Low-Altitude Fluid Antenna Network with Multi-Agent Reinforcement Learning
本文通过多智能体强化学习优化低空流体天线网络的动态重构，以解决动态信道、遮挡和干扰问题，提高系统总速率。
0 citationsRead paper
Selected work
Representative Papers
Safety-Gated Autoscaling: A Multi-Layered Defense Architecture for Kubernetes Vertical Resource Optimization
This work addresses significant cost inefficiencies in Kubernetes arising from resource over-provisioning and the limitations of existing autoscaling mechanisms, which often exhibit delayed responses and may obscure critical issues such as memory leaks. The authors propose a container vertical resource optimization approach grounded in a five-layer safety pipeline that introduces gating mechanisms prior to scaling decisions, with memory leak detection serving as a pivotal blocking criterion. The framework integrates SLA monitoring, circuit breakers, policy engines, and human-in-the-loop approval to ensure system reliability. Memory leaks are identified through a combination of linear regression (using R² scores) and percentile-based analysis, while scaling recommendations are generated via Holt-Winters forecasting and multi-objective Pareto optimization. The solution also incorporates conflict detection between Horizontal Pod Autoscaler (HPA) and Pod Disruption Budget (PDB) policies. Evaluated on Google Kubernetes Engine, the method achieves 20–40% cost savings, demonstrates 83% accuracy in memory leak detection, and validates reliability through 1,118 test cases covering 80.3% of the system.
0 citationsRead paper
A multiverse-consensus pipeline for reproducible feature selection in untargeted LC-MS metabolomics
This study addresses the high sensitivity of feature selection in untargeted LC-MS metabolomics to preprocessing pipelines, which renders results vulnerable to analytical degrees of freedom. To mitigate this, we introduce multi-universe analysis—a novel, auditable, configuration-driven consensus framework. Building upon a ten-stage quality control filter, our approach integrates four distinct preprocessing strategies with four feature-ranking methods, coupled with bootstrap-based stability selection and label permutation testing to retain only features consistently identified across multiple analytical paths. Among 30,370 initial features, individual pipelines selected 4–20 features with low Jaccard consistency (as low as 0.05), whereas the multi-universe consensus yielded 15 robust features reproducible in at least two out of four pathways. Notably, one feature was stable across all pathways and showed no false positives in 50 permutation tests, substantially enhancing result reliability and reproducibility.
0 citationsRead paper
Optimization Algorithms for Joint OFDM Waveform Design and RIS Configuration in 6G Networks: From Convex Relaxation to Foundation Models
This study addresses the multi-objective mixed-integer nonlinear programming problem of jointly designing OFDM waveforms and configuring reconfigurable intelligent surfaces (RIS) for 6G. Synthesizing insights from 78 studies published between 2021 and 2026, it proposes the first cross-paradigm classification framework encompassing model-based convex relaxation, heuristic search, deep reinforcement and unsupervised learning, as well as emerging approaches integrating foundation models, diffusion-based generative AI, and quantum optimization. The work identifies a key property of neural network inference—maintaining constant latency under antenna array scaling (N = 16–128)—and establishes a standardized benchmark. Results demonstrate that machine learning methods achieve 95–99% of the spectral efficiency of model-driven approaches while accelerating inference by 10²–10⁴ times, and highlight six open challenges, including the lack of unified benchmarks, hardware-aware deployment constraints, and safety concerns regarding large models in real-time control.
0 citationsRead paper
Patient-Level Elbow Abnormality Detection: Leakage-Aware Evaluation of Learned Preprocessing, Calibration, and Triage-Oriented Operating Points
This study addresses evaluation bias arising from data leakage in musculoskeletal abnormality detection on elbow X-rays by introducing a leakage-aware, patient-level evaluation protocol. Using the MURA dataset with DenseNet121 as the backbone, the work systematically assesses the impact of both learning-based (e.g., DnCNN) and conventional (e.g., CLAHE) preprocessing strategies on model discriminative performance, calibration, and clinical utility. Comprehensive evaluation at high-specificity operating points employs multiple metrics, including AUROC, PR-AUC, Expected Calibration Error (ECE), and Brier score. Results indicate that preprocessing yields only marginal improvements highly dependent on specific configurations, and the baseline model using raw inputs remains competitive across most metrics, with no single preprocessing method consistently outperforming the baseline.
