 후 제출된 솔루션의 인구 및 장치 변화를 평가하는 데 사용되었습니다. MARIO 챌린지에는 두 가지 과제가 포함되었습니다. 첫 번째 과제는 두 개의 연속적인 2D OCT B-스캔 간의 진행 상황을 분류하는 것이었습니다. 두 번째 과제는 항-혈관 내피 성장 인자(VEGF) 치료를 받는 환자에 대해 3개월 동안의 미래 AMD 진행 상황을 예측하는 것이었습니다. 35개 팀이 참가하여 상위 12개 결선 진출 팀이 자체 방법을 발표했습니다. 본 논문에서는 챌린지의 구조, 과제, 데이터 특성 및 우승 방법론을 설명하고 OCT, 적외선 이미징 및 임상 데이터(방문 횟수, 연령, 성별 등)를 사용한 AMD 모니터링을 위한 벤치마크를 설정합니다. 이 챌린지의 결과는 인공 지능(AI)이 AMD 진행 상황 측정(과제 1)에서 의사와 동등한 수준의 성능을 발휘하지만 아직 미래 진행 상황 예측(과제 2)에는 미흡하다는 것을 보여줍니다.

시사점, 한계점

•

시사점:

◦

OCT, 적외선 영상 및 임상 데이터를 사용한 AMD 모니터링을 위한 새로운 벤치마크를 설정했습니다.

◦

AI가 AMD 진행 상황 측정(과제 1)에서 의사 수준의 성능을 보임을 확인했습니다.

◦

다양한 데이터 세트(프랑스와 알제리)를 사용하여 알고리즘의 일반화 성능을 평가했습니다.

◦

상위 12개 팀의 방법론을 통해 AMD 진행 상황 분석을 위한 다양한 접근 방식을 제시했습니다.

•

한계점:

◦

AI가 AMD의 미래 진행 상황 예측(과제 2)에 있어서는 아직 의사 수준의 성능을 보이지 못했습니다.

◦

사용된 데이터 세트의 제한으로 인해 알고리즘의 일반화 성능에 대한 추가 연구가 필요합니다.

◦

과제 2의 어려움은 AI 기반 AMD 진행 예측 모델 개발의 어려움을 시사합니다.

Made with Slashpage