With the increasing connectivity between the Electronic Control Units (ECUs) and the outside world, safety and security have become stringent problems. The Controller Area Network (CAN) bus is the most commonly used in-vehicle network protocol, which lacks security mechanisms by design, so it is vulnerable to various attacks. In this paper, we propose a novel intrusion detection model called CNN-LSTM with Attention model (CLAM) for the in-vehicle network, especially CAN. The CLAM model uses one-dimensional convolution (Conv1D) to extract the abstract features of the signal values at each time step and feeds it into the Bidirectional Long Short Term Memory (Bi-LSTM) to extract the time dependence. By combining the attention mechanisms, we calculate the weight of each hidden state output by Bi-LSTM and perform a weighted summation, so that the model focuses only on locally important time steps, which can improve the convergence speed of the model and prediction accuracy. The proposed model uses the bit flip rate to extract continuous signal boundaries in the 64-bit CAN data, so it does not need to parse the CAN communication matrix and is suitable for different vehicles. The extensive evaluation results demonstrate that our proposed CLAM model can effectively detect CAN attacks, with an average F1-score of 0.951 and an error rate of 2.16%. Compared with the work in related research fields, the accuracy of attack detection is improved by 2.5%.

Although many network traffic protection methods have been developed to protect user privacy, encrypted traffic can still reveal sensitive user information with sophisticated analysis. In this paper, we propose ETC-PS , a novel encrypted traffic classification method with path signature. We first construct the traffic path with a session packet length sequence to represent the interactions between the client and the server. Then, path transformations are conducted to exhibit its structure and obtain different information. A multiscale path signature is finally computed as a kind of distinctive feature to train the traditional machine learning classifier, which achieves highly robust accuracy and low training overhead. Six publicly available datasets with different traffic types of HTTPS/1, HTTPS/2, QUIC, VPN, non-VPN, Tor, and non-Tor are used to conduct closed-world and open-world evaluations to verify the effectiveness of ETC-PS . The experimental results demonstrate that ETC-PS is superior to the state-of-the-art methods in terms of accuracy, $f_{1}$ score, time complexity, and stability.

The rapid growth of interconnected smart devices and advanced computing technologies in the industrial Internet of Things (IIoT) has significantly enhanced operational resilience and performance but also increased cybersecurity risks. While deep learning shows promise in IIoT security, it faces challenges due to the lack of labeled data and reliance on human expertise for unsupervised anomaly detection. To address these challenges, a novel automated adversarial deep learning-based unsupervised anomaly detection method called EvoAAE is proposed to optimize the hyperparameters and neural architectures of adversarial variational autoencoder (VAE) for securing IIoT. Specifically, a generative adversarial network-based VAE is employed to adversarially generate multivariate time series. Then, particle swarm optimization with an efficient binary encoding strategy is designed to evolve hyperparameters and neural architectures in adversarial VAE including batch size, learning rate, the type of optimizer, the number of convolutional layer, the number of kernels of convolutional layer, kernel size, the type of normalization layer, and the type of active function. The experimental results indicate that EvoAAE achieves notable performance across four IIoT datasets in industrial control domain, i.e., secure water treatment, water distribution, Mars Science Laboratory, and power system domain, i.e., power system attack with precision of 0.949, 0.8356, 0.972, and 0.981, recall of 0.971, 0.9214, 0.964, and 0.979, and $F_{1}$-score of 0.960, 0.8764, 0.968, and 0.980, respectively.

Conjugate gradient (CG) methods are a class of important methods for solving linear equations and nonlinear optimization problems. In this paper, we propose a new stochastic CG algorithm with variance reduction and we prove its linear convergence with the Fletcher and Reeves method for strongly convex and smooth functions. We experimentally demonstrate that the CG with variance reduction algorithm converges faster than its counterparts for four learning models, which may be convex, nonconvex or nonsmooth. In addition, its area under the curve performance on six large-scale data sets is comparable to that of the LIBLINEAR solver for the L2-regularized L2-loss but with a significant improvement in computational efficiency.

Dynamic neural networks are considered as an effective method in the field of scientific computing, among which gradient neural networks (GNNs) are an efficient method for solving static problems. However, when solving dynamic problems, the current GNNs are often subject to lagging errors. In this article, we develop a finite-time convergent GNN (FTCGNN) model for solving static and time-varying systems of linear equations. Different from zeroing neural networks (ZNNs) dedicated to time-varying problem solving, the FTCGNN model has finite-time convergence regardless of the existence of time-varying noises. Simulation results show that the FTCGNN model is effective, among which the comparisons with existing GNNs and ZNNs validate the advantages of the FTCGNN model.

The k-winners-take-all (k-WTA) problem refers to the selection of k winners with the first k largest inputs over a group of n neurons, where each neuron has an input. In existing k-WTA neural network models, the positive integer k is explicitly given in the corresponding mathematical models. In this article, we consider another case where the number k in the k-WTA problem is implicitly specified by the initial states of the neurons. Based on the constraint conversion for a classical optimization problem formulation of the k-WTA, via modifying the traditional gradient descent, we propose an initialization-based k-WTA neural network model with only n neurons for n-dimensional inputs, and the dynamics of the neural network model is described by parameterized gradient descent. Theoretical results show that the state vector of the proposed k-WTA neural network model globally asymptotically converges to the theoretical k-WTA solution under mild conditions. Simulative examples demonstrate the effectiveness of the proposed model and indicate that its convergence can be accelerated by readily setting two design parameters.