Improvement In F1 Score Research Articles

Deep learning resilience inference for complex networked systems

Resilience, the ability to maintain fundamental functionality amidst failures and errors, is crucial for complex networked systems. Most analytical approaches rely on predefined equations for node activity dynamics and simplifying assumptions on network topology, limiting their applicability to real-world systems. Here, we propose ResInf, a deep learning framework integrating transformers and graph neural networks to infer resilience directly from observational data. ResInf learns representations of node activity dynamics and network topology without simplifying assumptions, enabling accurate resilience inference and low-dimensional visualization. Experimental results show that ResInf significantly outperforms analytical methods, with an F1-score improvement of up to 41.59% over Gao-Barzel-Barabási framework and 14.32% over spectral dimension reduction. It also generalizes to unseen topologies and dynamics and maintains robust performance despite observational disturbances. Our findings suggest that ResInf addresses an important gap in resilience inference for real-world systems, offering a fresh perspective on incorporating data-driven approaches to complex network modeling.

A novel image classification framework based on variational quantum algorithms

Image classification is a crucial task in machine learning with widespread practical applications. The existing classical framework for image classification typically utilizes a global pooling operation at the end of the network to reduce computational complexity and mitigate overfitting. However, this operation often results in a significant loss of information, which can affect the performance of classification models. To overcome this limitation, we introduce a novel image classification framework that leverages variational quantum algorithms (VQAs) hybrid approaches combining quantum and classical computing paradigms within quantum machine learning. The major advantage of our framework is the elimination of the need for the global pooling operation at the end of the network. In this way, our approach preserves more discriminative features and fine-grained details in the images, which enhances classification performance. Additionally, employing VQAs enables our framework to have fewer parameters than the classical framework, even in the absence of global pooling, which makes it more advantageous in preventing overfitting. We apply our method to different state-of-the-art image classification models and demonstrate the superiority of the proposed quantum architecture over its classical counterpart through a series of state vector simulation experiments on public datasets. Our experiments show that the proposed quantum framework achieves up to a 9.21% increase in accuracy and up to a 15.79% improvement in F1 score, compared to the classical framework. Additionally, we explore the impact of shot noise on our method through shot-based simulation and find that increasing the number of measurements does not always lead to better results. Selecting an appropriate number of measurements can yield optimal results, even surpassing those obtained from state vector simulation.

Large language model-based optical network log analysis using LLaMA2 with instruction tuning

The optical network encompasses numerous devices and links, generating a significant volume of logs. Analyzing these logs is significant for network optimization, failure diagnosis, and health monitoring. However, the large-scale and diverse formats of optical network logs present several challenges, including the high cost and difficulty of manual processing, insufficient semantic understanding in existing analysis methods, and the strict requirements for data security and privacy. Generative artificial intelligence (GAI) with powerful language understanding and generation capabilities has the potential to address these challenges. Large language models (LLMs) as a concrete realization of GAI are well-suited for analyzing DCI logs, replacing human experts and enhancing accuracy. Additionally, LLMs enable intelligent interactions with network administrators, automating tasks and improving operational efficiency. Moreover, fine-tuning with open-source LLMs protects data privacy and enhances log analysis accuracy. Therefore, we introduce LLMs and propose a log analysis method with instruction tuning using LLaMA2 for log parsing, anomaly detection and classification, anomaly analysis, and report generation. Real log data extracted from the field-deployed network was used to design and construct instruction tuning datasets. We utilized the dataset for instruction tuning and demonstrated and evaluated the effectiveness of the proposed scheme. The results indicate that this scheme improves the performance of log analysis tasks, especially a 14% improvement in exact match rate for log parsing, a 13% improvement in F1-score for anomaly detection and classification, and a 23% improvement in usability for anomaly analysis, compared with the best baselines.

Data Augmentation Method Using Room Transfer Function for Monitoring of Domestic Activities

Monitoring domestic activities helps us to understand user behaviors in indoor environments, which has garnered interest as it aids in understanding human activities in context-aware computing. In the field of acoustics, this goal has been achieved through studies employing machine learning techniques, which are widely used for classification tasks involving sound recognition and other objectives. Machine learning typically achieves better performance with large amounts of high-quality training data. Given the high cost of data collection, development datasets often suffer from imbalanced data or lack high-quality samples, leading to performance degradations in machine learning models. The present study aims to address this data issue through data augmentation techniques. Specifically, since the proposed method targets indoor activities in domestic activity detection, room transfer functions were used for data augmentation. The results show that the proposed method achieves a 0.59% improvement in the F1-Score (micro) from that of the baseline system for the development dataset. Additionally, test data including microphones that were not used during training achieved an F1-Score improvement of 0.78% over that of the baseline system. This demonstrates the enhanced model generalization performance of the proposed method on samples having different room transfer functions to those of the trained dataset.