0 citationsRead paper
Recent publications
Latest Papers
Low-Altitude Fluid Antenna Network with Multi-Agent Reinforcement Learning
本文通过多智能体强化学习优化低空流体天线网络的动态重构，以解决动态信道、遮挡和干扰问题，提高系统总速率。
0 citationsRead paper
Safety-Gated Autoscaling: A Multi-Layered Defense Architecture for Kubernetes Vertical Resource Optimization
This work addresses significant cost inefficiencies in Kubernetes arising from resource over-provisioning and the limitations of existing autoscaling mechanisms, which often exhibit delayed responses and may obscure critical issues such as memory leaks. The authors propose a container vertical resource optimization approach grounded in a five-layer safety pipeline that introduces gating mechanisms prior to scaling decisions, with memory leak detection serving as a pivotal blocking criterion. The framework integrates SLA monitoring, circuit breakers, policy engines, and human-in-the-loop approval to ensure system reliability. Memory leaks are identified through a combination of linear regression (using R² scores) and percentile-based analysis, while scaling recommendations are generated via Holt-Winters forecasting and multi-objective Pareto optimization. The solution also incorporates conflict detection between Horizontal Pod Autoscaler (HPA) and Pod Disruption Budget (PDB) policies. Evaluated on Google Kubernetes Engine, the method achieves 20–40% cost savings, demonstrates 83% accuracy in memory leak detection, and validates reliability through 1,118 test cases covering 80.3% of the system.
0 citationsRead paper
A multiverse-consensus pipeline for reproducible feature selection in untargeted LC-MS metabolomics
This study addresses the high sensitivity of feature selection in untargeted LC-MS metabolomics to preprocessing pipelines, which renders results vulnerable to analytical degrees of freedom. To mitigate this, we introduce multi-universe analysis—a novel, auditable, configuration-driven consensus framework. Building upon a ten-stage quality control filter, our approach integrates four distinct preprocessing strategies with four feature-ranking methods, coupled with bootstrap-based stability selection and label permutation testing to retain only features consistently identified across multiple analytical paths. Among 30,370 initial features, individual pipelines selected 4–20 features with low Jaccard consistency (as low as 0.05), whereas the multi-universe consensus yielded 15 robust features reproducible in at least two out of four pathways. Notably, one feature was stable across all pathways and showed no false positives in 50 permutation tests, substantially enhancing result reliability and reproducibility.
0 citationsRead paper
Optimization Algorithms for Joint OFDM Waveform Design and RIS Configuration in 6G Networks: From Convex Relaxation to Foundation Models
This study addresses the multi-objective mixed-integer nonlinear programming problem of jointly designing OFDM waveforms and configuring reconfigurable intelligent surfaces (RIS) for 6G. Synthesizing insights from 78 studies published between 2021 and 2026, it proposes the first cross-paradigm classification framework encompassing model-based convex relaxation, heuristic search, deep reinforcement and unsupervised learning, as well as emerging approaches integrating foundation models, diffusion-based generative AI, and quantum optimization. The work identifies a key property of neural network inference—maintaining constant latency under antenna array scaling (N = 16–128)—and establishes a standardized benchmark. Results demonstrate that machine learning methods achieve 95–99% of the spectral efficiency of model-driven approaches while accelerating inference by 10²–10⁴ times, and highlight six open challenges, including the lack of unified benchmarks, hardware-aware deployment constraints, and safety concerns regarding large models in real-time control.
0 citationsRead paper
Patient-Level Elbow Abnormality Detection: Leakage-Aware Evaluation of Learned Preprocessing, Calibration, and Triage-Oriented Operating Points
This study addresses evaluation bias arising from data leakage in musculoskeletal abnormality detection on elbow X-rays by introducing a leakage-aware, patient-level evaluation protocol. Using the MURA dataset with DenseNet121 as the backbone, the work systematically assesses the impact of both learning-based (e.g., DnCNN) and conventional (e.g., CLAHE) preprocessing strategies on model discriminative performance, calibration, and clinical utility. Comprehensive evaluation at high-specificity operating points employs multiple metrics, including AUROC, PR-AUC, Expected Calibration Error (ECE), and Brier score. Results indicate that preprocessing yields only marginal improvements highly dependent on specific configurations, and the baseline model using raw inputs remains competitive across most metrics, with no single preprocessing method consistently outperforming the baseline.
0 citationsRead paper