Abstracts:Route optimization is widely utilized in consumer electronics like Google Maps and Facebook connections. However, due to the limitations of consumer devices, users often rely on online service providers to process large-model-enabled route services, which may involve sensitive information such as location and social relationships. This raises privacy concerns, necessitating the need for a system that ensures expected utility while protecting the privacy of both consumers and service providers. This problem, known as Symmetric Private Shortest Path Retrieval, has been studied in the literature, either focusing on specific data types or a simple private setting where the data user and owner are the same. In this paper, we propose a low-latency symmetric private shortest path query for generic graph data in a more realistic public service model. We propose several novel techniques, including a chained encryption method and an oblivious key transfer mechanism to efficiently reveal the queried shortest path without leaking additional information to either the consumers or service provider. Experimental results demonstrate that our scheme outperforms the state-of-the-art by being approximately 46 to 53 times faster and consuming roughly 36 to 74 times less bandwidth for a city navigation service.

Abstracts:The proliferation of consumer electronics systems in swarm environments has heightened the need for robust privacy-preserving techniques in federated learning (FL) frameworks. Traditional FL methods, while effective in distributing model training, often fall short in efficiently handling data removal requests and safeguarding against inference attacks. Existing solutions, such as FL with Differential Privacy (FL+DP), offer some improvements but still face challenges in balancing performance, privacy, and communication efficiency. In this paper, we introduce FL+Fuzzy Unlearning, a novel FL framework that integrates Fuzzy Logic to address these challenges in federated multi-task learning for swarm consumer electronics systems. Our method aims to maintain high model performance while ensuring robust privacy and efficient communication. We evaluated FL+Fuzzy Unlearning against Traditional FL and FL+DP using two benchmark datasets: MNIST and MovieLens. Our results demonstrate that the proposed approach attains comparable or superior accuracy (0.86 on MNIST and 0.88 on MovieLens) while achieving robust unlearning performance. Moreover, FL+Fuzzy Unlearning substantially enhances privacy protection, evidenced by elevated inference attack failure rates (0.75 for MNIST and 0.78 for MovieLens), and effectively reduces communication overhead, underscoring its efficiency in data exchange. Overall, FL+Fuzzy Unlearning presents a promising, balanced solution for privacy, efficiency, and adaptability, making it highly suitable for real-world applications in swarm consumer electronics systems.

Abstracts:Contrastive loss and its variants are very popular for visual representation learning in an unsupervised scenario, where positive and negative pairs are produced to train a feature encoder based on data augmentation. However, treating only the original image and its augmented versions as positive pairs may lead to problems, since two training images that share the same object category may be misclassified as negative pairs. To address this issue, this paper introduces a plug-in-play affinity similarity module (ASM) used in any contrastive learning framework for unsupervised visual representation learning. In this approach, positive and negative pairs are determined not only based on data augmentation, but also according to the similarity of their feature embeddings. The core idea is that two training samples with higher feature similarities are likely to belong to the same object category, suggesting that they should be classified as a positive pair with high probability, and vice versa for negative pairs. Accordingly, we have also developed an improved affinity similarity-based contrastive loss (ASCL) that leverages the newly generated positive and negative training pairs produced by the ASM. To demonstrate the effectiveness of our method, we extensively evaluate its performance on the ImageNet dataset, showing that it achieves satisfactory results.

Abstracts:The Internet of Things (IoT) and unmanned aerial vehicles (UAVs) continue to advance the low-carbon smart agriculture technologies for next-generation consumer electronics and unlock more informed agricultural practices. Reinforcement learning (RL), federated learning (FL), and federated reinforcement learning (FRL) have demonstrated notable achievements in resolving complex problems, including resource allocation, energy efficiency, anomaly detection, and bandwidth utilization for multimodal tasks. This research explores multimodal data analysis and resource optimization using FRL for agricultural consumer electronics. The proposed framework employs IoT devices to monitor temperature, humidity, soil temperature, and soil moisture in real time, while UAVs provide aerial imagery for soil moisture, crop growth, and pest identification across three fields. This framework supports distributed learning, which trains local RL models on each node and combines them into the global model. The proposed FRL model demonstrated significant enhancements, including a 17% reduction in energy consumption for IoT devices and a 15% reduction for UAVs compared to non-FRL methods. This research emphasizes the effectiveness of FRL in integrating IoT and UAV for efficient resource allocation, energy efficiency, and reduced carbon emissions for low-carbon agricultural consumer electronics.

Abstracts:Quantum computers leverage the principles of quantum mechanics, including superposition and entanglement, allowing them to execute specific computations significantly faster than classical computers. The gate model is a common way to implement quantum algorithms, where the algorithms are broken down into a sequence of simple gates that operate on one or more quantum bits. This manipulation of a quantum computer involves a succession of unitary transformations that affect the different components of the superposition simultaneously, enabling significant parallel data processing and reducing the time of execution. As a result of these capabilities, quantum technology is expected to provide abilities and performance that are currently unattainable by classical methods. However, quantum hardware is under development and is prone to errors, which can negatively impact the performance of quantum methods. To address this challenge, error mitigation techniques are developed to reduce the impact of errors on the final outcomes. By leveraging the speedup offered by quantum hardware and using effective error mitigation techniques, quantum computing holds the promise of outperforming classical methods in various consumer applications (CA).

