Train Classifiers Research Articles

Time series clustering to improve one-class classifier performance

The improvement of one-class classifiers’ performance through clustering of multivariate time series is considered in this paper. Datasets arising from real processes come from the available sensors and are affected by many factors, such as aging of the process, changes in the operation region, and equipment malfunction. Despite that, one expects that the classes represented by such diverse data can be unveiled via trained classifiers. This work hypothesizes that the overall performance can be improved by training sets of one-class classifiers with subsets of data clustered by similarity. The proposed method is applied to one class classifiers since they are trained only with the target class, which is clustered based on time series similarity using Dynamic Time Warping and k-means. The advantages of the techniques are illustrated through their application to a public dataset from the oil industry with instances characterizing eight classes of data represented by five time series. Seven classes are selected to train LSTM classifiers using the variables and instances clustered using time series clustering algorithms. The results show that the increase in the similarity of training data tends to improve the performance of the LSTM classifier, achieving an increase of 10% in the overall performance. In a specific case, where the clustering model raised the similarity by 84%, the classification performance improved by 21%.

Model-informed classification of broadband acoustic backscatter from zooplankton in an in situ mesocosm

Abstract Classification of zooplankton to species with broadband echosounder data could increase the taxonomic resolution of acoustic surveys and reduce the dependence on net and trawl samples for ‘ground truthing’. Supervised classification with broadband echosounder data is limited by the acquisition of validated data required to train machine learning algorithms (‘classifiers’). We tested the hypothesis that acoustic scattering models could be used to train classifiers for remote classification of zooplankton. Three classifiers were trained with data from scattering models of four Arctic zooplankton groups (copepods, euphausiids, chaetognaths, and hydrozoans). We evaluated classifier predictions against observations of a mixed zooplankton community in a submerged purpose-built mesocosm (12 m3) insonified with broadband transmissions (185–255 kHz). The mesocosm was deployed from a wharf in Ny-Ålesund, Svalbard, during the Arctic polar night in January 2022. We detected 7722 tracked single targets, which were used to evaluate the classifier predictions of measured zooplankton targets. The classifiers could differentiate copepods from the other groups reasonably well, but they could not differentiate euphausiids, chaetognaths, and hydrozoans reliably due to the similarities in their modelled target spectra. We recommend that model-informed classification of zooplankton from broadband acoustic signals be used with caution until a better understanding of in situ target spectra variability is gained.

Classification of Hemilabile Ligands Using Machine Learning.

Hemilabile ligands have the capacity to partially disengage from a metal center, providing a strategy to balance stability and reactivity in catalysis, but they are not straightforward to identify. We identify ligands in the Cambridge Structural Database that have been crystallized with distinct denticities and are thus identifiable as hemilabile ligands. We implement a semi-supervised learning approach using a label-spreading algorithm to augment a small negative set that is supported by heuristic rules of ligand and metal co-occurrence. We show that a heuristic based on coordinating atom identity alone is not sufficient to identify whether a ligand is hemilabile, and our trained machine-learning classification models are instead needed to predict whether a bi-, tri-, or tetradentate ligand is hemilabile with high accuracy and precision. Feature importance analysis of our models shows that the second, third, and fourth coordination spheres all play important roles in ligand hemilability.

A novel unsupervised dynamic feature domain adaptation strategy for cross-individual myoelectric gesture recognition

Objective. Surface electromyography pattern recognition (sEMG-PR) is considered as a promising control method for human-machine interaction systems. However, the performance of a trained classifier would greatly degrade for novel users since sEMG signals are user-dependent and largely affected by a number of individual factors such as the quantity of subcutaneous fat and the skin impedance. Approach. To solve this issue, we proposed a novel unsupervised cross-individual motion recognition method that aligned sEMG features from different individuals by self-adaptive dimensional dynamic distribution adaptation (SD-DDA) in this study. In the method, both the distances of marginal and conditional distributions between source and target features were minimized through automatically selecting the optimal feature domain dimension by using a small amount of unlabeled target data. Main results. The effectiveness of the proposed method was tested on four different feature sets, and results showed that the average classification accuracy was improved by above 10% on our collected dataset with the best accuracy reached 90.4%. Compared to six kinds of classic transfer learning methods, the proposed method showed an outstanding performance with improvements of 3.2%-13.8%. Additionally, the proposed method achieved an approximate 9% improvement on a publicly available dataset. Significance. These results suggested that the proposed SD-DDA method is feasible for cross-individual motion intention recognition, which would provide help for the application of sEMG-PR based system.