Improving the industrial defect recognition in radiographic testing by pre-training on medical radiographs

Deep learning methodologies have gained substantial traction for defect recognition in industrial radiographic testing including welds, castings and other fields. Regardless of the deep learning utilized, it has emerged as a standard configuration to use a model pre-trained from ImageNet to accelerate convergence and enhance recognition accuracy. However, there is a significant gap between the domain of natural images and industrial radiographs, raising the question of whether there might be a superior pre-training method than relying on ImageNet pre-training. Fortunately, medical radiographs are more similar to industrial radiographs than natural images because of the same imaging method. In this paper, we initially utilize numerous medical radiographic images from CheXpert dataset to train a pre-trained CNN model. Then, we apply this model to four distinct tasks within two radiographic testing scenarios to validate its advantages and generalization capabilities. Finally, experiments on multiple datasets indicate that our method brings more benefits than ImageNet pre-training or training from scratch, with a F1 score improvement of 3.41 %–13.72 % for defect classification and a mIoU improvement of 1.05 %–6.58 % for defect segmentation. It demonstrates that pre-training from medical radiographs provides a cost-free improvement for all kinds of tasks in industrial defect recognition.

Influence of Structure from Motion Algorithm Parameters on Metrics for Individual Tree Detection Accuracy and Precision

Uncrewed aerial system (UAS) structure from motion (SfM) monitoring strategies for individual trees has rapidly expanded in the early 21st century. It has become common for studies to report accuracies for individual tree heights and DBH, along with stand density metrics. This study evaluates individual tree detection and stand basal area accuracy and precision in five ponderosa pine sites against the range of SfM parameters in the Agisoft Metashape, Pix4DMapper, and OpenDroneMap algorithms. The study is designed to frame UAS-SfM individual tree monitoring accuracy in the context of data processing and storage demands as a function of SfM algorithm parameter levels. Results show that when SfM algorithms are properly tuned, differences between software types are negligible, with Metashape providing a median F-score improvement over OpenDroneMap of 0.02 and PIX4DMapper of 0.06. However, tree extraction performance varied greatly across algorithm parameters, with the greatest extraction rates typically coming from parameters causing increased density in dense point clouds and minimal point cloud filtering. Transferring UAS-SfM forest monitoring into management will require tradeoffs between accuracy and efficiency. Our analysis shows that a one-step reduction in dense point cloud quality saves 77–86% in point cloud processing time without decreasing tree extraction (F-score) or basal area precision using Metashape and PIX4DMapper but the same parameter change for OpenDroneMap caused a ~5% loss in precision. Providing reproducible processing strategies is a vital step in successfully transferring these technologies into usage as management tools.

Detection and isolation of brain tumors in cancer patients using neural network techniques in MRI images

MRI imaging primarily focuses on the soft tissues of the human body, typically performed prior to a patient's transfer to the surgical suite for a medical procedure. However, utilizing MRI images for tumor diagnosis is a time-consuming process. To address these challenges, a new method for automatic brain tumor diagnosis was developed, employing a combination of image segmentation, feature extraction, and classification techniques to isolate the specific region of interest in an MRI image corresponding to a brain tumor. The proposed method in this study comprises five distinct steps. Firstly, image pre-processing is conducted, utilizing various filters to enhance image quality. Subsequently, image thresholding is applied to facilitate segmentation. Following segmentation, feature extraction is performed, analyzing morphological and structural properties of the images. Then, feature selection is carried out using principal component analysis (PCA). Finally, classification is performed using an artificial neural network (ANN). In total, 74 unique features were extracted from each image, resulting in a dataset of 144 observations. Principal component analysis was employed to select the top 8 most effective features. Artificial Neural Networks (ANNs) leverage comprehensive data and selective knowledge. Consequently, the proposed approach was evaluated and compared with alternative methods, resulting in significant improvements in precision, accuracy, and F1 score. The proposed method demonstrated notable increases in accuracy, with improvements of 99.3%, 97.3%, and 98.5% in accuracy, Sensitivity and F1 score. These findings highlight the efficiency of this approach in accurately segmenting and classifying MRI images.

