Federated learning, a potent paradigm for collaborative machine learning across multiple parties, offers significant promise for contemporary industries. Nonetheless, its collaborative essence necessitates addressing concerns pertaining to data security and privacy. Sensitive user information, encompassing preferences, behaviors, and identities, remains vulnerable to adversarial analysis, thereby revealing the inadequacies of conventional privacy preservation strategies within federated learning frameworks. To mitigate these challenges, this paper proposes GSFL, an innovative federated learning architecture that amalgamates smart contracts with group signatures. GSFL facilitates secure and reliable distributed machine learning data sharing, while concurrently bolstering privacy protection. Furthermore, its enhanced decentralization fosters greater user participation in federated learning initiatives. Empirical analysis and testing validate GSFL's efficacy in satisfying the prerequisites for data sharing and privacy preservation in federated learning contexts.

In the digital era, the internet and social media have transformed communication but have also facilitated the spread of hate speech and disinformation, leading to radicalization, polarization, and toxicity. This is especially concerning for media outlets due to their significant role in shaping public discourse. This study examines the topics, sentiments, and hate prevalence in 337,807 response messages (website comments and tweets) to news from five Spanish media outlets (La Vanguardia, ABC, El País, El Mundo, and 20 Minutos) in January 2021. These public reactions were originally labeled as distinct types of hate by experts following an original procedure, and they are now classified into three sentiment values (negative, neutral, or positive) and main topics. The BERTopic unsupervised framework was used to extract 81 topics, manually named with the help of Large Language Models (LLMs) and grouped into nine primary categories. Results show social issues ( 22.22%), expressions and slang ( 20.35%), and political issues ( 11.80%) as the most discussed. Content is mainly negative ( 62.7%) and neutral ( 28.57%), with low positivity ( 8.73%). Toxic narratives relate to conversation expressions, gender, feminism, and COVID-19. Despite low levels of hate speech ( 3.98%), the study confirms high toxicity in online responses to social and political topics.

As benchmark image datasets expand in sample size and feature complexity, the challenge of managing increased dimensionality becomes apparent. Contrary to the expectation that more features equate to enhanced information and improved outcomes, the curse of dimensionality often hampers performance. This paper reviews existing literature on filter feature selection techniques applied to image features, highlighting when they are applied to both classical and deep-learning-based feature extraction methods. Additionally, this study explores how different feature selection methods behave when applied to image features through big data technologies. Different experiments were performed to compare the results when using feature selection techniques with various reduction percentages. Experimental results demonstrated that an important reduction of the extracted features provides classification results similar to those obtained with the full set of features. Furthermore, applying dimensionality reduction techniques outperforms, in some cases, the results achieved using broad feature vectors.

The understanding of financial behavior, especially its application in the stock market, has become increasingly important in recent years due to its significant impact on the global economy. One field that explores the intersection of finance and computer science to create predictive models is known as stock market prediction. This field aims to predict the behavior of various securities in the financial market. Deep Learning, one of the most renowned and utilized techniques, consists of various deep neural network structures that facilitate learning from non-linear models. In this study, we utilized open data from some of Brazil’s largest companies – Petrobras (PETR4), Itausa (ITSA4), and Vale (VALE3) – provided by BovDB, which includes stock quote data for all companies listed on the Brazilian stock market from 2000 to 2020. The data for the years in question were processed using a recurrent neural network to assess the impact of a price correction factor that accounts for the influence of past events not included in the training and validation results of the RNN model. The findings indicate a strong correlation of the model with temporal data and suggest a positive effect on reducing noise and forecast errors during model training.

Image inpainting represents a fundamental task in computer vision, primarily focusing on the generation of missing content within an image to restore its integrity and aesthetics. Existing GAN-based approaches often yield content with ambiguity and entail high training costs. They tend to concentrate narrowly on damaged regions, leading to distortions along edges, which consequently hampers generalization. To overcome these challenges and achieve high-fidelity image inpainting, we introduce an image editing algorithm to the image inpainting task by designing two distinct networks. The first network, Edge-e4e, uses pretrained StyleGAN2 for global image generation, mitigating edge distortions in damaged regions and reducing training costs. Simultaneously, we incorporate contour information in the damaged areas to ensure the correctness of the restoration content. The second network, the Appending network, includes two style-based encoders and a generator to refine the images restored by the Edge-e4e network. Specifically, we subtract the restored images from the input images in the channel dimension to obtain a distortion map, which serves as a prior to refine the restored images. The encoders extract features from the input images and distortion map, while the generator is employed to generate optimized images. To enhance the quality of refined images, we propose integrating plugin and modulate plugin modules into the Appending network for style extraction and fusion, leveraging the available information from input images and blending it into the generator. Experimental results demonstrate that our algorithm achieves high-fidelity restoration and excellent generalization, with optimal FID and Lpips metrics of 0.0631 and 0.875, respectively.