Automated Peripheral Smear Recognition and Classification of LGL Leukemia Using Machine Learning

Background: Machine learning can augment clinical decision support, especially in cases where complex diagnostic criteria are necessary. In making a diagnosis of LGL leukemia, a pathologist must review the peripheral smear thoroughly, which is a tedious process. Supervised machine learning has been successfully implemented in such cases by training on very large image-based datasets, followed by a corrective phase of updating the model weights and finally demonstrating a high degree of successful predictions in an independent dataset. We sought to assess whether a supervised machine learning model can automate histopathological image analysis and provide a degree of confidence in supporting a pathologist's manual review. Methods: We searched the published literature from 2016-2022 for high-definition peripheral smear images with confirmed cases. The samples were downloaded, pre-processed with the Segment Anything Model (SAM), and labeled as training, testing, or validation data. Non-pathological controls, defined as peripheral smears confirmed negative for LGLL, were also obtained from published literature and labeled manually. The neural network was written and trained on TensorFlow using the open-source Keras library with parallel threading and on a multicore GPU machine. The resulting predictive model had a binary classification task: Does the presented image have cells that resemble LGL cells, returning true or false. Every “true” label returned from the classifier also creates a stack of probable positives that need pathology review. Feedback from each pathologist review further updates the model weights to increase the probability of successful prediction in the future. Results: We screened 57 publications to obtain 163 high-definition peripheral smear images that had confirmed LGLL and 100 non-pathological smears. Two cohorts were created: 200 smears in a training cohort comprised of 100 LGLL smears and 100 non-pathological smears, and 63 smears in the validation cohort. To remain agnostic towards image orientation, we created 7 views of each smear, and 1400 total images underwent feature extraction using SAM. The resulting data was labeled and used for model training and our process is depicted visually below in Figure 1. During validation, our model accurately predicted LGL-cells in 56/63 cases with an 8% false positive rate, 2% false negatives and 2% unclassifiable. After updating model weights and re-training with Adaptive Boosting, the accuracy increased to 60/63 smears with a false-positive rate of 3%. Our trained classification model achieved an AUC of 0.8144. Conclusion: Supervised machine learning model with Adaptive Boosting can be used to label a peripheral smear image with a high likelihood of containing LGL cells and, in turn, help a pathologist review smears with higher priority in a timely fashion. Our initial training data was limited, which hindered the performance of our model; however, with a sufficiently large training dataset, we can drastically improve our accuracy. A more extensive training set with robust hardware can transform our image-recognition model into a foundational model where other clinical parameters can be added, and the final diagnostic probability will be based on multiple data streams. Our work takes a meaningful step in this direction, and future work at Karmanos Cancer Center will focus on extending our classifier to more sophisticated deep-learning algorithms. Our open-source platform will eventually support several niches in the future that are currently only served by commercial applications.

Variant Classification Discordance: Contributing Factors and Predictive Models

An ever-growing catalog of human variants is hosted in the ClinVar database. In this database, submissions on a variant are combined into a multi-submitter record and in the case of discordance in variant classification between submitters, the record is labelled as conflicting. We used ClinVar data to identify characteristics that would make variants more likely to be associated with the conflict class of variants. Further, we used the Extreme Gradient Boosting (XGBoost) algorithm to train classifier models to provide prediction of classification discordance for single submission variants in ClinVar database.We showed that population allele frequency, the gene harboring the variant, variant type, consequence on protein, variant deleteriousness score, first submitter identity and submission count are associated with conflict in variant classification. Using such features, the optimized classifier showed accuracy on the test set of 88% with the weighted average of precision, recall and f1-score of 0.84, 0.88 and 0.85, respectively.There are pronounced associations between variant classification discordance and allele frequency, gene type, and the identity of the first submitter. We provide the predicted discordance status for single-submitter variants deposited in ClinVar. Our approach can be used to assess whether single-submitter variants are likely to be supported, or in conflict with, future entries; this knowledge might help laboratories with clinical variant assessment.

