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.

Optimizing LLMs for Resource-Constrained Environments: A Survey of Model Compression Techniques

Humans can learn to detect AI-generated texts, or at least learn when they can't

Large Language Models Understanding: an Inherent Ambiguity Barrier

Data Therapist: Eliciting Domain Knowledge from Subject Matter Experts Using Large Language Models

SeriesBench: A Benchmark for Narrative-Driven Drama Series Understanding

Nexus-Gen: A Unified Model for Image Understanding, Generation, and Editing

PINN-MEP: Continuous Neural Representations for Minimum-Energy Path Discovery in Molecular Systems

Building Trustworthy Multimodal AI: A Review of Fairness, Transparency, and Ethics in Vision-Language Tasks

GeoUni: A Unified Model for Generating Geometry Diagrams, Problems and Problem Solutions

A highly maneuverable flying squirrel drone with agility-improving foldable wings

VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning

FindAnything: Open-Vocabulary and Object-Centric Mapping for Robot Exploration in Any Environment

Perils of Label Indeterminacy: A Case Study on Prediction of Neurological Recovery After Cardiac Arrest

CodeIF-Bench: Evaluating Instruction-Following Capabilities of Large Language Models in Interactive Code Generation

Novel Deep Neural OFDM Receiver Architectures for LLR Estimation

NaFM: Pre-training a Foundation Model for Small-Molecule Natural Products

Benchmarking Open-Source Large Language Models on Healthcare Text Classification Tasks

Atyaephyra at SemEval-2025 Task 4: Low-Rank Negative Preference Optimization

Integrating AI for Human-Centric Breast Cancer Diagnostics: A Multi-Scale and Multi-View Swin Transformer Framework

Negotiative Alignment: Embracing Disagreement to Achieve Fairer Outcomes -- Insights from Urban Studies

Large Language Models for Outpatient Referral: Problem Definition, Benchmarking and Challenges

Semantic Shift Estimation via Dual-Projection and Classifier Reconstruction for Exemplar-Free Class-Incremental Learning

LIVS: A Pluralistic Alignment Dataset for Inclusive Public Spaces

Faster, Cheaper, Better: Multi-Objective Hyperparameter Optimization for LLM and RAG Systems

FLARE: A Framework for Stellar Flare Forecasting using Stellar Physical Properties and Historical Records

TLOB: A Novel Transformer Model with Dual Attention for Price Trend Prediction with Limit Order Book Data

Correcting Noisy Multilabel Predictions: Modeling Label Noise through Latent Space Shifts

Safety Evaluation of DeepSeek Models in Chinese Contexts

DejAIvu: Identifying and Explaining AI Art on the Web in Real-Time with Saliency Maps

Texture Image Synthesis Using Spatial GAN Based on Vision Transformers

Toward Task Generalization via Memory Augmentation in Meta-Reinforcement Learning

The Right to AI

Communicating Activations Between Language Model Agents

Guaranteed Recovery of Unambiguous Clusters

ValuesRAG: Enhancing Cultural Alignment Through Retrieval-Augmented Contextual Learning

Vision Transformers for Efficient Indoor Pathloss Radio Map Prediction

Quantifying Risk Propensities of Large Language Models: Ethical Focus and Bias Detection through Role-Play

E2E-AFG: An End-to-End Model with Adaptive Filtering for Retrieval-Augmented Generation

Jailbreaking and Mitigation of Vulnerabilities in Large Language Models

WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines

CATCH: Channel-Aware multivariate Time Series Anomaly Detection via Frequency Patching

Learning to Compare Hardware Designs for High-Level Synthesis

To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning

Automated detection of underdiagnosed medical conditions via opportunistic imaging

Exploring the Trade-Offs: Quantization Methods, Task Difficulty, and Model Size in Large Language Models From Edge to Giant

On Synthetic Texture Datasets: Challenges, Creation, and Curation

Lexicon3D: Probing Visual Foundation Models for Complex 3D Scene Understanding

XG-NID: Dual-Modality Network Intrusion Detection using a Heterogeneous Graph Neural Network and Large Language Model

Recite, Reconstruct, Recollect: Memorization in LMs as a Multifaceted Phenomenon

Enhancing Differential Testing With LLMs For Testing Deep Learning Libraries

HORAE: A Domain-Agnostic Language for Automated Service Regulation

DEGAP: Dual Event-Guided Adaptive Prefixes for Templated-Based Event Argument Extraction with Slot Querying

Analyzing Consumer IoT Traffic from Security and Privacy Perspectives: a Comprehensive Survey

DyCE: Dynamically Configurable Exiting for Deep Learning Compression and Real-time Scaling

The Inadequacy of Similarity-based Privacy Metrics: Privacy Attacks against "Truly Anonymous" Synthetic Datasets

Connecting NTK and NNGP: A Unified Theoretical Framework for Wide Neural Network Learning Dynamics

An automated end-to-end deep learning-based framework for lung cancer diagnosis by detecting and classifying the lung nodules

Label-Efficient Deep Learning in Medical Image Analysis: Challenges and Future Directions

Transformer-based assignment decision network for multiple object tracking

An alignment safety case sketch based on debate

The Power of Stories: Narrative Priming Shapes How LLM Agents Collaborate and Compete

A Survey of Slow Thinking-based Reasoning LLMs using Reinforced Learning and Inference-time Scaling Law

Agentic Neurodivergence as a Contingent Solution to the AI Alignment Problem

Theoretical Foundations for Semantic Cognition in Artificial Intelligence

Approximate Lifted Model Construction

MultiMind: Enhancing Werewolf Agents with Multimodal Reasoning and Theory of Mind

