The rapid growth of online short texts has made specialized analysis essential, as these texts are sparse and information-limited. Short text clustering (STC) is critical for automatically grouping unlabeled texts into meaningful clusters, supporting applications such as sentiment analysis, spam filtering, and social media personalization. In the context of massive online content, deep clustering seeks to uncover semantic categories by measuring distances in the representation space. Consequently, aligning clustering pseudo-labels with the true category distribution is crucial for effective self-supervised training, particularly under class imbalance and distribution skew commonly observed in web data. To address this challenge, we propose the Cluster Identification-Guided Dual Correction (CIDC) framework, which generates reliable pseudo-labels to guide deep clustering. Specifically, given cluster partitions and model-estimated class distributions, we perform Cluster Category Identification (CCI) at each training epoch to determine the most probable category for each cluster. This identification provides the foundation for the Pseudo-Label Correction (PLC) and Prototype-Based Correction (PBC) modules, which jointly enhance pseudo-label reliability and representation learning. In the PLC module, samples whose model-estimated class distributions conflict with the assigned cluster category are corrected, thereby improving semantic alignment within clusters. In the PBC module, representative and reliable prototypes are selected according to cluster categories and model predictions to guide training, further strengthening representation discriminability. Extensive experiments demonstrate that CIDC consistently outperforms existing methods in terms of clustering accuracy and mutual information, particularly in unsupervised settings characterized by class imbalance and noisy data.

Lithium-ion batteries (LIBs) are widely used in electric vehicles and energy storage systems, where accurate State of Health (SoH) prediction is essential for safety and reliability. While deep learning has shown promise for SoH modeling, obtaining high-quality training data remains challenging due to high testing costs and privacy constraints. This study proposes an enhanced federated learning (FL) approach, enabling multiple institutions to collaboratively train SoH models while preserving local data privacy. To address communication overhead in resource-constrained Internet of Things (IoT) environments and data heterogeneity across non-IID battery datasets, we introduce an adaptive wavelet-based compression framework. An Agent-specific Adaptive Top-K Sparsification (AATKS) mechanism dynamically adjusts the sparsification rate for each agent, mitigating the impact of data heterogeneity by preventing excessive information loss in critical updates while ensuring efficient communication. Additionally, a Layer-wise Energy Weighted Aggregation (LEWAgg) strategy enhances global model consistency by prioritizing high-energy coefficients, leading to more stable learning across heterogeneous clients. Experiments on three real-world LIBs datasets demonstrate that the proposed WCFL framework achieves a 14W reduction in communication cost with less than 0.75% accuracy loss compared to centralized training. This study provides an efficient and scalable solution for collaborative battery health management in practical IoT settings.

Abstract Copy-move is a common image forgery operation, which copies and moves a block of an image from one position to another place. Image hashing refers to extracting a unique number sequence from the image by using various image features. In practical application, image hashing is used to replace the image itself, which effectively reduces the cost of image storage and computational complexity. In this paper, we propose a novel image hash extraction scheme: constructing image hashing by combining local feature based on Canny operator and global feature based on tensor. In addition, instead of using the traditional correlation coefficient or Hamming distance, a novel method is proposed to calculate the hash distances. A large number of experiments have proved that our image hashing can achieve a better balance between robustness and discrimination with a shorter hash length. What’s more, we can directly locate the forgery areas from the hashing for copy-move forged images.

As the proliferation of 5th Generation Mobile Communication Technology (5G) accelerates the adoption of Internet of Things (IoT) applications, building robust and secure communication channel becomes increasingly crucial with the exponential growth of connected devices. The 3rd Generation Partnership Project (3GPP) has established security standards for 5G systems, including mechanisms such as the 5G-Authentication and Key Agreement (5G-AKA), which enables establish secure sessions in untrustworthy participants or insecure channels. The private key which untrustworthy parties have independently or transmitted through insecure channels, may involve risk of information leakage in 5G-AKA. Motivated by this challenge, we propose a practical and secure dual-server key agreement scheme based on password authentication for IoT devices in 5G networks. The scheme ensures secure reliable key storage and key transmission, mitigating risks associated with key information leakage through a dual-server architecture and three-lock security policy. Importantly, we avoid ownership of the complete key by any untrustworthy entity in insecure 5G network to ensure key security. The scheme can resilience to various security threats prevalent in 5G networks through rigorous formal security. We analyze the communication and computational loads to illustrate the protocol’s practicality and efficacy.