Modeling information flow from multispectral remote sensing images to land use and land cover maps for understanding classification mechanism

ABSTRACTInformation on Land Use and Land Cover Map (LULCM) is essential for environment and socioeconomic applications. Such maps are generally derived from Multispectral Remote Sensing Images (MRSI) via classification. The classification process can be described as information flow from images to maps through a trained classifier. Characterizing the information flow is essential for understanding the classification mechanism, providing solutions that address such theoretical issues as “what is the maximum number of classes that can be classified from a given MRSI?” and “how much information gain can be obtained?” Consequently, two interesting questions naturally arise, i.e. (i) How can we characterize the information flow? and (ii) What is the mathematical form of the information flow? To answer these two questions, this study first hypothesizes that thermodynamic entropy is the appropriate measure of information for both MRSI and LULCM. This hypothesis is then supported by kinetic-theory-based experiments. Thereafter, upon such an entropy, a generalized Jarzynski equation is formulated to mathematically model the information flow, which contains such parameters as thermodynamic entropy of MRSI, thermodynamic entropy of LULCM, weighted F1-score (classification accuracy), and total number of classes. This generalized Jarzynski equation has been successfully validated by hypothesis-driven experiments where 694 Sentinel-2 images are classified into 10 classes by four classical classifiers. This study provides a way for linking thermodynamic laws and concepts to the characterization and understanding of information flow in land cover classification, opening a new door for constructing domain knowledge.

Feature Extraction Based Machine Learning Approach for Bone Cancer Detection

Osteosarcoma is a type of cancer that develops in the bones. Though it can happen in any bone, it commonly happens in long bones like the legs and arms. As a result, early detection and categorization of bone cancers have become critical for treating patients. A wavelet-based segmentation algorithm was utilized in this work to detect bone cancers. The segmented bone cancer components were then processed further for categorization. The enhanced convolutional neural network (ECNN) classification was employed in this investigation to differentiate between benign and malignant bone cancers. Collect multiple photos and use wavelet transform features to extract a trained classifier model. Sensitivity (97%), Specificity (97%), Precision (98%), Accuracy (97.5%), and F1Score (97.5) are the performance metrics for the ECNN deep learning (DL) algorithm. According to the results, ECNN deep learning beats deep learning methods, including SVM, ANN, and RNN. As a result, the ECNN deep learning technology can be used to diagnose bone cancer more accurately. Based on histology pictures, our enhanced model performs at the cutting edge of detecting osteosarcoma cancer.

XL1R-Net: Explainable AI-driven improved [formula omitted]-regularized deep neural architecture for NSCLC biomarker identification

Background and Objective:Non-small cell lung cancer (NSCLC) exhibits intrinsic molecular heterogeneity, primarily driven by the mutation of specific biomarkers. Identification of these biomarkers would assist not only in distinguishing NSCLC into its major subtypes — Adenocarcinoma and Squamous Cell Carcinoma, but also in developing targeted therapy. Medical practitioners use one or more types of omic data to identify these biomarkers, copy number variation (CNV) being one such type. CNV provides a measure of genomic instability, which is considered a hallmark of carcinoma. However, the CNV data has not received much attention for biomarker identification. This paper aims to identify biomarkers for NSCLC using CNV data. Methods:An eXplainable AI (XAI)-driven L1-regularized deep learning architecture, XL1R-Net, is proposed that introduces a novel modification of the standard L1-regularized gradient descent algorithm to arrive at an improved deep neural classifier for NSCLC subtyping. Further, XAI-based feature identification has been used to leverage the trained classifier to uncover a set of twenty NCSLC-relevant biomarkers. Results:The identified biomarkers are evaluated based on their classification performance and clinical relevance. Using Multilayer Perceptron (MLP)-based model, a classification accuracy of 84.95% using 10-fold cross-validation is achieved. Moreover, the statistical significance test on the classification performance also revealed the superiority of the MLP model over the competitive machine learning models. Further, the publicly available Drug–Gene Interaction Database reveals twelve of the identified biomarkers as potentially druggable. The K–M Plotter tool was used to verify eighteen of the identified biomarkers with a high probability of predicting NSCLC patients’ likelihood of survival. While nine of the identified biomarkers confirm the recent literature, five find mention in the OncoKB Gene List. Conclusion:A set of seven novel biomarkers that have not been reported in the literature could be investigated for their potential contribution towards NSCLC therapy. Given NSCLC’s genetic diversity, using only one omics data type may not adequately capture the tumor’s complexity. Multiomics data and its integration with other sources will be examined in the future to better understand NSCLC heterogeneity.

