Search

Modern Artificial Intelligence (AI) technologies, led by Machine Learning (ML), have gained unprecedented momentum over the past decade. Following this wave of ``AI summer", the network research community has also embraced AI/ML algorithms to address many problems related to network operations and management. However, compared to their counterparts in other domains, most ML-based solutions have yet to receive large-scale deployment due to insufficient maturity for production settings. This paper concentrates on the practical issues of developing and operating ML-based solutions in real networks. Specifically, we enumerate the key factors hindering the integration of AI/ML in real networks and review existing solutions to uncover the missing considerations. We also highlight two potential directions, i.e., MLOps and Causal ML, that can close the gap. We believe this paper spotlights the system-related considerations on implementing \& maintaining ML-based solutions and invigorate their full adoption in future networks., This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

International audience; Small IoT devices, such as drones and lightweight battery-powered robots, are emerging as a major platform for the deployment of AI/ML capabilities. Autonomous and semiautonomous device operation relies on the systematic use of deep neural network models for solving complex tasks, such as image classification. The challenging restrictions of these devices in terms of computing capabilities, network connectivity, and power consumption are the main limits to the accuracy of latencysensitive inferences. This paper presents ReBEL, a split computing architecture enabling the dynamic remote offload of partial computations or, in alternative, a local approximate labeling based on a jointly-trained classifier. Our approach combines elements of head network distillation, early exit classification, and bottleneck injection with the goal of reducing the average endto-end latency of AI/ML inference on constrained IoT devices.

ObjectivesThis study aims to evaluate the diagnostic performance of machine-learning-based contrast-enhanced CT radiomic analysis for categorizing benign and malignant ovarian tumors.MethodsA total of 1,329 patients with ovarian tumors were randomly divided into a training cohort (N=930) and a validation cohort (N=399). All tumors were resected, and pathological findings were confirmed. Radiomic features were extracted from the portal venous phase images of contrast-enhanced CT. The clinical predictors included age, CA-125, HE-4, ascites, and margin of tumor. Both radiomics model (including selected radiomic features) and mixed model (incorporating selected radiomic features and clinical predictors) were constructed respectively. Six classifiers [k-nearest neighbor (KNN), support vector machines (SVM), random forest (RF), logistic regression (LR), multi-layer perceptron (MLP), and eXtreme Gradient Boosting (XGBoost)] were used for each model. The mean relative standard deviation (RSD) and area under the receiver operating characteristic curve (AUC) were applied to evaluate and select the best classifiers. Then, the performances of the two models with selected classifiers were assessed in the validation cohort.ResultsThe MLP classifier with the least RSD (1.21 and 0.53, respectively) was selected as the best classifier in both radiomics and mixed models. The two models with MLP classifier performed well in the validation cohort, with the AUCs of 0.91 and 0.96 and with accuracies (ACCs) of 0.83 and 0.87, respectively. The Delong test showed that the AUC of mixed model was statistically different from that of radiomics model (pConclusionsMachine-learning-based CT radiomic analysis could categorize ovarian tumors with good performance preoperatively. The mixed model with MLP classifier may be a potential tool in clinical applications.