In modern industrial machining processes, cutting tools play an important role, and their health directly affects the quality of machined parts. However, changing cutting tools based on experience not only increases costs but also reduces productivity. Therefore, predicting the future health of the cutting tools in advance can enable cutting tools changes to be carried out at the right time. The existing health status assessment of a cutting tools consists of three main areas: cutting tools wear prediction, health stages division, and reliability assessment. But traditional deep learning models usually process these three tasks separately, ignoring the correlation that exists between the three tasks. In order to solve this problem, this paper proposes a multi-task model based on temporal convolutional network (TCN) and progressive layered extraction (PLE) network. Firstly, the pre-processed data are subjected to feature extraction and feature selection and the redundant information between the features is reduced by using an autoencoder. Secondly, the TCN network module is used to extract the correlation feature information of multiple tasks in time, the PLE network module learns the difference information between each task, and finally, the prediction output is made by each sub-task module for each sub-task. Finally, the dynamic weight average method is used to adjust the weights of the loss function, which avoids manual adjustment of the weights. The experimental results show that the method not only achieves the prediction of multiple tasks but also has high prediction accuracy.

Text-image retrieval is a key challenge in computer vision and natural language processing, aiming to retrieve the most semantically relevant image or text given a query in the opposite modality. However, growing privacy and security concerns make traditional centralized learning approaches increasingly unsuitable for handling sensitive multimodal data. In this paper, we propose FedBi-GNNs, a federated learning framework for bimodal graph neural networks, which enables collaborative training across decentralized clients without sharing private data. Each client independently constructs heterogeneous graphs from local text and image data and learns correspondences via bimodal graph matching. These local representations are then aggregated at a central server using a heterogeneous federated aggregation scheme. Empirical results on the MSCOCO benchmark demonstrate that FedBi-GNNs significantly outperform existing state-of-the-art methods, offering improved retrieval accuracy, enhanced privacy preservation, and greater robustness to data heterogeneity across clients.

In this paper, an unscented Kalman filtering problem is considered for a class of nonlinear systems with stochastic nonlinearities under the FlexRay protocol. The phenomenon of stochastic nonlinearities is characterized by the statistical means to account for engineering practice. Moreover, with the FlexRay protocol implemented between the sensors and the filter, an appropriate measurement model is established to characterize the measurement outputs after data transmission via the FlexRay protocol. By considering the stochastic nonlinearities and the FlexRay protocol, an tailored unscented Kalman filtering algorithm is designed where the influence of the stochastic nonlinearities and the FlexRay protocol is quantified. In the end, the effectiveness of the proposed filtering algorithm is verified in estimating the state of nonlinear systems through simulation experiments.

This article focuses on examining the sampled-data based containment control (CC) issue for nonlinear multiagent systems (MASs) with dynamic leaders and input saturation. The proposed control protocol requires that the information is exchanged and calculated only at the sampling instants with the aim of conserving communication resources, and the protocol incorporates the control saturation as well. The CC is analyzed by means of the algebraic graph theory, m-matrix theory and Halanay-type inequality, etc. Some criteria are derived to ensure the MAS can realize the CC under the control protocol, and in the meantime, a CC region is also given ensuring that all the followers with their initial stacked states in it will converge ultimately to the convex hull formed by the leaders. Furthermore, the design of the control gain can be carried out by searching for feasible solutions to a group of matrix inequalities. Finally, a numerical illustration is provided to substantiate the efficacy of the theoretical findings.

This paper addresses the problem of distributed recursive filtering for state-saturated systems in a networked communication environment. An output mask function is employed to safeguard the privacy of interaction data during node exchange in sensor networks. Scaled uniform quantization is introduced to facilitate the digital communication and optimize the network resource usage. The primary objective of the study is to design a distributed recursive filter that ensures the filtering error covariance remains bounded over a finite horizon. Specifically, by using Riccati-like equations, an upper bound for the filtering error covariance is derived, which depends on the network topology, the output mask function, and the quantization level. The desired gain matrix is then solved recursively. Finally, the effectiveness of the proposed filtering algorithm is demonstrated through a three-tank simulation example.

The integration of large language models (LLMs) into creative applications has unlocked new capabilities but also introduced vulnerabilities, notably prompt injections. These are malicious inputs designed to manipulate model responses, posing threats to security, privacy, and functionality. This paper delves into the mechanisms of prompt injections, their impacts, and presents novel detection strategies. More specifically, the necessity for robust detection systems is outlined, a predefined list of banned terms is combined to embed techniques for similarity search, and a BERT (Bidirectional Encoder Representations from Transformers) model is built to identify and mitigate prompt injections effectively with the aim to neutralize prompt injections in real-time. The research highlights the challenges in balancing security with usability, evolving attack vectors, and LLM limitationsm, and emphasizes the significance of securing LLM-integrated applications against prompt injections to preserve data privacy, user trust, and uphold ethical standards. This work aims to foster collaboration for developing standardized security frameworks, contributing to more safer and reliable AI-driven systems.

This paper investigates the least-squares linear estimation problem for multirate systems with stochastic parameter matrices, under the influence of random denial-of-service (DoS) attacks. These attacks can severely impair the performance of estimation algorithms by causing intermittent loss of measurement data. To counteract the adverse effect of DoS attacks, two compensation strategies-hold-input and prediction compensation- are used. For each of these strategies, specific recursive filtering and smoothing algorithms are designed. A key advantage of the proposed methodology is its ability to operate without requiring a detailed signal evolution model, relying only on the mean and covariance functions of the involved processes. The effectiveness of the proposed approaches is validated through numerical simulations, which highlight how common network-induced phenomena, such as missing observations, can be incorporated into the framework of systems with random parameter matrices and, additionally, they provide insights into estimation performance under different attack probabilities.