SynDroid: An adaptive enhanced Android malware classification method based on CTGAN-SVM

Android mobile phones have the highest market share nowadays, bringing a boom of Android application programming as well as malicious software (malware) issues. Traditional machine learning and deep learning methods are widely used in Android malware detection and classification, both require plenty of application samples to train classifiers. However, the collected samples are always imbalanced, because the distribution of malware categories differs hugely in the real environment. For Android malware samples with high dimension features, and Class Imbalance Ratio coming to more than 100:1, traditional methods become weak. To fill the gap, we propose an Android malware classification model named SynDroid. The core step of SynDroid uses CTGAN-SVM to generate qualified high-dimension samples, and adaptively discards bad results. Besides, we propose KS-CIR test to help the model to determine which classes of data needed to be enhanced most. This new proposed test can measure the data in respect of both samples' quality and quantity. Lastly, Random Forest is taken as a classifier to finish the classification task. The performance of SynDroid is evaluated on CCCS-CIC-AndMal2020 on the accuracy, precision, recall and F1-score. Both longitudinal and horizontal comparison experiments have been done with traditional oversampling methods, cost-sensitive learning, and other complicated methods. The result shows that the proposed method gets 12% more accuracy than the method on the same dataset and alleviates imbalanced data problems.

Pest recognition based on multi-image feature localization and adaptive filtering fusion.

Accurate recognition of pest categories is crucial for effective pest control. Due to issues such as the large variation in pest appearance, low data quality, and complex real-world environments, pest recognition poses challenges in practical applications. At present, many models have made great efforts on the real scene dataset IP102, but the highest recognition accuracy is only 75%. To improve pest recognition in practice, this paper proposes a multi-image fusion recognition method. Considering that farmers have easy access to data, the method performs fusion recognition on multiple images of the same pest instead of the conventional single image. Specifically, the method first uses convolutional neural network (CNN) to extract feature maps from these images. Then, an effective feature localization module (EFLM) captures the feature maps outputted by all blocks of the last convolutional stage of the CNN, marks the regions with large activation values as pest locations, and then integrates and crops them to obtain the localized features. Next, the adaptive filtering fusion module (AFFM) learns gate masks and selection masks for these features to eliminate interference from useless information, and uses the attention mechanism to select beneficial features for fusion. Finally, the classifier categorizes the fused features and the soft voting (SV) module integrates these results to obtain the final pest category. The principle of the model is activation value localization, feature filtering and fusion, and voting integration. The experimental results indicate that the proposed method can train high-performance feature extractors and classifiers, achieving recognition accuracy of 73.9%, 99.8%, and 99.7% on IP102, D0, and ETP, respectively, surpassing most single models. The results also show that thanks to the positive role of each module, the accuracy of multi-image fusion recognition reaches the state-of-the-art level of 96.1%, 100%, and 100% on IP102, D0, and ETP using 5, 2, and 2 images, respectively, which meets the requirements of practical applications. Additionally, we have developed a web application that applies our research findings in practice to assist farmers in reliable pest identification and drive the advancement of smart agriculture.

Prediction of therapeutic intensity level from automatic multiclass segmentation of traumatic brain injury lesions on CT-scans