Advancing Embodied Agent Security: From Safety Benchmarks to Input Moderation

Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute

Recursive Inference Scaling: A Winning Path to Scalable Inference in Language and Multimodal Systems

Generating Symbolic World Models via Test-time Scaling of Large Language Models

Imagining and building wise machines: The centrality of AI metacognition

Public Perceptions of Fairness Metrics Across Borders

Combating Confirmation Bias: A Unified Pseudo-Labeling Framework for Entity Alignment

Flow-GRPO: Training Flow Matching Models via Online RL

StreamBridge: Turning Your Offline Video Large Language Model into a Proactive Streaming Assistant

ComPO: Preference Alignment via Comparison Oracles

TransProQA: an LLM-based literary Translation evaluation metric with Professional Question Answering

TokLIP: Marry Visual Tokens to CLIP for Multimodal Comprehension and Generation

Reasoning Models Don't Always Say What They Think

Crosslingual Reasoning through Test-Time Scaling

CART-ELC: Oblique Decision Tree Induction via Exhaustive Search

Threshold Modulation for Online Test-Time Adaptation of Spiking Neural Networks

Time of the Flight of the Gaussians: Optimizing Depth Indirectly in Dynamic Radiance Fields

High-fidelity Grain Growth Modeling: Leveraging Deep Learning for Fast Computations

Feature-Augmented Deep Networks for Multiscale Building Segmentation in High-Resolution UAV and Satellite Imagery

Mapping User Trust in Vision Language Models: Research Landscape, Challenges, and Prospects

Scalable Chain of Thoughts via Elastic Reasoning

Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection

PlaceIt3D: Language-Guided Object Placement in Real 3D Scenes

Software Development Life Cycle Perspective: A Survey of Benchmarks for CodeLLMs and Agents

T-T: Table Transformer for Tagging-based Aspect Sentiment Triplet Extraction

Enhancing Cooperative Multi-Agent Reinforcement Learning with State Modelling and Adversarial Exploration

Put CASH on Bandits: A Max K-Armed Problem for Automated Machine Learning

Incentive-Aware Machine Learning; Robustness, Fairness, Improvement & Causality

LAPSO: A Unified Optimization View for Learning-Augmented Power System Operations

Concept-Based Unsupervised Domain Adaptation

Revealing Weaknesses in Text Watermarking Through Self-Information Rewrite Attacks

Biomed-DPT: Dual Modality Prompt Tuning for Biomedical Vision-Language Models

Stochastic Variational Propagation: Local, Scalable and Efficient Alternative to Backpropagation

Dukawalla: Voice Interfaces for Small Businesses in Africa

Inference-Time Policy Steering through Human Interactions

Created by

Haebom

저자

Yanwei Wang, Lirui Wang, Yilun Du, Balakumar Sundaralingam, Xuning Yang, Yu-Wei Chao, Claudia Perez-D'Arpino, Dieter Fox, Julie Shah

개요

인간의 시범 학습을 통해 훈련된 생성 정책은 다중 모드의 장기간 작업을 자율적으로 수행할 수 있습니다. 하지만 추론 과정에서 인간이 정책 실행 루프에서 제외되는 경우가 많아, 사전 훈련된 정책을 여러 예측 중 특정 하위 목표 또는 궤적 형태로 안내하는 능력이 제한됩니다. 단순한 인간 개입은 분포 이동을 악화시켜 제약 위반이나 실행 실패로 이어질 수 있습니다. 본 논문에서는 상호 작용 데이터에 대한 정책 미세 조정 대신 인간 상호 작용을 활용하여 생성 샘플링 과정에 편향을 주는 추론 시간 정책 조향(ITPS) 프레임워크를 제안하여 인간 의도와 정책 출력을 더 잘 정렬하고 분포 외 오류를 유발하지 않습니다. 세 가지 시뮬레이션 및 실제 벤치마크에서 세 가지 형태의 인간 상호 작용 및 관련 정렬 거리 측정 기준을 테스트하여 ITPS를 평가합니다. 여섯 가지 샘플링 전략 중 확산 정책을 사용한 제안된 확률적 샘플링이 정렬과 분포 이동 사이에서 최상의 절충안을 달성합니다. 비디오는 https://yanweiw.github.io/itps/ 에서 확인할 수 있습니다.

Inference-Time Policy Steering

Inference-Time Policy Steering with Human Interactions

yanweiw.github.io

시사점, 한계점

•

시사점:

◦

인간 상호 작용을 통해 생성 정책의 출력을 효과적으로 조향하는 새로운 프레임워크(ITPS) 제시.

◦

미세 조정 없이 인간의 의도를 반영하여 정책 출력을 개선하는 방법 제시.

◦

확산 정책을 이용한 확률적 샘플링이 정렬과 분포 이동 사이 최적의 균형을 제공함을 실험적으로 증명.

◦

다양한 실제 및 시뮬레이션 환경에서 성능 검증.

•

한계점:

◦

제안된 방법이 모든 유형의 작업 및 상호 작용 방식에 대해 최적으로 작동하는지는 추가 연구 필요.

◦

실제 환경에서의 범용성 및 확장성에 대한 추가적인 검증 필요.

◦

특정 유형의 인간 상호 작용에 대해서만 평가되었으므로, 다른 유형의 상호 작용에 대한 일반화 가능성에 대한 추가 연구 필요.

Made with Slashpage