Augmenting biomedical named entity recognition with general-domain resources

ObjectiveTraining a neural network-based biomedical named entity recognition (BioNER) model usually requires extensive and costly human annotations. While several studies have employed multi-task learning with multiple BioNER datasets to reduce human effort, this approach does not consistently yield performance improvements and may introduce label ambiguity in different biomedical corpora. We aim to tackle those challenges through transfer learning from easily accessible resources with fewer concept overlaps with biomedical datasets. MethodsWe proposed GERBERA, a simple-yet-effective method that utilized general-domain NER datasets for training. We performed multi-task learning to train a pre-trained biomedical language model with both the target BioNER dataset and the general-domain dataset. Subsequently, we fine-tuned the models specifically for the BioNER dataset. ResultsWe systematically evaluated GERBERA on five datasets of eight entity types, collectively consisting of 81,410 instances. Despite using fewer biomedical resources, our models demonstrated superior performance compared to baseline models trained with additional BioNER datasets. Specifically, our models consistently outperformed the baseline models in six out of eight entity types, achieving an average improvement of 0.9% over the best baseline performance across eight entities. Our method was especially effective in amplifying performance on BioNER datasets characterized by limited data, with a 4.7% improvement in F1 scores on the JNLPBA-RNA dataset. ConclusionThis study introduces a new training method that leverages cost-effective general-domain NER datasets to augment BioNER models. This approach significantly improves BioNER model performance, making it a valuable asset for scenarios with scarce or costly biomedical datasets. We make data, codes, and models publicly available via https://github.com/qingyu-qc/bioner_gerbera.

SF-GPT: A training-free method to enhance capabilities for knowledge graph construction in LLMs

Knowledge graphs (KGs) are constructed by extracting knowledge triples from text and fusing knowledge, enhancing information retrieval efficiency. Current methods for knowledge triple extraction include ”Pretrain and Fine-tuning” and Large Language Models (LLMs). The former shifts effort from manual extraction to dataset annotation and suffers from performance degradation with different test and training set distributions. LLMs-based methods face errors and incompleteness in extraction. We introduce SF-GPT, a training-free method to address these issues. Firstly, we propose the Entity Extraction Filter (EEF) module to filter triple generation results, addressing evaluation and cleansing challenges. Secondly, we introduce a training-free Entity Alignment Module based on Entity Alias Generation (EAG), tackling semantic richness and interpretability issues in LLM-based knowledge fusion. Finally, our Self-Fusion Subgraph strategy uses multi-response self-fusion and a common entity list to filter triple results, reducing noise from LLMs’ multi-responses. In experiments, SF-GPT showed a 55.5% increase in recall and a 32.6% increase in F1 score on the BDNC dataset compared to the UniRel model trained on the NYT dataset and achieved a 5% improvement in F1 score compared to GPT-4+EEF baseline on the WebNLG dataset in the case of a fusion round of three. SF-GPT offers a promising way to extract knowledge from unstructured information.

Fusion model of gray level co-occurrence matrix and convolutional neural network faced for histopathological images.

The image recognition of cancer cells plays an important role in diagnosing and treating cancer. Deep learning is suitable for classifying histopathological images and providing auxiliary technology for cancer diagnosis. The convolutional neural network is employed in the classification of histopathological images; however, the model's accuracy may decrease along with the increase in network layers. Extracting appropriate image features is helpful for image classification. In this paper, different features of histopathological images are represented by extracting features of the gray co-occurrence matrix. These features are recombined into a 16 × 16 × 3 matrix to form an artificial image. The original image and the artificial image are fused by summing the softmax output. The histopathological images are divided into the training set, validation set, and testing set. Each training dataset consists of 1500 images, while the validation dataset and test dataset each consist of 500 images. The results indicate that the effectiveness of our fusion model is demonstrated through significant improvements in accuracy, precision, recall, and F1-score, with an average accuracy reaching 99.31%. This approach not only enhances the classification performance of tissue pathology images but also holds promise for advancing computer-aided diagnosis in cancer pathology.

