This paper proposes the Server–Client Machine Learning Intrusion Detection System (SC-MLIDS), a novel fusion framework designed to enhance security in Wireless Sensor Networks (WSNs), which are inherently vulnerable to various security threats due to their distributed nature and resource constraints . Traditional Intrusion Detection Systems (IDSs) often face challenges with high computational demands and privacy issues. SC-MLIDS addresses these problems by integrating Federated Learning (FL) with a multi-sensor fusion approach to implementing two layers of defence that operate independently of specific attack types. Moreover, this framework leverages a server–client architecture to efficiently manage and process data from sensor nodes , sink nodes, and gateways within the network. The core innovation of SC-MLIDS lies in its dual model aggregation algorithms at the gateway: one assesses model performance and weight, while the other uses majority voting to integrate predictions from both client and server models. As a result, this approach reduces redundant data transmissions and enhances detection accuracy, making it more effective than conventional methods in WSNs. Our proposed framework outperforms current state-of-the-art techniques, achieving F1-scores of 99.78% and 98.80% for the two aggregation algorithms, namely, Weighted Score and Majority Voting. This validation demonstrates the effectiveness of SC-MLIDS in providing accurate intrusion detection and robust data management.

Federated Learning (FL) has emerged as a novel distributed Machine Learning (ML) approach, to tackle the challenges associated with data privacy and overload in MLbased intrusion detection systems (IDSs). Drawing inspiration from the FL architecture, we have introduced a hybrid ML IDS tailored for Wireless Sensor Networks (WSNs). This system is crafted to leverage ML for achieving a two-layer intrusion detection mechanism in WSNs free from constraints posed by specific attack types. The architecture follows a server-client model compatible with the configuration of sensor nodes, sink nodes, and gateways in WSNs. In this setup, client models located at sink nodes undergo training using sensing data while the server model at the gateway is trained using network traffic data. This two-layer training approach amplifies the efficiency of intrusion detection and ensures comprehensive network coverage. The results derived from our simulation experiments corroborate the effectiveness of the proposed hybrid ML IDS. It generates precise aggregation predictions and leads to a substantial reduction in redundant data transmissions. Furthermore, the system exhibits efficacy in detecting intrusions through a dual validation process.

We propose an approach to estimate the number of samples required for a model to reach a target performance. We find that the power law, the de facto principle to estimate model performance, leads to a large error when using a small dataset (e.g., 5 samples per class) for extrapolation. This is because the log-performance error against the log-dataset size follows a nonlinear progression in the few-shot regime followed by a linear progression in the high-shot regime. We introduce a novel piecewise power law (PPL) that handles the two data regimes differently. To estimate the parameters of the PPL, we introduce a random forest regressor trained via meta learning that generalizes across classification/detection tasks, ResNet/ViT based architectures, and random/pre-trained initializations. The PPL improves the performance estimation on average by 37% across 16 classification and 33% across 10 detection datasets, compared to the power law. We further extend the PPL to provide a confidence bound and use it to limit the prediction horizon that reduces over-estimation of data by 76% on classification and 91% on detection datasets.

TUNERCAR is a toolchain that jointly optimizes racing strategy, planning methods, control algorithms, and vehicle parameters for an autonomous racecar. In this paper, we detail the target hardware, software, simulators, and systems infrastructure for this toolchain. Our methodology employs a parallel implementation of CMA-ES which enables simulations to proceed 6 times faster than real-world rollouts. We show our approach can reduce the lap times in autonomous racing, given a fixed computational budget. For all tested tracks, our method provides the lowest lap time, and relative improvements in lap time between 7-21%. We demonstrate improvements over a naive random search method with equivalent computational budget of over 15 seconds/lap, and improvements over expert solutions of over 2 seconds/lap. We further compare the performance of our method against hand-tuned solutions submitted by over 30 international teams, comprised of graduate students working in the field of autonomous vehicles. Finally, we discuss the effectiveness of utilizing an online planning mechanism to reduce the reality gap between our simulation and actual tests.

Model Predictive Control (MPC) plays an important role in optimizing operations of complex cyber-physical systems because of its ability to forecast system’s behavior and act under system level constraints. However, MPC requires reasonably accurate underlying models of the system. In many applications, such as building control for energy management, Demand Response, or peak power reduction, obtaining a high-fidelity physics-based model is cost and time prohibitive, thus limiting the widespread adoption of MPC. To this end, we propose a data-driven control algorithm for MPC that relies only on the historical data. We use multi-output regression trees to represent the system’s dynamics over multiple future time steps and formulate a finite receding horizon control problem that can be solved in real-time in closed-loop with the physical plant. We apply this algorithm to peak power reduction in buildings to optimally trade-off peak power reduction against thermal comfort without having to learn white/grey box models of the systems dynamics.