Classification Task Research Articles

Facial Emotion Recognition of 16 Distinct Emotions From Smartphone Videos: Comparative Study of Machine Learning and Human Performance.

The development of automatic emotion recognition models from smartphone videos is a crucial step toward the dissemination of psychotherapeutic app interventions that encourage emotional expressions. Existing models focus mainly on the 6 basic emotions while neglecting other therapeutically relevant emotions. To support this research, we introduce the novel Stress Reduction Training Through the Recognition of Emotions Wizard-of-Oz (STREs WoZ) dataset, which contains facial videos of 16 distinct, therapeutically relevant emotions. This study aimed to develop deep learning-based automatic facial emotion recognition (FER) models for binary (positive vs negative) and multiclass emotion classification tasks, assess the models' performance, and validate them by comparing the models with human observers. The STREs WoZ dataset contains 14,412 facial videos of 63 individuals displaying the 16 emotions. The selfie-style videos were recorded during a stress reduction training using front-facing smartphone cameras in a nonconstrained laboratory setting. Automatic FER models using both appearance and deep-learned features for binary and multiclass emotion classification were trained on the STREs WoZ dataset. The appearance features were based on the Facial Action Coding System and extracted with OpenFace. The deep-learned features were obtained through a ResNet50 model. For our deep learning models, we used the appearance features, the deep-learned features, and their concatenation as inputs. We used 3 recurrent neural network (RNN)-based architectures: RNN-convolution, RNN-attention, and RNN-average networks. For validation, 3 human observers were also trained in binary and multiclass emotion recognition. A test set of 3018 facial emotion videos of the 16 emotions was completed by both the automatic FER model and human observers. The performance was assessed with unweighted average recall (UAR) and accuracy. Models using appearance features outperformed those using deep-learned features, as well as models combining both feature types in both tasks, with the attention network using appearance features emerging as the best-performing model. The attention network achieved a UAR of 92.9% in the binary classification task, and accuracy values ranged from 59.0% to 90.0% in the multiclass classification task. Human performance was comparable to that of the automatic FER model in the binary classification task, with a UAR of 91.0%, and superior in the multiclass classification task, with accuracy values ranging from 87.4% to 99.8%. Future studies are needed to enhance the performance of automatic FER models for practical use in psychotherapeutic apps. Nevertheless, this study represents an important first step toward advancing emotion-focused psychotherapeutic interventions via smartphone apps.

Leak localization in water distribution networks using fusion classifier techniques

ABSTRACT This article presents a model-based fusion classifier technique to estimate the approximate position of a leak in a water distribution network (WDN). This technique uses residuals obtained by comparing pressure measurements with the nominal behavior estimated from a mathematical model. These residuals are analyzed as a classification task to associate the information with an approximate leak location. The classification task is performed by three different classifiers: the K-Nearest Neighbors algorithm, the Multilayer Perceptron, and the Decision Tree classifier. The outputs of these classifiers are combined using two ensemble/fusion methods, the Majority Voting and the Naive Bayes fusion, to improve the accuracy of the leak position estimation. Results from a benchmark problem comprising 126 nodes, 8 valves, 2 tanks, 2 reservoirs, 3 pumps, and 168 pipes, subject to four variable demand patterns, illustrate the performance improvement of the two ensemble methods over an individual classifier. Additionally, to create a more realistic scenario, the performance of the proposed leak localization scheme is evaluated in a perturbed scenario by noise and uncertainties in the demand estimation.

On the similarity of bandwidth-tuned quantum kernels and classical kernels

Abstract Quantum kernels (QK) are widely used in quantum machine learning applications; yet, their potential to surpass classical machine learning methods on classical datasets remains uncertain. This limitation can be attributed to the exponential concentration phenomenon, which can impair generalization. A common strategy to alleviate this is bandwidth tuning, which involves rescaling data points in the quantum model to improve generalization. In this work, we numerically demonstrate that optimal bandwidth tuning results in QKs that closely resemble radial basis function (RBF) kernels, leading to a lack of quantum advantage over classical methods. Moreover, we reveal that the size of optimal bandwidth tuning parameters further simplifies QKs, causing them to behave like polynomial kernels, corresponding to a low-order Taylor approximation of a RBF kernel. We thoroughly investigate this for fidelity quantum kernels and projected quantum kernels using various data encoding circuits across several classification datasets. We provide numerical evidence and derive a simple analytical model that elucidates how bandwidth tuning influences key quantities in classification tasks. Overall, our findings shed light on the mechanisms that render QK methods classically tractable.

Using Of Deep Learning Models In Acoustic Scene Classification

Ambient sound analysis has become more prominent with the rise of portable and wearable devices. It provides valuable insights into a person's environment by analyzing surrounding sounds. Recently, deep learning methods, frequently used in image and text processing, have been applied to this field and are proving more effective than traditional machine learning techniques. In this study, we evaluated the performance of different deep learning models using mel-spectrograms of 3 classes of stage sounds based on TAU Acoustic Scene 2023 dataset. Our results indicate that a simple Convolutional Neural Network (CNN) model gives better classification results compared to other more complex models in classification tasks. Despite having the fewest parameters, the CNN model achieved the highest success with 59% accuracy. This suggests that simpler models can be highly effective for acoustic scene classification, highlighting the value of more efficient and computationally feasible approaches in this domain

A two-stage EEG zero-shot classification algorithm guided by class reconstruction.