Machine learning-based layer-wise detection of overheating anomaly in LPBF using photodiode data

Overheating anomaly detection is essential for the quality and reliability of parts produced by laser powder bed fusion (LPBF) additive manufacturing (AM). In this research, we focus on the detection of overheating anomalies which can lead to various defects in the part including geometric distortion, and poor surface roughness, among others, using photodiode sensor data. Photodiode sensors can collect high-frequency data from the melt pool, reflecting the process dynamics and thermal history. Hence, the proposed method offers a machine learning (ML) framework to utilize photodiode sensor data for layer-wise detection of overheating anomalies. In doing so, three sets of features are extracted from the raw photodiode data: MSMM (mean, standard deviation, median, maximum), MSQ (mean, standard deviation, quartiles), and MSD (mean, standard deviation, deciles). These three datasets are used to train several ML classifiers. Cost-sensitive learning is used to handle the class imbalance between the “anomalous” layers (affected by overheating) and “nominal” layers in the benchmark dataset. To boost detection accuracy, our proposed ML framework involves utilizing the majority voting ensemble (MVE) approach. First, the top three ML classifiers are identified from an initial pool of classifiers, based on their performance in k-fold cross-validation. Next, final predictions are generated using majority voting from the individual predictions of the top three classifiers. We performed 100 iterations to generate statistically reliable results. This proposed method is demonstrated using a case study including an open benchmark dataset of photodiode measurements from an LPBF specimen with deliberate overheating anomalies at some layers. The results from the case study demonstrate that the MSD features yield the best performance for all classifiers, and the MVE classifier (with a mean F1-score of 0.8654) surpasses the individual ML classifiers. Moreover, our machine learning methodology achieves superior results (9.66% improvement in mean F1-score) in detecting layer-wise overheating anomalies, surpassing the existing methods in the literature that use the same benchmark dataset. Finally, based on our results, we provide useful insights and recommendations for future research on applying machine learning techniques for defect detection in AM.

Improving Log Anomaly Detection via Spatial Pooling: Combining SPClassifier with Ensemble Method

In the ever-updating field of software development, new bugs emerge daily, requiring significant time for analysis. As a result, research is being conducted on automating bug resolution using techniques such as anomaly detection through deep learning applied to text logs. This study focuses on anomaly detection using text logs and aims to address current challenges. Specifically, we aim to improve the accuracy of SPClassifier, a robust and lightweight AI model capable of handling dynamic log datasets through ad-hoc learning. We employ three ensemble learning methods to enhance the accuracy of SPClassifier. The method that achieved the greatest improvement was Improved Bagging, which combines the non-overlapping sampling of Pasting with the overlapping sampling of Bagging, resulting in a maximum F1-score improvement of 155%. Additionally, on certain datasets, the F1-score surpassed that of well-known DNN methods by 130%. Furthermore, the proposed method demonstrated lower variance compared to DNN methods, indicating its advantage, particularly in environments where datasets frequently fluctuate, such as development fields. These results highlight the clear superiority of the proposed method, which is lightweight in terms of computational resources and supports ad-hoc learning.

A multi-strategy ontology mapping method based on cost-sensitive SVM

As the core of ontology integration, the task of ontology mapping is to find the semantic relationship between ontologies. Nevertheless, most existing ontology mapping methods only rely ontext information to calculate entity similarity, thereby disregarding semantic nuances and necessitating substantial manual intervention. Therefore, this paper introduces an ontology mapping method. Based on the traditional ontology mapping method, the process employs a deep learning model to mine the semantic information of entity concepts, entity properties and ontology structure to obtain the embedding vector. We use the similarity mechanism to calculate the similarity between different embedding vectors, and combine the similarity values obtained from multiple strategy entities into a similarity matrix. The similarity matrix serves as input to the support vector machine (SVM), and the ontology mapping problem is finally transformed into a binary classification problem. However, since the number of mapped pairs is much larger than the number of non-mapped pairs, the number of positive samples in the data set is much smaller than the number of negative samples. Therefore, based on the traditional SVM, the paper adopts cost-sensitive strategy to deal with the class imbalance problem. In comparative evaluations against contemporary ontology mapping techniques, our method demonstrates a noteworthy 5.0% enhancement in recall and a 3.0% improvement in F1 score when tested on both public benchmark datasets and domain-specific datasets.