The prediction of the therapeutic intensity level (TIL) for severe traumatic brain injury (TBI) patients at the early phase of intensive care unit (ICU) remains challenging. Computed tomography images are still manually quantified and then underexploited. In this study, we develop an artificial intelligence-based tool to segment brain lesions on admission CT-scan and predict TIL within the first week in the ICU. A cohort of 29 head injured patients (87 CT-scans; Dataset1) was used to localize (using a structural atlas), segment (manually or automatically with or without transfer learning) 4 or 7 types of lesions and use these metrics to train classifiers, evaluated with AUC on a nested cross-validation, to predict requirements for TIL sum of 11 points or more during the 8 first days in ICU. The validation of the performances of both segmentation and classification tasks was done with Dice and accuracy scores on a sub-dataset of Dataset1 (internal validation) and an external dataset of 12 TBI patients (12 CT-scans; Dataset2). Automatic 4-class segmentation (without transfer learning) was not able to correctly predict the apparition of a day of extreme TIL (AUC = 60 ± 23%). In contrast, manual quantification of volumes of 7 lesions and their spatial location provided a significantly better prediction power (AUC = 89 ± 17%). Transfer learning significantly improved the automatic 4-class segmentation (DICE scores 0.63 vs 0.34) and trained more efficiently a 7-class convolutional neural network (DICE = 0.64). Both validations showed that segmentations based on transfer learning were able to predict extreme TIL with better or equivalent accuracy (83%) as those made with manual segmentations. Our automatic characterization (volume, type and spatial location) of initial brain lesions observed on CT-scan, publicly available on a dedicated computing platform, could predict requirements for high TIL during the first 8 days after severe TBI. Transfer learning strategies may improve the accuracy of CNN-based segmentation models.Trial registrations Radiomic-TBI cohort; NCT04058379, first posted: 15 august 2019; Radioxy-TC cohort; Health Data Hub index F20220207212747, first posted: 7 February 2022.

Learning from class-imbalanced data using misclassification-focusing generative adversarial networks

This paper presents a novel end-to-end oversampling-classification approach, which we refer to as imbalanced data-classifying generative adversarial network (ImbGAN), for imbalanced data classification. ImbGAN has a classifier-embedded structure within a GAN and consists of five components: (1) generator, (2) discriminator, (3) classifier, (4) storage for misclassified minority class data, and (5) storage for artificial minority class data. By iterative interaction with the embedded classifier, the first two components generate artificial minority class instances that are similar to minority class instances misclassified by the classifier. Therefore, these three networks are iteratively and simultaneously trained. The misclassified and artificial minority class instances are stored in the fourth and fifth components, respectively. These two components are also updated as iterations proceed. Our method obtains the final classification model from a single learning process, while most artificial data generation methods for imbalanced data classification go through an additional process for training classifiers after artificial data generation. Numerical experiments based on tabular, image, and text datasets confirm that the proposed method outperforms well-known synthetic sampling methods.

Artificial intelligence in pediatric echocardiography-Automated view classification and image anonymization in rare cardiac malformations on the example of borderline HLHS

Abstract Background/Introduction Artificial intelligence (AI) has already much potential in adult echocardiography. However, there is still few experience in how to effectively implement AI methods in pediatric echocardiography. Since children with congenital heart defects show a large anatomical variance, AI-based image analysis tools adapted to this group are necessary. For the fundamentally needed steps of an automatic image data anonymization and a view classification there are still no reliable tools available. Purpose We have developed an anonymization tool for pediatric echocardiography data. For effective automated analysis of echocardiographies in children, we established an algorithm using bHLHS as an example that requires only small numbers of patients for a good view classification accuracy. Thus, we expect that the algorithm will also be applicable to other cardiac malformations. Method Raw data were obtained from visage imaging platforms and echocardiography examinations were conducted on Siemens machines. An anonymization tool was developed using the Python programming language and the Tkinter library. It converts medical images and crops the embedded data from the non-anatomical top region. The classification neural network was evaluated with data obtained from 5 patients with bHLHS. After data cleaning, each patient's image is labelled according to different Echo views: 4CH, SA, and LA, creating a total of 6600 images of 3 separate views. Due to the limited echocardiography dataset, a data augmentation approach was employed to improve the generalization of the AI model. The transfer learning approach was used to develop the AI model, which utilized a pre-trained VGG 13 model that was modified to classify the three different views of the pediatric echocardiography dataset. The modified model was fine-tuned on the dataset with a GPU-enhanced Intel i9 platform and Pycharm 2022.2.4. We use 70% dataset for training, 20% dataset for testing, and 10% dataset for validating. Results Table 1 shows that the expert clinician validated the anonymization tool by comparing identified patient information in the raw image and anonymized image. The trained pediatric echocardiography view classification model achieved an overall accuracy of 0.776 with precision values of 0.725, 0.889, and 0.775 for LA, SA and 4CH classes, respectively. The F1 score values were 0.835, 0.690, and 0.778 for the same classes. ROC curve were plotted (Figure 1) to visualize the performance of the model. Conclusion This study shows the importance of data anonymization and view classification in automating the diagnosis of HLHS using echocardiography images. The developed deep learning algorithm showed promising performance in accurately classifying different views of echocardiography images for diagnosing HLHS. Future studies can build upon this work by developing a complete diagnostic system to automatically analyze cardiac structures of patients with cardiac malformations.