Abstracts:Accurate exchange rate prediction plays a crucial role in promoting global economic stability and growth. Exchange rates being volatile have an impact on trade balance, inflation rates and monetary policies and have an impact on global economies as well as the domestic economy. This ability to analyze and predict such changes enables the policy-makers, economists, financial analysts and the national banks draw better policies. However, it is hard to forecast exchange rate because of the complexity of the financial systems and the overwhelming number of factors that constitutes an indicator to the exchange rates. Traditional models often struggle to capture non-linear relationship and high volatility prevailing into the financial data. Problems related to variability and the existence of a large number of economic indicators make analysis with these models less accurate. Moreover, they fail at incorporating macro-economic aspects and other market indicators effectively into your plans which leads to more often than not incomplete and sometimes even incorrect prognoses. To tackle these challenges, this research presents the F-CER (Forecasting Currency Exchange Rates) model which utilizes more advanced features like Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), Multi-Head Attention, and Reinforcement learning (RL). By using the analysis of the historical rate of exchange of Chinese Yuan and U.S. Dollar from 2000 to June 2023, and by incorporating relevant macroeconomic and market factors, F-CER successfully overcomes the shortcomings of the above models. The F-CER model applies advanced techniques for feature engineering as well as Principal Component Analysis and adaptive normalization to overcome the data dimensionality and variability issues. It finds significant shapes in noisy and massive time series and focuses on key variables that contribute to the enhanced accuracy of forecasts. Experimental results indicate that F-CER model perform well, and achieving a Mean Squared Error (MSE) of 0.005, Test Mean Absolute Error (MAE) of 0.089, Demonstrating its advanced predictive capabilities.

Abstracts:Motor imagery (MI)-based brain-computer interfaces (BCIs) using electroencephalography (EEG) have been widely adopted due to their portability and safety. However, the non-stationary nature of EEG signals introduces significant variability between individuals, making it difficult to directly use data from other subjects (source domain) to decode the intentions of a new subject (target domain). To tackle this issue, this paper proposes a domain adaptation framework by aligning the inverse Gram matrices for cross-subject motor imagery classification (DAIG), which consists of a feature extraction module and the aligns the inverse Gram matrices module. Specifically, in the feature extraction module, EEGBiMamba is proposed to capture the temporal dependencies in one-dimensional EEG time-series representations with a bidirectional (forward and backward) modeling. For the extracted EEG features from both domains, the model aligns their inverse Gram matrices from the perspectives of angle and scale, effectively minimizing the differences in feature distributions. Furthermore, the multilayer perceptron (MLP) layer is replaced with a Kolmogorov-Arnold Network (KAN) to generate an M-dimensional vector for classification, improving the model’s expressiveness and adaptability. The proposed method is evaluated on two public datasets, BCI IV IIa and BCI IV IIb, along with the AMI_MD dataset collected in our laboratory. The results show that the proposed method achieved average cross-subject decoding accuracies of 68.58%, 86.72%, and 81.65% across different datasets, representing remarkable performance improvements of 17.06%, 13.6%, and 15.76% over baseline methods, respectively.

Abstracts:This paper investigates the data-driven decentralized resilient control problem for large-scale systems (LSS) under randomly occurring Denial-of-Service (DoS) attacks. A min-max optimization criterion is established based on zero-sum differential game theory, and the corresponding optimal control strategy is derived. Global asymptotic stability of the closed-loop LSS is theoretically guaranteed under the proposed control scheme. A two-stage adaptive dynamic programming (ADP) algorithm, integrating reinforcement learning techniques with local state feedback, is proposed to derive the optimal control policy without requiring prior knowledge of the system model. Simulations are conducted in MATLAB on a multimachine power system benchmark. In particular, the two-stage ADP controller shortens the settling time by up to 7.7% and reduces overshooting by over 14.5% compared to the existing methods, thereby validating its robustness and superior performance in dynamic and adversarial environments.

Abstracts:In vehicular edge computing (VEC) Consumer Electronics networks, the integration of holographic counterpart technology presents significant challenges due to its stringent requirements for high data transmission rates and communication reliability. Traditional task offloading methods, constrained by suboptimal communication link quality and energy limitations, are inadequate to meet these demands. This paper introduces a groundbreaking system that synergistically combines wireless power transfer (WPT) and reconfigurable intelligent surfaces (RIS) to significantly enhance both communication performance and computational efficiency. Leveraging deep reinforcement learning (DRL), our system achieves joint optimization of task offloading strategies and resource allocation. Departing from conventional dynamic RIS designs, we implement a fixed phase shift matrix approach, which not only simplifies system implementation but also reduces computational complexity, thereby enhancing both task offloading efficiency and system stability. Extensive simulation results demonstrate that our optimized RIS-assisted approach achieves a remarkable 38.30% improvement in computational rates compared to non-RIS schemes and a 4.83% enhancement over random-phase RIS configurations. These substantial improvements highlight the transformative potential of RIS in boosting computation rates and providing robust solutions for high-demand task offloading scenarios. Our innovative system design represents a significant advancement in intelligent vehicular networks and edge computing technologies, offering substantial application potential for holographic projection task offloading in next-generation vehicular systems.