2833 papers
This paper proposes a parameter-efficient, speaker-aware hybrid framework that combines hierarchical extractive condensation, LoRA-based adaptation, and reinforcement learning with a novel speaker-attribution reward to significantly improve faithfulness and reduce hallucinations in dialogue summarization while maintaining competitive lexical quality.
This study analyzes behavioral signals from over 1,000,000 interactions on major short-video platforms to demonstrate that dynamic user engagement patterns, particularly rationality and curiosity indicators, are more effective than demographics in predicting membership in echo chambers surrounding misleading AI-generated videos, thereby supporting privacy-conscious early-warning systems.
This study introduces Progressive Residual LightGBM (PRLGB), a three-stage ensemble model that significantly outperforms standard tree-based algorithms in predicting 5G Quality-of-Service resource allocation by effectively learning residual errors and fusing uncertainty-aware features to achieve superior accuracy metrics.
This study introduces an English–Arabic benchmark corpus derived from U.S. presidential speeches and a corresponding evaluation framework designed to systematically assess the ability of AI translation systems to preserve implicit political meaning, particularly strategies of indirectness and name avoidance, which are often overlooked in favor of lexical accuracy.
IZSafe is an embedded deep-learning system that utilizes inertial sensors on firefighter oxygen-bottle carriers to achieve real-time, micro-level situational awareness by accurately recognizing individual movements with 93.85% accuracy on an ARM Cortex-M4 microcontroller.
This study proposes a condition-aware multi-scale Long- and Short-term Time-series Network (CA-MS-LSTNet), optimized by an improved Bat Algorithm, to significantly enhance the ultra-short-term prediction accuracy of wind turbine parameters by effectively modeling nonlinear, heterogeneous, and operating-condition-dependent temporal characteristics in SCADA data.
This study proposes and validates a compact convolutional neural network framework that converts software-defined radio captures of drone video transmissions into time-domain images for real-time, high-accuracy detection with significantly reduced computational costs compared to traditional spectrogram-based methods.
This paper presents a compact, CPU-optimized ALPR system tailored for challenging Vietnamese road conditions, integrating a YOLOv8n detector, ByteTrack, rapid deskewing, and ONNX-accelerated OCR with spatiotemporal voting to achieve high accuracy and real-time performance despite adverse weather, motion blur, and diverse plate formats.
This study audits 173,369 AI datasets to reveal that while their creation costs follow a non-monotonic trend, their scholarly impact is largely independent of financial or labor investment, being instead driven by author count and topical prevalence.
This paper introduces Stable Group-Relative Policy Optimization (S-GRPO), a novel method that mitigates the "Think-Answer Mismatch" in large reasoning models by deriving noise-aware advantage weights, thereby achieving superior robustness and performance over standard GRPO under noisy reward conditions.