Public security is a crucial aspect of maintaining social order. Although crime rates in western cultures may be considered socially acceptable, it is important to continually improve security measures to prevent potential risks.  With the advancements in artificial intelligence methods, particularly in deep learning and computer vision, it has become possible to detect abnormal event patterns in groups of people. This paper presents a systematic review of deep learning techniques employed for identifying gatherings of people and detecting anomalous events to enhance public security. Some of the open research areas are identified, including the lack of works addressing multiple cases of anomalies in large concentrations of people, which leaves open an important avenue for future scientific work.

Knowledge graphs convey precise semantic information that can be effectively interpreted by neural networks, and generating descriptive text based on these graphs places significant emphasis on content consistency. However, knowledge graphs are inadequate for providing additional linguistic features such as paragraph structure and expressive modes, making it challenging to ensure content coherence in generating text that spans multiple sentences. This lack of coherence can further compromise the overall consistency of the content within a paragraph. In this work, we present the generation of scientific abstracts by leveraging knowledge graphs, with a focus on enhancing both content consistency and coherence. In particular, we construct the ACL Abstract Graph Dataset (ACL-AGD) which pairs knowledge graphs with text, incorporating sentence labels to guide text structure and diverse expressions. We then implement a Siamese network to complement and concretize the entities and relations based on paragraph structure by accomplishing two tasks: graph-to-text generation and entity alignment. Extensive experiments demonstrate that the logical paragraphs generated by our method exhibit entities with a uniform position distribution and appropriate frequency. In terms of content, our method accurately represents the information encoded in the knowledge graph, prevents the generation of irrelevant content, and achieves coherent and non-redundant adjacent sentences, even with a shared knowledge graph.

Background With the rapid development of artificial intelligence technology, highly intelligent and unmanned factories have become an important trend. In the complex environments of smart factories, the long-term tracking and inspection of specified targets such as operators and special goods, as well as comprehensive visual recognition and decision-making capabilities throughout the whole production process, are critical components of automated unmanned factories. It is inevitable that issues such as target occlusion and disappearance will occur, thus exacerbating the long-term tracking challenge. Currently, there are no long-term tracking studies specifically addressing occlusions in these environments. Methods We first construct three new benchmark datasets in the complex workshop environment of a smart factory (referred to as SF-Complex3 data), which include challenging conditions such as complete occlusion and partial occlusion of targets. Next, utilising a brain memory inspired approach, we determine uncertainty estimation parameters: confidence, peak-to-sidelobe ratio (PSR), and average peak to correlation energy (APCE), to derive a continual learning based adaptive model update method. Additionally, we design a lightweight target detection model to automatically detect and locate targets in the initial frame and during re-detection. Finally, we integrate the algorithm with ground mobile robots and UAV-based imaging and processing equipment to build a new visual detection and tracking framework, termed SFC-RT (Smart Factory Complex Tracking and Identification). Results We conducted extensive tests on the benchmark UAV20L and SF-Complex3 datasets. Compared to state-of-the-art tracking algorithms, our proposed algorithm demonstrates an average performance improvement of 6% when facing key challenging attributes. Moreover, it can smoothly run on embedded platforms, including mobile robots and UAVs, at a real-time speed of 36.4 fps. Conclusions The proposed SFC-RT framework is shown to efficiently and accurately address the challenges of target loss and occlusion for long-term tracking within complex smart factory environments. It meets the requirements of real-time performance, robustness, and lightweight design.

A document by Kunal Chakma. Click on the document to view its contents.

In the global monitoring of smart cities, the demands of global object detection systems based on cloud and fog computing in intelligent systems can be satisfied by photographs with globally recognized properties. Nevertheless, conventional techniques are constrained by the imaging depth of field and can produce artifacts or indistinct borders, which can be disastrous for accurately detecting the object. In light of this, this paper proposes an artificial intelligence-based gradient learning network that gathers and enhances domain information at different sizes in order to produce globally focused fusion results. Gradient features, which provide a lot of boundary information, can eliminate the problem of border artifacts and blur in multi-focus fusion. The multiple-receptive module (MRM) facilitates effective information sharing and enables the capture of object properties at different scales. In addition, with the assistance of the global enhancement module (GEM), the network can effectively combine the scale features and gradient data from various receptive fields and reinforce the features to provide precise decision maps. Numerous experiments have demonstrated that our approach outperforms the seven most sophisticated algorithms currently in use.