Researchers have long been dedicated to decoding human visual representations from neural signals. These studies are crucial in uncovering the mechanisms of visual processing in the human brain. Electroencephalogram(EEG) signals have garnered widespread attention recently due to their noninvasive nature and low cost. EEG classification is one of the most popular topics in brain-computer interface(BCI) research. However, most traditional EEG classification algorithms are difficult to generalize to unseen classes that were not involved in the training phase. The main objective of this work is to improve the performance of these EEG classification algorithms for unseen classes. In this work, we propose a two-stage zero-shot EEG classification algorithm guided by class reconstruction. The method is specifically designed with a two-stage training strategy based on class reconstruction. This structure and training strategy enable the model to thoroughly learn the relations and distinctions among EEG embeddings of different classes. The Contrastive Language-Image Pre-training(CLIP) model has a well-aligned latent space and powerful cross-modality generalization ability. The method bridges the modality gap between EEG, images, and text using CLIP features. It significantly improves the model's performance in unseen classes. We conducted experiments on the ImageStimulus-EEG dataset to evaluate the performance of the proposed method. Meanwhile, it was compared with the state-of-the-art model and the baseline model. The experimental results demonstrate that our model achieves superior performance in among Top-1, Top-3, and Top-5 classification accuracy for a 50-way zero-shot classification task, reaching 17.77%, 38.76% and 54.75%, respectively. These results further validate the effectiveness of the proposed method in EEG zero-shot classification.

Collaborative multitasking framework for enhanced hippocampus segmentation and Alzheimer's disease classification.

Collaborative multitasking framework for enhanced hippocampus segmentation and Alzheimer's disease classification.

A vision transformer-convolutional neural network framework for decision-transparent dual-energy X-ray absorptiometry recommendations using chest low-dose CT.

A vision transformer-convolutional neural network framework for decision-transparent dual-energy X-ray absorptiometry recommendations using chest low-dose CT.

Unraveling the neural dynamics of mathematical interference in english reading: A novel approach with deep learning and fNIRS data.

Unraveling the neural dynamics of mathematical interference in english reading: A novel approach with deep learning and fNIRS data.

Congestion Forecasting Using Machine Learning Techniques: A Systematic Review

Traffic congestion constitutes a substantial global issue, adversely impacting economic productivity and quality of life, with associated costs estimated at approximately 2% of GDP in various nations. This systematic review investigates the application of machine learning (ML) in traffic congestion forecasting from 2010 to 2024, adhering to the PRISMA 2020 guidelines. A comprehensive search of three major databases (IEEE Xplore, SpringerLink, and ScienceDirect) yielded 9695 initial records, with 115 studies meeting the inclusion criteria following rigorous screening. Data extraction encompassed methodological approaches, ML techniques, traffic characteristics, and forecasting periods, with quality assessment achieving near-perfect inter-rater reliability (Cohen’s κ = 0.89). Deep Neural Networks were the predominant technical approach (47%), with supervised learning being the most prevalent (57%). Classification tasks were the most common (42%), primarily addressing recurrent congestion scenarios (76%) and passenger vehicles (90%). The quality of publications was notably high, with 85% appearing in Q1-ranked journals, demonstrating exponential growth from minimal activity in 2010 to 18 studies in 2022. Significant research gaps persist: reinforcement learning is underutilized (8%), rural road networks are underrepresented (2%), and industry–academia collaboration is limited (3%). Future research should prioritize multimodal transportation systems, real-time adaptation mechanisms, and enhanced practical implementation to advance intelligent transportation systems (ITSs). This review was not registered because it focused on mapping the research landscape rather than intervention effects.

Evaluation of the effectiveness of artificial intelligence models in radiopaque and radiolucent lesions of the maxillofacial region on panoramic radiographs.

The aim of this study is to evaluate the success of algorithms used in deep learning (DL), a technique of artificial intelligence (AI), in the classification, detection, and segmentation of radiopaque, and radiolucent lesions in the maxillofacial region on panoramic radiographs (PR). This study included PRs of individuals aged 12 to 80 years who presented with radiopaque or radiolucent findings in the maxillofacial region based on radiological examination. Lesions were classified on the dataset obtained from the PRs using AlexNet, VGG16, and GoogleNet architectures. The location detection and segmentation of lesions were performed using the YOLOv8 architecture. The classification, object detection, and segmentation performances of the DL architectures were evaluated. In the classification tasks using full PR, GoogleNet achieved the highest accuracy of 95.6%, with 97.1% precision and 95.5% F1 score in two-class lesion classification (lesion vs. no lesion). In distinguishing radiopaque and radiolucent lesions, VGG16 performed best, with 68.4% accuracy and 81.0% F1 score. For three-class and four-class classifications, GoogleNet again outperformed others with 61.6 and 75.7% accuracy, respectively. In cropped lesion-based classification, both GoogleNet and AlexNet achieved 96.5% accuracy. The YOLOv8m model demonstrated the best performance in object detection and segmentation, with 71.5% and 72.1% mean Average Precision (mAP), respectively. These findings suggest that DL architectures, particularly GoogleNet for classification and YOLOv8m for object detection and segmentation, demonstrate strong potential in the automated analysis of maxillofacial lesions on panoramic radiographs. Their high performance in distinguishing lesion types and accurately localizing pathological areas indicates that such models could assist clinicians in early diagnosis and treatment planning, potentially reducing reliance on more complex imaging methods.

