Original Training Dataset Research Articles

Scoping Review on the Multimodal Classification of Depression and Experimental Study on Existing Multimodal Models.

Depression is a prevalent comorbidity in patients with severe physical disorders, such as cancer, stroke, and coronary diseases. Although it can significantly impact the course of the primary disease, the signs of depression are often underestimated and overlooked. The aim of this paper was to review algorithms for the automatic, uniform, and multimodal classification of signs of depression from human conversations and to evaluate their accuracy. For the scoping review, the PRISMA guidelines for scoping reviews were followed. In the scoping review, the search yielded 1095 papers, out of which 20 papers (8.26%) included more than two modalities, and 3 of those papers provided codes. Within the scope of this review, supported vector machine (SVM), random forest (RF), and long short-term memory network (LSTM; with gated and non-gated recurrent units) models, as well as different combinations of features, were identified as the most widely researched techniques. We tested the models using the DAIC-WOZ dataset (original training dataset) and using the SymptomMedia dataset to further assess their reliability and dependency on the nature of the training datasets. The best performance was obtained by the LSTM with gated recurrent units (F1-score of 0.64 for the DAIC-WOZ dataset). However, with a drop to an F1-score of 0.56 for the SymptomMedia dataset, the method also appears to be the most data-dependent.

Comparison of Machine Learning and Deep Learning Methods for the Prediction of Osteoradionecrosis Resulting from Head and Neck Cancer Radiation Therapy

<h3>Purpose/Objective(s)</h3> To test the hypothesis that deep learning (DL) techniques, using full dose distributions, can outperform machine learning (ML) methods, using dose summary statistics, in the prediction of osteoradionecrosis (ORN) resulting from head and neck cancer (HNC) radiotherapy (RT). <h3>Materials/Methods</h3> 1259 subjects from a single institution were identified who received HNC RT with curative intent. All 1259 subjects were included in the ML study and 1236 subjects with available dose maps and mandible contours were included in the DL study. After two years of follow-up, 173 patients developed ORN of any grade and 1086 remained ORN free (171 ORN+/1064 ORN- in the DL cohort). The ML methods, including logistic regression (LR), random forest (RF), support vector machine (SVM), principal component regression (PCR), and XGBoost, predict ORN status using subject dose summary statistics. The DL methods, including ResNet, DenseNet, DenseNet+ResNet ensemble, and autoencoder architectures, used subject 3D dose maps constrained to a bounding box around the mandible contour to predict ORN status. The autoencoder architecture uses bottleneck features with convolutional layers for prediction. The impact of training set size on DL performance was evaluated by retraining the architectures on decreasing ratios of the original training dataset (100% to 10% in 10% decrements). Model prediction performance was quantified using recall, precision, balanced accuracy, and area under the precision recall curve (AUPRC). The ML results are the average of 10-fold stratified cross-validation with 3 repeats whereas DL results are from a withheld test set (650/217/369 train/validation/test case split with 111/12/48 ORN+ cases per set, respectively). Class imbalance in the DL models was handled by randomly oversampling ORN+ cases in the training set to match the number of ORN- cases. <h3>Results</h3> The table shows the ML and DL ORN prediction results. Decreasing the amount of training data had no impact on DL performance; in the extreme of training the DL models on 10% of the training data, the balanced accuracy and F1 score did not decrease. <h3>Conclusion</h3> The traditional ML models had superior performance compared to the DL models. The lack of improvement in DL performance when increasing the amount of available training data suggests that either significantly more data is needed for DL model construction and/or that low-level dose image features are not powerful for this task. The poor DL performance despite a relatively large training cohort suggest additional imaging modalities in conjunction with 3D dose maps should be explored.

Assessment of pretreatment methods on CIA-K and CALMAG indices and the effects on paleoprecipitation estimates