Interactive computer-aided diagnosis on medical image using large language models

Computer-aided diagnosis (CAD) has advanced medical image analysis, while large language models (LLMs) have shown potential in clinical applications. However, LLMs struggle to interpret medical images, which are critical for decision-making. Here we show a strategy integrating LLMs with CAD networks. The framework uses LLMs’ medical knowledge and reasoning to enhance CAD network outputs, such as diagnosis, lesion segmentation, and report generation, by summarizing information in natural language. The generated reports are of higher quality and can improve the performance of vision-based CAD models. In chest X-rays, an LLM using ChatGPT improved diagnosis performance by 16.42 percentage points compared to state-of-the-art models, while GPT-3 provided a 15.00 percentage point F1-score improvement. Our strategy allows accurate report generation and creates a patient-friendly interactive system, unlike conventional CAD systems only understood by professionals. This approach has the potential to revolutionize clinical decision-making and patient communication.

Fine-grained vulnerability detection for medical sensor systems

The Internet of Things (IoT) has revolutionized the healthcare system by connecting medical sensors to the internet, while also posing challenges to the security of medical sensor networks (MSN). Given the extreme sensitivity of medical data, any vulnerability may result in data breaches and misuse, impacting patient safety and privacy. Therefore, safeguarding MSN security is critical. As medical sensor devices rely on smart healthcare software systems for data management and communication, precisely detecting system code vulnerabilities is essential to ensuring network security. Effective software vulnerability detection targets two key objectives: (i) achieving high accuracy and (ii) directly identifying vulnerable code lines for developers to fix. To address these challenges, we introduce Vulcoder, a novel vulnerability-oriented, encoder-driven model based on the Bidirectional Encoder Representations from Transformers (BERT) architecture. We propose a one-to-one mapping function to capture code semantics through abstract syntax trees (AST). Combined with multi-head attention, Vulcoder achieves precise function- and line-level detection of software vulnerabilities in MSN. This accelerates the vulnerability remediation process, thereby strengthening network security. Experimental results on various datasets demonstrate that Vulcoder outperforms previous models in identifying vulnerabilities within MSN. Specifically, it achieves a 1%–419% improvement in function-level prediction F1 scores and a 12.5%–380% increase in line-level localization precision. Therefore, Vulcoder helps enhance security defenses and safeguard patient privacy in MSN, facilitating the development of smart healthcare.

A diffusion model multi-scale feature fusion network for imbalanced medical image classification research

Background and objectiveMedicine image classification are important methods of traditional medical image analysis, but the trainable data in medical image classification is highly imbalanced and the accuracy of medical image classification models is low. In view of the above two common problems in medical image classification. This study aims to: (i) effectively solve the problem of poor training effect caused by the imbalance of class imbalanced data sets. (ii) propose a network framework suitable for improving medical image classification results, which needs to be superior to existing methods. MethodsIn this paper, we put in the diffusion model multi-scale feature fusion network (DMSFF), which mainly uses the diffusion generation model to overcome imbalanced classes (DMOIC) on highly imbalanced medical image datasets. At the same time, it is processed according to the cropped image augmentation strategy through cropping (IASTC). Based on this, we use the new dataset to design a multi-scale feature fusion network (MSFF) that can fully utilize multiple hierarchical features. The DMSFF network can effectively solve the problems of small and imbalanced samples and low accuracy in medical image classification. ResultsWe evaluated the performance of the DMSFF network on highly imbalanced medical image classification datasets APTOS2019 and ISIC2018. Compared with other classification models, our proposed DMSFF network achieved significant improvements in classification accuracy and F1 score on two datasets, reaching 0.872, 0.731, and 0.906, 0.836, respectively. ConclusionsOur newly proposed DMSFF architecture outperforms existing methods on two datasets, and verifies the effectiveness of generative model inverse balance for imbalance class datasets and feature enhancement by multi-scale feature fusion. Further, the method can be applied to other class imbalanced data sets where the results will be improved.

Evaluating Synthetic Data Augmentation to Correct for Data Imbalance in Realistic Clinical Prediction Settings.

Predictive modeling holds a large potential in clinical decision-making, yet its effectiveness can be hindered by inherent data imbalances in clinical datasets. This study investigates the utility of synthetic data for improving the performance of predictive modeling on realistic small imbalanced clinical datasets. We compared various synthetic data generation methods including Generative Adversarial Networks, Normalizing Flows, and Variational Autoencoders to the standard baselines for correcting for class underrepresentation on four clinical datasets. Although results show improvement in F1 scores in some cases, even over multiple repetitions, we do not obtain statistically significant evidence that synthetic data generation outperforms standard baselines for correcting for class imbalance. This study challenges common beliefs about the efficacy of synthetic data for data augmentation and highlights the importance of evaluating new complex methods against simple baselines.

