4157 papers
This study empirically validates an integrated energy-aware software framework by demonstrating that code-level interventions, CI/CD energy measurements, and regression detection can function together as a cohesive feedback loop to identify and localize energy inefficiencies in real-world Spring Boot applications.
This paper introduces YQX+, a unified benchmarking framework for probabilistic modeling of expressive piano performance that standardizes prediction tasks and evaluation protocols to compare diverse models, alongside PerformanceMOS, a data-driven estimator for perceived expressive quality, demonstrating that advanced models like XGBoost and -VAE outperform prior baselines in capturing human-like musical expression.
This paper demonstrates that the purported benefits of experience replay in multi-agent LLM pipelines are largely artifacts of single-run evaluations, as rigorous replication across multiple runs reveals the effects to be unstable, non-reproducible, and governed by retrieval resolution rather than store size.
This comparative evaluation of five large language models against the 2025 ACC/AHA/ACEP/NAEMSP/SCAI Acute Coronary Syndromes guidelines reveals that DeepSeek significantly outperformed other models in quality and guideline adherence, while overall model reliability remains variable and necessitates further research into their real-world clinical impact.
This paper introduces AgentSentinel, an autonomous multi-agent framework that leverages large language models and a specialized feedback mechanism to detect anomalies, identify root causes, and execute self-healing repairs in distributed systems, demonstrating high-speed detection and remediation with zero false positives in a controlled feasibility study.
DASMark is a robust diffusion image watermarking framework that employs dual-anchor geometric synchronization and reproducible user-specific noise to achieve highly accurate deepfake detection and scalable user identification even under severe geometric attacks like rotation and cropping.
This paper presents a structured integrative review and the Runtime Security Monitoring and Continuous Assurance (RSMCA) framework, which addresses the dynamic security risks of long-lived AI agents by defining eight monitoring surfaces, multi-layered evaluation methods, and graduated response mechanisms to ensure observable, enforceable, and auditable agent operations in network and systems management environments.
This paper proposes a blockchain-based federated learning framework that eliminates central server vulnerabilities and ensures data privacy through decentralized consensus and hybrid model architectures, thereby enabling secure, collaborative surgical video analysis to advance surgical embodied intelligence.
This study proposes an explainable and uncertainty-aware hybrid deep learning framework that fuses engineered MRI features with multiple neural learners to achieve high accuracy in prostate MRI image classification, while explicitly noting that its current performance is methodological and requires clinical validation with pathology-confirmed labels before being used for cancer detection.
This study proposes an intelligence-driven hierarchical adaptive large-neighborhood search (H-ALNS) algorithm that optimizes three-dimensional route loading for homogeneous vehicles by prioritizing fleet size minimization over distance, successfully reducing total vehicle dispatches by 15.31% across 600 benchmark instances to achieve resource-efficient freight operations.