The chemical compositions of paleosols are frequently used to reconstruct paleoclimatic and paleoenvironmental conditions using proxy relationships derived from modern soils. However, through time, some of the proxy relationships have been applied to paleosol types that were not part of the original training data sets, or they have been analyzed following pretreatment steps that were not employed by the original studies without critical evaluation of whether those methodological changes have impacted the interpretation of the data. Here we test pretreatment methods critically using chemical compositions of thirty-two modern soils from California and Texas to evaluate their impact on two of the most commonly employed climofunction indices to estimate mean annual precipitation, CIA-K and CALMAG. Chemical data are generated from paired soils which have underdone two different pretreatment styles, including using weak glacial acetic acid to digest calcite and using no pretreament methods to keep all calcium in the soil, including from pedogenic carbonate. There is little change in major element oxide values from untreated to acid-treated aliquots, except for calcium oxide depletion after pretreatment. In all soils with >2% CaO in the matrix, calcium oxide is significantly depleted by the acid-digestion pretreatment. This acid-digestion pretreatment induces a substantial increase in estimates of mean annual precipitation, which are not in accordance with published meteorological values. The extent to which precipitation estimates are skewed toward higher values as a function of CaO loss is parametrically correlated with calcium carbonate content of the soil material. This has important implications for the appropriate application of CIA-K and CALMAG indices as viable proxies of paleoprecipitation estimates from paleosol profiles that contain pedogenic and/or diagenetic carbonates and we suggest quantitative thresholds for CaO in untreated samples to minimize potential sources of error.

DCML: Deep contrastive mutual learning for COVID-19 recognition

COVID-19 is a form of disease triggered by a new strain of coronavirus. Automatic COVID-19 recognition using computer-aided methods is beneficial for speeding up diagnosis efficiency. Current researches usually focus on a deeper or wider neural network for COVID-19 recognition. And the implicit contrastive relationship between different samples has not been fully explored. To address these problems, we propose a novel model, called deep contrastive mutual learning (DCML), to diagnose COVID-19 more effectively. A multi-way data augmentation strategy based on Fast AutoAugment (FAA) was employed to enrich the original training dataset, which helps reduce the risk of overfitting. Then, we incorporated the popular contrastive learning idea into the conventional deep mutual learning (DML) framework to mine the relationship between diverse samples and created more discriminative image features through a new adaptive model fusion method. Experimental results on three public datasets demonstrate that the DCML model outperforms other state-of-the-art baselines. More importantly, DCML is easier to reproduce and relatively efficient, strengthening its high practicality.

Endoscopy Artefact Detection by Deep Transfer Learning of Baseline Models.

To visualise the tumours inside the body on a screen, a long and thin tube is inserted with a light source and a camera at the tip to obtain video frames inside organs in endoscopy. However, multiple artefacts exist in these video frames that cause difficulty during the diagnosis of cancers. In this research, deep learning was applied to detect eight kinds of artefacts: specularity, bubbles, saturation, contrast, blood, instrument, blur, and imaging artefacts. Based on transfer learning with pre-trained parameters and fine-tuning, two state-of-the-art methods were applied for detection: faster region-based convolutional neural networks (Faster R-CNN) and EfficientDet. Experiments were implemented on the grand challenge dataset, Endoscopy Artefact Detection and Segmentation (EAD2020). To validate our approach in this study, we used phase I of 2,200 frames and phase II of 331 frames in the original training dataset with ground-truth annotations as training and testing dataset, respectively. Among the tested methods, EfficientDet-D2 achieves a score of 0.2008 (mAPd[Formula: see text]0.6+mIoUd[Formula: see text]0.4) on the dataset that is better than three other baselines: Faster-RCNN, YOLOv3, and RetinaNet, and competitive to the best non-baseline result scored 0.25123 on the leaderboard although our testing was on phase II of 331 frames instead of the original 200 testing frames. Without extra improvement techniques beyond basic neural networks such as test-time augmentation, we showed that a simple baseline could achieve state-of-the-art performance in detecting artefacts in endoscopy. In conclusion, we proposed the combination of EfficientDet-D2 with suitable data augmentation and pre-trained parameters during fine-tuning training to detect the artefacts in endoscopy.

Learning Task-Parameterized Skills From Few Demonstrations

Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task-parameterized learning improves the generalization of motion policies by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demonstrations in different situations. To comprehensively create different situations is non-trivial thus renders the method less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. This paper presents a novel concept to augment the original training dataset with synthetic data for policy improvements, thus allows learning task-parameterized skills with few demonstrations.