Intrusion detection systems play a very important role in industrial Internet network security. However, in the large‐scale, complex, and heterogeneous industrial Internet of Things (IoT), it is becoming more and more difficult to defend network intrusion threats due to the insufficiency of high‐quality attack samples. To solve the problem, an efficient federated network intrusion method called EFedID is proposed for industrial IoT, which can allow different industrial agents to collaboratively train a comprehensive detection model. Specifically, the adaptive gradient sparsification method is introduced to alleviate the communication and computation overheads. To protect the data privacy of the agents, a CKKS cryptosystem‐based secure communication protocol is designed to encrypt the model parameters through the federated training process. Our proposed system demonstrates exceptional detection performance on the NSL‐KDD, KDD CUP 99, and CICIDS 2017 datasets. Notably, on the NSL‐KDD dataset, the model compression rate reaches 9 times while the model accuracy reaches 84.31%. On the KDD CUP 99 dataset, the model compression rate reaches 8.9 times while the model accuracy reaches 97.3%. Lastly, on the CICIDS 2017 dataset, the model compression rate reached 6.173 times while the model accuracy reached 95.51%. The experimental results demonstrate that the proposed method is very suitable for effectively developing a high‐accuracy detection model while protecting the data information of industrial agents. Furthermore, the method can be extended to other recent deep learning networks for intrusion detection.

Federated learning (FL) as an emerging learning paradigm, has been achieved widespread attention since it allows distributed industrial agents to collaboratively develop a global model while keeping their data locally. Although various FL-based algorithms were proposed to solve engineering tasks in industrial cyber-physical systems, existing works rarely study a practical problem that the training samples collected by certain industrial agents (called unreliable industrial agents) may be of low quality. Obviously, the unreliable industrial agent would degrade the model accuracy. In this article, we propose a privacy-preserving momentum federated learning considering unreliable industrial agents, named DetectPMFL. In DetectPMFL, we design a detection method to alleviate the adverse effect of the unreliable agents. In addition, the privacy issues are analyzed by the mathematical description, especially for the convolution neural network. Based on this, Cheon-Kim-Kim-Song (CKKS) homomorphic encryption is used to protect the private information of the agents. The proposed approach is evaluated by two common datasets for recognition tasks. The security analysis and experiment results indicate that DetectPMFL is robust against unreliable industrial agents, and achieves high accuracy while preserving privacy.

Mengting Hu, Shiwan Zhao, Honglei Guo, Chao Xue, Hang Gao, Tiegang Gao, Renhong Cheng, Zhong Su. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

A publicly verifiable scheme for the copyright protection of digital image is proposed in this paper. Combining with some cryptographic techniques such as digital signature and timestamp, the scheme features an idea of registering watermark information to a trusted authority (TA) rather than embedding it into the host image, which overcomes many deficiencies of the conventional watermarking algorithm. In the scheme, generalized brain-state-in-a-box neural network (gBSB) and error diffusion halftoning are employed to extract the robust feature, which is further used to generate the verification information registered to TA, from the original image. Experimental results demonstrate that the feature is adequately robust to make sure that the verifier will be able to extract the logo mark from the attacked image correctly. The proposed scheme is competent to be applied to the copyright protection of digital multimedia.

The design of an efficient one-way hash function with good performance is a hot spot in modern cryptography researches. In this paper, a hash function construction method based on cell neural network (CNN) with hyper-chaos characteristics is proposed. The chaos sequence generated by iterating CNN with Runge-Kutta algorithm, then the sequence iterates with every bit of the plaintext continually. Then hash code is obtained through the corresponding transform of the latter chaos sequence from iteration. Hash code with different length could be generated from the former hash result. Simulation and analysis demonstrate that the new method has the merit of convenience, high sensitivity to initial values, good hash performance, especially the strong stability, even if the hash code length is short relatively.