In the realm of traditional representation learning for Natural Language Processing (NLP) tasks, the prevalent approach involves leveraging large language models like BERT, GPT-2, XLNet, and others, which demonstrate their prowess when trained on extensive datasets. However, the quest for effective representation learning becomes a formidable challenge when data is scarce or costly to acquire, such as in specialised domains and low-resource languages. This challenge stems from the scarcity of suitable algorithms and the inherent difficulties in learning representations from limited data. In this study, we introduce a novel and groundbreaking technique, the Attention-sequence Bidirectional Long Short-Term Memory (ABiLSTM), to tackle this challenge head-on. This method amalgamates the strengths of attention mechanisms and bidirectional Long Short-Term Memory (LSTM) networks, bolstering the model’s ability to capture intricate patterns and dependencies within sequential data, particularly in scenarios involving small or constrained datasets. Through empirical analysis, we delve into the intricate dynamics of neural network architectures, specifically examining the impact of varying numbers of hidden layers and the occurrence of training errors. These aspects are scrutinised in the context of unsupervised pretraining methods, with a focus on their effectiveness in generating robust and informative representations for open relation extraction tasks. Furthermore, we propose the use of knowledge graph techniques to tackle the challenge of imbalanced data during information extraction processes. Our study demonstrates that the use of ABiLSTM yields superior results in open relation extraction tasks, even when dealing with small or limited datasets. This advancement represents a significant contribution to the fields of natural language understanding and information extraction.

Predictions arising from deep neural networks may be very accurate but not very robust, leading to uncertainty in their outcome. This critical problem is receiving growing attention from the Machine Learning (ML) community. A practical solution that is increasingly applied is calculating confidence bounds for the ML predictions. Most confidence bounds available in the literature are theoretically sound but unfeasible from a practical viewpoint. In this paper, we contribute to the literature with probabilistic confidence bounds based on conditional probabilities, and we demonstrate their operational validity by means of a real-world application that concerns the prediction of the sleeping states of car drivers.

In the environmental security monitoring of smart cities, the infrared and visible image fusion method deployed on intelligent systems based on cloud and fog computing plays an vital role in providing enhanced images for target detection systems. However, the fusion quality can be significantly influenced by the illumination of the monitoring scenario in visible images. Therefore, conventional methods typically suffer a severe performance drop under the condition of insufficient illumination. To tackle this issue, we propose an illumination enhancement and target-aware fusion method (IETAFusion) based on artificial intelligence, which breaks the boundaries between the task of illumination enhancement and image fusion and provide a fusion result with better visual perception in nighttime scene. Specifically, we use a light-weight contrast enhancement module (CEM) restore the brightness of the visble image. Moreover, a Swin Transformer-based backbone network (STBNet) is utilized to facilitate information exchangement between the source images and enhance the capabilities of target awareness. Finally, the fused images are reconstructed by the contrast-texture retention module (CTRM) and reconstructor. The extensive experiments indicates that the proposed approach achieves improved performance both in human perception and quantitative analysis compared with the state-of-the-art (SOTA) methods.

Voice Liveness Detection (VLD) has become one of the hot research topics in the Internet of Things era. Conventional VLD methods are centralized solutions that are trained on abundant data collected from local clients and stored on the server, however, they have the risk of causing data islands and privacy leakage. To address this problem, we propose a novel word-level VLD framework based on asynchronous federated learning (FL) with pop noise, named FL-VLD. Structurally, FL-VLD takes the preprocessed voice for local model training and constructs a global model by only transmitting the learned weights after differential privacy with FL’s central server in an asynchronous manner. In addition, the local network of the framework incorporates the residual network and the spatial grouping enhancement module to optimize the complexity and accuracy of the global model. With the advantage of FL’s distributed structure, FL-VLD solves the data island problem in VLD scenario without threatening users’ privacy. Experimental results on the popular POCO dataset show that our proposal is clearly superior to the traditional centralized methods as well as overperforming other federated schemes in terms of fairness, stability, accuracy, and lightweightness. Further, for attacks involving far-field replay, synthesis, and conversion, FL-VLD has high generalisation capabilities. Finally, the ablation study attests to its efficacy.