BERT based severity prediction of bug reports for the maintenance of mobile applications

Mobile application maintenance is crucial to ensuring the accurate operation and continuous improvement of mobile applications (mobile apps). To effectively address issues and enhance the user experience, developers utilize issue-tracking systems that gather bug reports to refine mobile apps. Users can submit bugs through these systems, allowing them to determine the severity of each reported issue. The severity level plays a pivotal role in prioritizing bug resolution, enabling developers to address critical bugs promptly. Nonetheless, manually assessing the severity of each issue can be laborious and prone to errors. To overcome this challenge, this paper presents Bidirectional Encoder Representations from Transformers (BERT) based severity prediction of bug reports (called BERT-SBR) that leverages a deep neural network for automatic bug severity classification for mobile app maintenance. We collect the publicly available mobile apps bug reports dataset from the Hugging Face. BERT-SBR first computes the sentiment of reporters of bug reports and preprocesses them by leveraging BertTokenizer input formatting techniques. Next, it passes the formatted text and computed sentiment of each bug report to generate word embeddings. Then, it introduces a fine-tuned BERT classifier for bug report severity prediction. After that, it passes the generated word embeddings to the fine-tuned BERT classifier for training and testing. Finally, the proposed classifier’s performance is evaluated. The BERT-SBR assessment results confirm that the fine-tuned BERT classifies bug reports significantly more effectively than other deep learning classifiers. On average, BERT-SBR achieves a remarkable improvement of 40.43%, 67.78%, 40.71%, and 58.14% in the accuracy, precision, recall, and f-measure. This indicates its superiority in accurately predicting the severity of bug reports for mobile application maintenance.

Abstract 17924: Machine Learning Analysis of the Atrial ECG in Patients With Persistent Atrial Fibrillation Predicts Long-Term Outcomes

Background: The surface Atrial ECG (A-ECG) time and frequency domains features are not currently used clinically in patients with persistent atrial fibrillation (PeAF). Machine Learning (ML) can provide novel insight into various data sources. Research question: Can ML analysis of the A-ECG predict the outcome of catheter ablation (CA) of PeAF? Aim: To use ML Tools to assess A-ECG prediction of CA outcomes in PeAF patients at 1 and 2 years. Methods: Thirty-two patients (age 63±9 years; 71% males) with PeAF underwent two 5-minute epochs of surface ECG; 4 were excluded for significant artifacts and 2 more for unrelated clinical issues. The Pre-CA A-ECG was derived by QRST subtraction and independent component analysis filtering and divided into 50% overlap 10-second windows for spectral analysis. A-ECG mean dominant frequency DF, mean organization index and median amplitude M-AMP were computed for each lead. Freedom from atrial arrhythmia AA (Responders) was assessed at one and two years. The two 5-minute segments per patient were included in the data set. Classification Learner (Mathworks Inc., Natick, MA, USA) was used to train classifiers, looking for the minimum number of features able to explain the outcomes. Once the best (2-3) features were identified, a MATLAB program was used to visualize the decision regions and correlate with ablation outcomes. Results: ML identified A-ECG features DF and M-AMP of V 1 to well explain outcome. Specifically, an Artificial Neural Network (ANN) with 1 fully connected hidden layer of size 7 can explain the data with 90.4% accuracy. The figure shows the datasets for year 1 and year 2 CA outcomes, superimposed with ANN decision regions, for Responders and non-Responders. High DF and high M-AMP are predictors of responders and Low DF low M-AMP for non-Responders. Conclusion: ML analysis of the A-ECG is a promising tool to predict Responders to PVI CA for treatment of PeAF. This predictive algorithm needs to be tested on a prospective cohort.

