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.

VGR: Visual Grounded Reasoning

Improving Multimodal Learning Balance and Sufficiency through Data Remixing

Autonomous Computer Vision Development with Agentic AI

A Self-Refining Framework for Enhancing ASR Using TTS-Synthesized Data

Enabling On-Device Medical AI Assistants via Input-Driven Saliency Adaptation

Task-aligned prompting improves zero-shot detection of AI-generated images by Vision-Language Models

From Reasoning to Code: GRPO Optimization for Underrepresented Languages

Farseer: A Refined Scaling Law in Large Language Models

Monitoring Decomposition Attacks in LLMs with Lightweight Sequential Monitors

VideoDeepResearch: Long Video Understanding With Agentic Tool Using

Data Shifts Hurt CoT: A Theoretical Study

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

Extended Creativity: A Conceptual Framework for Understanding Human-AI Creative Relations

Disclosure Audits for LLM Agents

A Survey of Generative Categories and Techniques in Multimodal Large Language Models

Resa: Transparent Reasoning Models via SAEs

EmoNet-Voice: A Fine-Grained, Expert-Verified Benchmark for Speech Emotion Detection

LPO: Towards Accurate GUI Agent Interaction via Location Preference Optimization

Latent Multi-Head Attention for Small Language Models

Intra-Trajectory Consistency for Reward Modeling

Foundation Models in Medical Imaging -- A Review and Outlook

EdgeProfiler: A Fast Profiling Framework for Lightweight LLMs on Edge Using Analytical Model

Segment Concealed Objects with Incomplete Supervision

Inherently Faithful Attention Maps for Vision Transformers

DRAGged into Conflicts: Detecting and Addressing Conflicting Sources in Search-Augmented LLMs

Sparse Interpretable Deep Learning with LIES Networks for Symbolic Regression

Agentic Surgical AI: Surgeon Style Fingerprinting and Privacy Risk Quantification via Discrete Diffusion in a Vision-Language-Action Framework

Reparameterized LLM Training via Orthogonal Equivalence Transformation

Lightweight Sequential Transformers for Blood Glucose Level Prediction in Type-1 Diabetes

SurgBench: A Unified Large-Scale Benchmark for Surgical Video Analysis

Synthesize Privacy-Preserving High-Resolution Images via Private Textual Intermediaries

LeVo: High-Quality Song Generation with Multi-Preference Alignment

FAMSeg: Fetal Femur and Cranial Ultrasound Segmentation Using Feature-Aware Attention and Mamba Enhancement

SALT: A Lightweight Model Adaptation Method for Closed Split Computing Environments

MAGNET: A Multi-agent Framework for Finding Audio-Visual Needles by Reasoning over Multi-Video Haystacks

Towards Physics-informed Diffusion for Anomaly Detection in Trajectories

SAFE: Finding Sparse and Flat Minima to Improve Pruning

Tactile MNIST: Benchmarking Active Tactile Perception

SafeGenBench: A Benchmark Framework for Security Vulnerability Detection in LLM-Generated Code

Hierarchical Language Models for Semantic Navigation and Manipulation in an Aerial-Ground Robotic System

MELABenchv1: Benchmarking Large Language Models against Smaller Fine-Tuned Models for Low-Resource Maltese NLP

Deep Learning-Based Breast Cancer Detection in Mammography: A Multi-Center Validation Study in Thai Population

KVCache Cache in the Wild: Characterizing and Optimizing KVCache Cache at a Large Cloud Provider

Incentivizing Reasoning for Advanced Instruction-Following of Large Language Models

From Argumentative Text to Argument Knowledge Graph: A New Framework for Structured Argumentation

An Incremental Framework for Topological Dialogue Semantics: Efficient Reasoning in Discrete Spaces

Diffusion Graph Neural Networks for Robustness in Olfaction Sensors and Datasets

COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning

R-KV: Redundancy-aware KV Cache Compression for Reasoning Models

Quantum computing and artificial intelligence: status and perspectives

Bayesian Neural Scaling Law Extrapolation with Prior-Data Fitted Networks

On the performance of machine-learning-assisted Monte Carlo in sampling from simple statistical physics models

Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning

SOSBENCH: Benchmarking Safety Alignment on Scientific Knowledge

PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding

Transparency in Healthcare AI: Testing European Regulatory Provisions against Users' Transparency Needs

FoMoH: A clinically meaningful foundation model evaluation for structured electronic health records

Oversmoothing, Oversquashing, Heterophily, Long-Range, and more: Demystifying Common Beliefs in Graph Machine Learning

Learning to Rank Chain-of-Thought: An Energy-Based Approach with Outcome Supervision

Fixed Point Explainability

X-Sim: Cross-Embodiment Learning via Real-to-Sim-to-Real

JAEGER: Dual-Level Humanoid Whole-Body Controller

AgentVigil: Generic Black-Box Red-teaming for Indirect Prompt Injection against LLM Agents

AI Recommendations and Non-instrumental Image Concerns

Entropic Time Schedulers for Generative Diffusion Models

JEPA4Rec: Learning Effective Language Representations for Sequential Recommendation via Joint Embedding Predictive Architecture

Benchmarking Practices in LLM-driven Offensive Security: Testbeds, Metrics, and Experiment Design

Scholar Inbox: Personalized Paper Recommendations for Scientists

On Synthesizing Data for Context Attribution in Question Answering

Evaluating how LLM annotations represent diverse views on contentious topics

QualiSpeech: A Speech Quality Assessment Dataset with Natural Language Reasoning and Descriptions

ShED-HD: A Shannon Entropy Distribution Framework for Lightweight Hallucination Detection on Edge Devices