The expansion of machine learning applications across dif- ferent domains has given rise to a growing interest in tap- ping into the vast reserves of data that are being generated by edge devices. To preserve data privacy, federated learn- ing was developed as a collaboratively decentralized privacy- preserving technology to overcome the challenges of data silos and data sensibility. This technology faces certain lim- itations due to the limited network connectivity of mobile devices and malicious attackers. In addition, data samples across all devices are typically not independent and iden- tically distributed (non-IID), which presents additional chal- lenges to achieving convergence in fewer communication rounds. In this paper, we have simulated attacks, namely Byzantine, label flipping, and noisy data attacks, besides non-IID data. We proposed Robust federated learning against poisoning attacks (RFCaps) to increase safety and accelerate conver- gence. RFCaps incorporates a prediction-based clustering and a gradient quality evaluation method to prevent attack- ers from the aggregation phase by applying multiple filters and also accelerate convergence by using the highest quality gradients. In comparison to MKRUM, COMED, TMean, and FedAvg algorithms, RFCap has high robustness in the pres- ence of attackers and has achieved a higher accuracy of up to 80% on both MNIST and Fashion-MNIST datasets.

Cardiovascular disease (CVD) believes to be a major cause of transience and indisposition worldwide. Early diagnosis and timely intervention are critical in preventing the progression of CVD and improving patient outcomes. Machine learning (ML) algorithms have emerged as powerful tools in CVD recognition, with the potential to assist physicians in making accurate and efficient diagnoses. This research paper explores the combination of multiple ML algorithms for CVD recognition, utilizing diverse datasets such as the Cleveland, Hungarian, Switzerland, statlog, and VA Long Beach datasets. Additionally, a CVD dataset comprising 12 attributes and 70,000 records is employed, demonstrating improved results through the proposed and trained model compared to previous prediction techniques for CVD. The performance of various ML techniques, including support vector machines (SVM), Naive Bayes (NB), K-Nearest Neighbor (KNN), Random Forest (RF), and Logistic Regression (LR), is evaluated and compared. The impact of feature selection and feature scaling on the models’ performance is also examined. An ensemble bagging techniques is applied which is being embedded with other classifiers. LR classifier embedded with bagging techniques proved to be our proposed model. The findings reveal that the proposed Hybrid Linear Regression Bagging Model (HLRBM) outperforms other models. Furthermore, the study highlights the significance of data preprocessing techniques, such as data normalization and class balancing, which significantly enhance the performance of all models. To this end, standard scalar and Synthetic Minority Over-sampling Technique (SMOTE) are employed. The study emphasizes the importance of selecting an appropriate ensemble technique in conjunction with various ML algorithms and preprocessing methods for CVD prediction. Overall, the research provides valuable insights into the potential of ML in improving CVD risk assessment.

Anemia occurs when the hemoglobin (Hgb) value falls below a certain reference range. It requires many blood tests, radiological images, and tests for diagnosis and treatment. By processing medical data from patients with artificial intelligence and machine learning methods, disease predictions can be made for newly ill individuals and decision-support mechanisms can be created for physicians with these predictions. Thanks to these methods, which are very important in reducing the margin of error in the diagnoses made by doctors, the evaluation of data records in health institutions is also important for patients and hospitals. In this study, three hybrid models are proposed to classify non-anemia records, Hgb-anemia, folate deficiency anemia (FDA), iron deficiency anemia (IDA), and B12 deficiency anemia by combining artificial intelligence and machine learning methods TreeBagger with Crow Search Algorithm (CSO), Chicken Swarm Optimization Algorithm (CSO) and JAYA methods. The proposed hybrid models aim to achieve high performance by better emphasizing the importance of parameters. To solve the multiclass anemia classification problem, fuzzy logic-based parameter optimization is applied to improve the class-based accuracy as well as the overall accuracy in the dataset. The classification performances of the proposed methods are evaluated using accuracy, F-score, precision, and recall criteria to build a prediction model to identify the anemia type of anemic patients. A result of the study on the dataset taken from the Kaggle database found that the three proposed hybrid methods outperformed other studies using the same dataset and similar studies in the literature.

Binary combinatorial optimization plays a crucial role in various scientific and engineering fields. While deterministic approaches have traditionally been used to solve these problems, stochastic methods, particularly metaheuristics, have gained popularity in recent years for efficiently handling large problem instances. Ant Colony Optimization (ACO) is among the most successful metaheuristics and is frequently employed in non-binary combinatorial problems due to its adaptability. Although for binary combinatorial problems ACO can suffer from issues such as rapid convergence to local minima, its eminently parallel structure means that it can be exploited to solve large and complex problems also in this field. In order to provide a versatile ACO implementation that achieves competitive results across a wide array of binary combinatorial optimization problems, we introduce a parallel multicolony strategy with an improved cooperation scheme to maintain search diversity. We evaluate our proposal (Binary Parallel Cooperative ACO, BiPCACO) using a comprehensive benchmark framework, showcasing its performance and, most importantly, its flexibility as a successful all-purpose solver for binary combinatorial problems.