Integrating federated learning (FL) with differential privacy (DP) into regional carbon allowances management and trading enables collaborative model training that protects both local data privacy and model security. However, the noise injection in DP degrades model accuracy, necessitating additional local training and increasing the overall computational overhead for enterprises. This overhead reduces the willingness of rational and self-interested participants to contribute to model training. To address this challenge, this study models the interaction between the regulator and enterprises as a Stackelberg game and proposes a scalable, bandwidth-efficient FL framework for regional carbon allowances management and trading. The framework adaptively determines the number of local training iterations based on enterprises’ privacy budgets and employs a submodular enterprise selection strategy for aggregation, which decreases redundant uploads and enhances global model quality. Furthermore, a deep reinforcement learning–based multi-objective differential evolution with bi-level optimization (DRL-MODE2) algorithm is developed to efficiently solve large-scale, high-dimensional Stackelberg games. Comprehensive experiments were conducted on both a carbon emission time-series dataset and the MNIST benchmark dataset under different enterprise scales, incentive mechanisms, and optimization algorithms. Experimental results demonstrate that the proposed framework incentivizes enterprise participation and improves model accuracy, without compromising privacy preservation and communication efficiency.

Recent advancements in adversarial attacks have demonstrated their effectiveness in misleading speaker recognition models, making wrong predictions about speaker identities. On the other hand, defense techniques against speaker-adversarial attacks focus on reducing the effects of speaker-adversarial perturbations on speaker attribute extraction. These techniques do not seek to fully remove the perturbations and restore the original speech. To this end, this paper studies the removability of speaker-adversarial perturbations. Specifically, the investigation is conducted assuming various degrees of awareness of the perturbation generator across three scenarios: ignorant, semi-informed, and well-informed. Besides, we consider both the optimization-based and feedforward perturbation generation methods. Experiments conducted on the LibriSpeech dataset demonstrated that: 1) in the ignorant scenario, speaker-adversarial perturbations cannot be eliminated, although their impact on speaker attribute extraction is reduced, 2) in the semi-informed scenario, the speaker-adversarial perturbations cannot be fully removed, while those generated by the feedforward model can be considerably reduced, and 3) in the well-informed scenario, speaker-adversarial perturbations are nearly eliminated, allowing for the restoration of the original speech.

Federated learning (FL) is a framework for distributed privacy-preserving computation, which enables global model aggregation by receiving client-trained models without accessing clients’ raw data. Therefore, developing an efficient incentive mechanism to promote client participation within this secure framework is a primary challenge in this field. To address this, we consider an FL system with differential privacy (DP) and model the interactions between the server and clients as a two-stage Stackelberg game aimed at minimizing server operating costs and maximizing client rewards. Then, a quality-based adaptive differential evolution-2 algorithm (Q-ADE2) is proposed to approximately solve the game model, which involves optimizing the server’s reward allocation to clients and determining the clients’ privacy budgets. In this approach, the quality coefficient is determined by the effectiveness of the FL system, evaluated based on the improvement in model accuracy. Finally, the standard MNIST dataset is employed in three comparative experiments, demonstrating that the proposed algorithm exhibits strong solving performance and effectively incentivizes client participation in FL training.

Speaker attribute perturbation offers a feasible approach to asynchronous voice anonymization by employing adversarially perturbed speech as anonymized output. In order to enhance the identity unlinkability among anonymized utterances from the same original speaker, the targeted attack training strategy is usually applied to anonymize the utterances to a common designated speaker. However, this strategy may violate the privacy of the designated speaker who is an actual speaker. To mitigate this risk, this paper proposes an any-to-any training strategy. It is accomplished by defining a batch mean loss to anonymize the utterances from various speakers within a training mini-batch to a common pseudo-speaker, which is approximated as the average speaker in the mini-batch. Based on this, a speaker-adversarial speech generation model is proposed, incorporating the supervision from both the untargeted attack and the any-to-any strategies. The speaker attribute perturbations are generated and incorporated into the original speech to produce its anonymized version. The effectiveness of the proposed model was justified in asynchronous voice anonymization through experiments conducted on the LibriSpeech datasets. Additional experiments were carried out to explore the potential limitations of speaker-adversarial speech in voice privacy protection. With them, we aim to provide insights for future research on its protective efficacy against black-box speaker extractors and adaptive attacks, as well as generalization to out-of-domain datasets and stability. Audio samples and open-source code are published in https://github.com/VoicePrivacy/any-to-any-speaker-attribute-perturbation.