FaceGAN: Robust Face Recognition using Generative Adversarial Networks (GAN) Algorithm

Generative Adversarial Networks (GANs) are a type of neural network that can generate synthetic images that are often indistinguishable from real ones. The article explores GAN to augment existing datasets or generate new ones for training classifiers. The competitive training process of GANs results in a generator network that can produce increasingly realistic images to create more diverse and balanced datasets for training classifiers. The article discusses several successful applications of GANs in image classification, including object recognition, face classification, and medical image analysis. The datasets used in this article are CelebA and FER2013. The CelebA dataset consists of 202,599 celebrity images with 40 attributes, such as gender, age, and facial hair. The FER2013 dataset consists of 35,887 images of faces with seven other emotions, including anger, disgust, fear, happiness, sadness, surprise, and neutral. The dataset is divided into training, validation, and test sets. We resized the images to 64x64 pixels and normalized the pixel values between -1 and 1, then trained a GAN model using the dataset. We evaluate the performance of our approach and compare it with several state-of-the-art methods, including Support Vector Machines (SVM) and Convolutional Neural Networks (CNN). We evaluate the performance of our approach and compare it with several state-of-the-art methods, including Support Vector Machines (SVM) and Convolutional Neural Networks (CNN), with the results that our approach outperforms SVM and CNN methods on both datasets, achieving a classification accuracy of 89.2% on CelebA and 72.5% in FER2013. Meanwhile, classification accuracy on SVM was 82.3% on CelebA and 65.4% on FER2013. Classification accuracy on CNN is 87.9% on CelebA and 70.8% on FER2013.

Automatic detection of artifacts and improved classification models for emotional activity detection from multimodal physiological data

This manuscript proposes an automated artifacts detection and multimodal classification system for human emotion analysis from human physiological signals. First, multimodal physiological data, including the Electrodermal Activity (EDA), electrocardiogram (ECG), Blood Volume Pulse (BVP) and respiration rate signals are collected. Second, a Modified Compressed Sensing-based Decomposition (MCSD) is used to extract the informative Skin Conductance Response (SCR) events of the EDA signal. Third, raw features (edge and sharp variations), statistical and wavelet coefficient features of EDA, ECG, BVP, respiration and SCR signals are obtained. Fourth, the extracted raw features, statistical and wavelet coefficient features from all physiological signals are fed into the parallel Deep Convolutional Neural Network (DCNN) to reduce the dimensionality of feature space by removing artifacts. Fifth, the fused artifact-free feature vector is obtained for neutral, stress and pleasure emotion classes. Sixth, an artifact-free feature vector is used to train the Random Forest Deep Neural Network (RFDNN) classifier. Then, a trained RFDNN classifier is applied to classify the test signals into different emotion classes. Thus, leveraging the strengths of both RF and DNN algorithms, more comprehensive feature learning using multimodal psychological data is achieved, resulting in robust and accurate classification of human emotional activities. Finally, an extensive experiment using the Wearable Stress and Affect Detection (WESAD) dataset shows that the proposed system outperforms other existing human emotion classification systems using physiological data.

EEG-based multimodal emotion recognition with optimal trained hybrid classifier

EEG-based multimodal emotion recognition with optimal trained hybrid classifier

HyperSpikeASIC: Accelerating Event-Based Workloads With HyperDimensional Computing and Spiking Neural Networks

Today’s Machine Learning(ML) systems, running workloads such as Deep Neural Networks, which require billions of parameters and many hours to train a model, consume a significant amount of energy. Due to the complexity of computation and topology, even the quantized models are hard to deploy on edge devices under energy constraints. To combat this, researchers have been focusing on new emerging neuromorphic computing models. Two of those models are HDC and SNN, both with their own benefits. HDC has various desirable properties that other ML algorithms lack such as robustness to noise, simple operations, and high parallelism. SNN are able to process event-based signal data in an efficient manner. This work develops, which utilizes a single, randomly initialized, and untrained SNN layer as a feature extractor connected to a trained HDC classifier. HDC is used to enable more efficient classification as well as provide robustness to errors. We experimentally show that is on average 31.5× more robust to errors than traditional SNNs. On Intel’s Loihi Loihi, is 10× faster and 2.6× more energy efficient over traditional SNN networks. We further develop, a customized accelerator for . By decoupling the neuron and synapses, skips the inactive neurons and limits the neuron state updating to once per time step at most. is 601× faster and 3467× more energy efficient than running on Intel’s Loihi for SNN acceleration, and 12.2× faster and 211× more energy efficient than the state-of-the-art SNN ASIC implementation LSMCore.