Comparing three neural networks to predict depression treatment outcomes in psychological therapies.

Artificial neural networks have been used in various fields to solve classification and prediction tasks. However, it is unclear if these may be adequate methods to predict psychological treatment outcomes. This study aimed to evaluate the prognostic accuracy of neural networks using psychological treatment outcomes data. Three neural network models (TensorFlow, nnet, and monmlp) and a generalised linear regression model were compared in their ability to predict post-treatment remission of depression symptoms in a large naturalistic sample (n=69,489) of patients accessing low intensity cognitive behavioural therapy. Prognostic accuracy was evaluated using the area under the curve (AUC) in an external cross-validation design. The AUC of the neural networks in an external test sample ranged from 0.64 to 0.65 and the AUC of the linear regression model was 0.63. Neural networks can help predict symptom remission in new samples with moderate accuracy, although these models were no more accurate than a simpler inferential statistical linear regression model.

DRTN: Dual Relation Transformer Network with feature erasure and contrastive learning for multi-label image classification.

DRTN: Dual Relation Transformer Network with feature erasure and contrastive learning for multi-label image classification.

Adaptive node-level weighted learning for directed graph neural network.

Adaptive node-level weighted learning for directed graph neural network.

Graph Neural Networks for Tabular Data Learning: A Survey with Taxonomy and Directions

This survey dives into Tabular Data Learning (TDL) using Graph Neural Networks (GNNs), a domain where deep learning-based approaches have increasingly shown superior performance in both classification and regression tasks compared to traditional methods. We highlight a critical gap in deep neural TDL: the underrepresentation of latent correlations among data instances and feature values. GNNs, with their innate capability to model intricate relationships and interactions between diverse elements of tabular data, have garnered significant interest and application across TDL domains. Our survey provides a systematic review of the methods involved in designing and implementing GNNs for TDL (GNN4TDL). It encompasses a detailed investigation into the foundational aspects and an overview of GNN-based TDL methods, offering insights into their evolving landscape. We present a comprehensive taxonomy focused on constructing graph structures and representation learning within GNN-based TDL methods. We also examine various training plans, emphasize the integration of auxiliary tasks to enhance the representation quality. A critical part of our discussion is dedicated to the practical applications across a spectrum of GNN4TDL scenarios, exhibiting their versatility and impact. Last, we discuss the limitations and future directions, aiming to spur advancements in GNN4TDL. This survey serves as a resource for researchers and practitioners, offering a thorough understanding of GNNs’ role in revolutionizing TDL and pointing towards future innovations in this promising area.

Arch-Net: Model conversion and quantization for architecture agnostic model deployment.

Arch-Net: Model conversion and quantization for architecture agnostic model deployment.

Classifying IoT firmware security threats using image analysis and deep learning

<p>As the internet of things (IoT) grows, its embedded devices face increasing vulnerability to firmware-based attacks. The lack of robust security mechanisms in IoT devices makes them susceptible to malicious firmware updates, potentially compromising entire networks. This study addresses the classification of IoT firmware security threats using deep learning and image-based analysis techniques. A publicly available dataset of 32×32 grayscale images, derived from IoT firmware samples and categorized as benignware, hackware, and malware, was utilized. The grayscale images were converted into three-channel RGB format to ensure compatibility with convolutional neural networks (CNNs). We tested multiple pre-trained CNN architectures, including SqueezeNet, ShuffleNet, MobileNet, Xception, and ResNet50, employing transfer learning to adapt the models for this classification task. Both ResNet50 and ShuffleNet achieved exceptional performance, with 100% accuracy, precision, recall, and F1-score. These results validate the effectiveness of our methodology in leveraging transfer learning for IoT firmware classification while maintaining computational efficiency, making it suitable for deployment in resource-constrained IoT environments. T</p>

Information bottleneck-driven prompt on graphs for unifying downstream few-shot classification tasks

Information bottleneck-driven prompt on graphs for unifying downstream few-shot classification tasks

Dynamic semantic-geometric guidance and structure transfer network for cross-scene hyperspectral image classification.

Dynamic semantic-geometric guidance and structure transfer network for cross-scene hyperspectral image classification.

Task-augmented cross-view imputation network for partial multi-view incomplete multi-label classification.

Task-augmented cross-view imputation network for partial multi-view incomplete multi-label classification.

Artificial intelligence for optimal in vitro fertilization morphokinetics.

Artificial intelligence for optimal in vitro fertilization morphokinetics.