TransURL: Improving malicious URL detection with multi-layer Transformer encoding and multi-scale pyramid features

While machine learning progress is advancing the detection of malicious URLs, advanced Transformers applied to URLs face difficulties in extracting local information, character-level information, and structural relationships. To address these challenges, we propose a novel approach for malicious URL detection, named TransURL, that is implemented by co-training the character-aware Transformer with three feature modules—Multi-Layer Encoding, Multi-Scale Feature Learning, and Spatial Pyramid Attention. This special Transformer allows TransURL to extract embeddings that contain character-level information from URL token sequences, with three feature modules contributing to the fusion of multi-layer Transformer encodings and the capture of multi-scale local details and structural relationships. The proposed method is evaluated across several challenging scenarios, including class imbalance learning, multi-classification, cross-dataset testing, and adversarial sample attacks. The experimental results demonstrate a significant improvement compared to the best previous methods. For instance, it achieved a peak F1-score improvement of 40% in class-imbalanced scenarios, and exceeded the best baseline result by 14.13% in accuracy in adversarial attack scenarios. Additionally, we conduct a case study where our method accurately identifies all 30 active malicious web pages, whereas two pior SOTA methods miss 4 and 7 malicious web pages respectively. The codes and data are available at: https://github.com/Vul-det/TransURL/.

Integration of Relation Filtering and Multi-Task Learning in GlobalPointer for Entity and Relation Extraction

The rise of knowledge graphs has been instrumental in advancing artificial intelligence (AI) research. Extracting entity and relation triples from unstructured text is crucial for the construction of knowledge graphs. However, Chinese text has a complex grammatical structure, which may lead to the problem of overlapping entities. Previous pipeline models have struggled to address such overlap problems effectively, while joint models require entity annotations for each predefined relation in the set, which results in redundant relations. In addition, the traditional models often lead to task imbalance by overlooking the differences between tasks. To tackle these challenges, this research proposes a global pointer network based on relation prediction and loss function improvement (GPRL) for joint extraction of entities and relations. Experimental evaluations on the publicly available Chinese datasets DuIE2.0 and CMeIE demonstrate that the GPRL model achieves a 1.2–26.1% improvement in F1 score compared with baseline models. Further, experiments of overlapping classification conducted on CMeIE have also verified the effectiveness of overlapping triad extraction and ablation experiments. The model is helpful in identifying entities and relations accurately and can reduce redundancy by leveraging relation filtering and the global pointer network. In addition, the incorporation of a multi-task learning framework balances the loss functions of multiple tasks and enhances task interactions.

Federated Learning Using Multi-Modal Sensors with Heterogeneous Privacy Sensitivity Levels

Data from multi-modal sensors, such as RGB cameras, thermal cameras, microphones, and mmWave radars, have gradually been adopted in various classification problems for better accuracy. Some sensors, like RGB cameras and microphones, however, capture privacy-invasive data, which are less likely to be used in centralized learning. Although the Federated Learning (FL) paradigm frees clients from sharing their sensor data, doing so results in reduced classification accuracy and increased training time. In this article, we introduce a novel Heterogeneous Privacy Federated Learning (HPFL) paradigm to better capitalize on the less privacy-invasive sensor data, such as thermal images and mmWave point clouds, by uploading them to the server for closing the performance gap between FL and centralized learning. HPFL not only allows clients to keep the more privacy-invasive sensor data private, such as RGB images and human voices, but also gives each client total freedom to define the levels of their privacy concern on individual sensor modalities. For example, more sensitive users may prefer to keep their thermal images private, while others do not mind sharing these images. We carry out extensive experiments to evaluate the HPFL paradigm using two representative classification problems: semantic segmentation and emotion recognition. Several key findings demonstrate the merits of HPFL: (i) compared to FedAvg, it improves foreground accuracy by 18.20% in semantic segmentation and boosts the F1-score by 4.20% in emotion recognition, (ii) with heterogeneous privacy concern levels, it achieves an even larger F1-score improvement of 6.17–16.05% in emotion recognition, and (iii) it also outperforms the state-of-the-art FL approaches by 12.04–17.70% in foreground accuracy and 2.54–4.10% in F1-score.