Neural networks are commonly known to be vulnerable to adversarial attacks mounted through subtle perturbation on the input data. Recent development in voice-privacy protection has shown the positive use cases of the same technique to conceal speaker’s voice attribute with additive perturbation signal generated by an adversarial network. This paper examines the reversibility property where an entity generating the adversarial perturbations is authorized to remove them and restore original speech (e.g., the speaker him/herself). A similar technique could also be used by an investigator to deanonymize a voice-protected speech to restore criminals’ identities in security and forensic analysis. In this setting, the perturbation generative module is assumed to be known in the removal process. To this end, a joint training of perturbation generation and removal modules is proposed. Experimental results on the LibriSpeech dataset demonstrated that the subtle perturbations added to the original speech can be predicted from the anonymized speech while achieving the goal of privacy protection. By removing these perturbations from the anonymized sample, the original speech can be restored. Audio samples can be found in https://voiceprivacy.github.io/Perturbation-Generation-Removal/.

The industrial IoT communication environment is complex and diverse. There are many running industrial production equipment in the intelligent factory based on the Industrial Internet of Things. At the same time, a large number of servers are deployed in different areas of the factory to control the production of industrial equipment and user interfacing. These servers work together to provide a variety of services to industrial equipment and users. In intelligent factory communication architecture, multiple servers can provide similar or identical services to a single user or industrial device. However, the characteristics of the intelligent factory network based on the Industrial Internet of Things (IIoT) are dynamic and changeable. And due to space and time factors, the dynamic QoS properties of each server node in the industrial IoT are unstable. These factors make it difficult for users to select the service that meets their needs in the candidate service set and increase the task processing latency of communication nodes. To address the above issues, we focus on QoS analysis and prediction methods for industrial IoT and propose an enhanced QoS prediction model based on EMD-mLSTM (EMD-EmLSTM). This model combines EMD (Empirical Mode Decomposition) with LSTM (Long Short-Term Memory) methods. We also design the multivariate data input model to mine the potential association of QoS time series data from a finer granularity level. In the data preprocessing section, we propose an abnormal data processing method, which uses the isolated forest algorithm to detect outliers in the dataset. Meanwhile, a residual correction prediction model based on LSTM is constructed. This auxiliary model can correct the prediction results by residual prediction. It can help the model to enhance the QoS prediction accuracy. The experimental results show that the EMD-EmLSTM model can achieve single-step and multi-step prediction of dynamic QoS attributes of industrial IoT. The prediction effect is better than other methods. Moreover, our proposed method is able to select the optimal service from the set of candidate services. It effectively reduces the node task processing delay and ensures the best quality of service in a certain time period of the network nodes.

Recently, fake news with text and images have achieved more effective diffusion than text-only fake news, raising a severe issue of multimodal fake news detection. Current studies on this issue have made significant contributions to developing multimodal models, but they are defective in modeling the multimodal content sufficiently. Most of them only preliminarily model the basic semantics of the images as a supplement to the text, which limits their performance on detection. In this paper, we find three valuable text-image correlations in multimodal fake news: entity inconsistency, mutual enhancement, and text complementation. To effectively capture these multimodal clues, we innovatively extract visual entities (such as celebrities and landmarks) to understand the news-related high-level semantics of images, and then model the multimodal entity inconsistency and mutual enhancement with the help of visual entities. Moreover, we extract the embedded text in images as the complementation of the original text. All things considered, we propose a novel entity-enhanced multimodal fusion framework, which simultaneously models three cross-modal correlations to detect diverse multimodal fake news. Extensive experiments demonstrate the superiority of our model compared to the state of the art.