OS-Kairos: Adaptive Interaction for MLLM-Powered GUI Agents

MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion

Transformers without Normalization

Compute Optimal Scaling of Skills: Knowledge vs Reasoning

Identifying Trustworthiness Challenges in Deep Learning Models for Continental-Scale Water Quality Prediction

InfiniSST: Simultaneous Translation of Unbounded Speech with Large Language Model

From Euler to AI: Unifying Formulas for Mathematical Constants

C2-DPO: Constrained Controlled Direct Preference Optimization

Less is More: Improving LLM Alignment via Preference Data Selection

Quantifying Memorization and Parametric Response Rates in Retrieval-Augmented Vision-Language Models

ProMedTS: A Self-Supervised, Prompt-Guided Multimodal Approach for Integrating Medical Text and Time Series

Boosting Generalization in Diffusion-Based Neural Combinatorial Solver via Inference Time Adaptation

HARBOR: Exploring Persona Dynamics in Multi-Agent Competition

SelfCite: Self-Supervised Alignment for Context Attribution in Large Language Models

Truth Knows No Language: Evaluating Truthfulness Beyond English

PoGDiff: Product-of-Gaussians Diffusion Models for Imbalanced Text-to-Image Generation

Multi-Knowledge-oriented Nighttime Haze Imaging Enhancer for Vision-driven Intelligent Systems

Life-Code: Central Dogma Modeling with Multi-Omics Sequence Unification

Is attention all you need to solve the correlated electron problem?

Optimizing Temperature for Language Models with Multi-Sample Inference

Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactions

Adapt-Pruner: Adaptive Structural Pruning for Efficient Small Language Model Training

The Other Side of the Coin: Unveiling the Downsides of Model Aggregation in Federated Learning from a Layer-peeled Perspective

Satori: Reinforcement Learning with Chain-of-Action-Thought Enhances LLM Reasoning via Autoregressive Search

Activation-Informed Merging of Large Language Models

Layer by Layer: Uncovering Hidden Representations in Language Models

Search-Based Adversarial Estimates for Improving Sample Efficiency in Off-Policy Reinforcement Learning

Activation by Interval-wise Dropout: A Simple Way to Prevent Neural Networks from Plasticity Loss

Embracing Contradiction: Theoretical Inconsistency Will Not Impede the Road of Building Responsible AI Systems

Created by

Haebom

저자

Gordon Dai, Yunze Xiao

개요

본 논문은 책임있는 AI (RAI) 측정 지표 간의 이론적 불일치(상이한 공정성 정의나 정확성과 프라이버시 간의 트레이드오프 등)를 제거해야 할 결함이 아닌 귀중한 특징으로 받아들여야 한다고 주장하는 입장 논문입니다. 저자들은 측정 지표를 상이한 목표로 다룸으로써 이러한 불일치를 탐색하는 것이 세 가지 주요 이점을 제공한다고 주장합니다: (1) 규범적 다원주의: 상반될 수 있는 측정 지표의 전체 모음을 유지함으로써 RAI에 내재된 다양한 도덕적 입장과 이해 관계자의 가치를 적절히 반영합니다. (2) 인식론적 완전성: 때때로 상충되는 여러 측정 지표를 사용하면 다면적인 윤리적 개념을 더 포괄적으로 포착하여 단일의 단순화된 정의보다 이러한 개념에 대한 더 큰 정보 충실도를 유지할 수 있습니다. (3) 암묵적 규제: 이론적으로 상충되는 목표를 공동으로 최적화하면 특정 측정 지표에 대한 과적합을 방지하여 실제 세계의 복잡성 하에서 향상된 일반화 및 견고성을 갖춘 솔루션으로 모델을 유도합니다. 반대로, 측정 지표를 단순화하거나 제거하여 이론적 일관성을 강화하려는 노력은 이러한 가치 다양성을 좁히고, 개념적 깊이를 잃고, 모델 성능을 저하시킬 위험이 있습니다. 따라서 저자는 RAI 이론과 실제에서의 전환을 옹호합니다. 불일치에 갇히는 것이 아니라 허용 가능한 불일치 임계값을 특징짓고 실제로 견고하고 근사적인 일관성을 허용하는 메커니즘을 설명하는 것입니다.

시사점, 한계점

•

시사점:

◦

RAI 측정 지표 간의 불일치를 해결하는 대신, 이를 다양한 윤리적 관점과 이해관계자 가치를 반영하는 기회로 활용할 수 있음을 제시합니다.

◦

다양한 측정 지표를 통해 윤리적 개념에 대한 더욱 포괄적이고 정확한 이해를 얻을 수 있음을 시사합니다.

◦

상충되는 측정 지표들을 함께 최적화하여 모델의 일반화 및 견고성을 향상시킬 수 있음을 보여줍니다.

◦

RAI 연구 및 개발에서 단순화된 접근 방식을 넘어 다원적이고 복잡한 접근 방식을 채택해야 함을 주장합니다.

•

한계점:

◦

"허용 가능한 불일치 임계값"의 구체적인 정의와 측정 방법에 대한 논의가 부족합니다.

◦

실제 적용 시 발생할 수 있는 다양한 기술적 및 실무적 어려움에 대한 구체적인 해결 방안 제시가 부족합니다.

◦

제안된 접근 방식의 효과성을 검증하기 위한 실험적 또는 경험적 증거가 제시되지 않았습니다.

◦

상충되는 측정 지표들을 효율적으로 통합하고 최적화하는 구체적인 방법론에 대한 설명이 부족합니다.

Made with Slashpage