GAN-regularized augmentation strategy for spectral datasets

Abstract Artificial neural networks are used in various fields including spectral unmixing, which is used to determine the proportions of substances involved in a mixture, and achieve promising results. This is especially true if there is a non-linear relationship between the spectra of mixtures and the spectra of the substances involved (pure spectra). To achieve sufficient results, neural networks need lots of representative training data. We present a method that extends existing training data for spectral unmixing consisting of spectra of mixtures by learning the mixing characteristic using an artificial neural network. Spectral variability is considered by random inputs. The network structure used is a generative adversarial net that takes the dependence on the abundances of pure substances into account by an additional term in its objective function, which is minimized during training. After training further data for abundance vectors for which there is no real measurement data in the original training dataset can be generated. A neural network trained with the augmented training dataset shows better performance in spectral unmixing compared to being trained with the original dataset. The presented network structure improves already existing results obtained with a generative convolutional neural network, which is superior to model-based approaches.

Bounding-box object augmentation with random transformations for automated defect detection in residential building façades

This study proposes a novel bounding-box object augmentation (BoxAug) method to improve the performance of deep learning models in detecting defects in residential building façades. The most significant characteristic of the method is that it augments objects in images, rather than augmenting images, to solve the data imbalance problem. Moreover, it employs the bounding-box form for object detection, instead of the segmentation mask form. To evaluate the method, 7635 images obtained using unmanned aerial vehicles were utilized as the original training dataset. The faster region-based convolutional neural network model trained with the augmented training dataset using the method exhibited better performance than the model trained with the original dataset. Particularly, the class with the least objects in the original dataset displayed a markedly improved performance. Thus, the method can serve as an auxiliary method for effectively augmenting real-world image datasets with an unbalanced number of objects.

Early Detection of Network Intrusions Using a GAN-Based One-Class Classifier

Early detection of network intrusions is a very important factor in network security. However, most studies of network intrusion detection systems utilize features for full sessions, making it difficult to detect intrusions before a session ends. To solve this problem, the proposed method uses packet data for features to determine if packets are malicious traffic. Such an approach inevitably increases the probability of falsely detecting normal packets as an intrusion or an intrusion as normal traffic for the initial session. As a solution, the proposed method learns the patterns of packets that are unhelpful in order to classify network intrusions and benign sessions. To this end, a new training dataset for Generative Adversarial Network (GAN) is created using misclassified data from an original training dataset by the LSTM-DNN model trained using the original one. The GAN trained with this dataset has ability to determine whether the currently received packet can be accurately classified in the LSTM-DNN. If the GAN determines that the packet cannot be classified correctly, the detection process is canceled and will be tried again when the next packet is received. Meticulously designed classification algorithm based on LSTM-DNN and validation model using GAN enable the proposed algorithm to accurately perform network intrusion detection in real time without session termination or delay time for collecting a certain number of packets. Various experiments confirm that the proposed method can detect intrusions very early (before the end of the session) while maintaining detection performance at a level similar to that of the existing methods.

Retracted] Comparative Analysis of Volleyball Serve Action Based on Human Posture Estimation

Serving is one of the most crucial techniques in volleyball. Serving is a method that does not require team interaction and is difficult for the opponent to immediately interfere with. The feature migration module with a fixed offset is suggested in this work. This module can be thought of as a cross‐channel dilated convolution approximation of dilated convolution. The reason cross‐channel dilated convolution is not worse than standard dilated convolution with few parameters is discussed in this article. An improved random forest model is put forth to address the issue of the human pose estimation system’s high memory consumption when utilizing random forest as the classifier. This model presents the Poisson process and incorporates it with the depth data to create a filter before using Bootstrap sampling. In order to optimize and reconstruct the training dataset, a portion of the feature sample points that do not contribute positively to subsequent classification is removed from the original training dataset. This allows the training dataset to better account for the repeated sampling of the random forest during the sampling process. Resampling has some drawbacks, but they are not very representative. The effectiveness of the optimization model, which significantly lowers the system’s time and space complexity and increases the system’s applicability, is demonstrated by experiments.

Multi layered Stacked Ensemble Method with Feature Reduction Technique for Multi-Label Classification

Nowadays, multi-label classification can be considered as one of the important challenges for classification problem. In this case instances are assigned more than one class label. Ensemble learning is a process of supervised learning where several classifiers are trained to get a better solution for a given problem. Feature reduction can be used to improve the classification accuracy by considering the class label information with principal Component Analysis (PCA). In this paper, stacked ensemble learning method with augmented class information PCA (CA PCA) is proposed for classification of multi-label data (SEMML). In the initial step, the dimensionality reduction step is applied, then the number of classifiers have to be chosen to apply on the original training dataset, then the stacking method is applied to it. By observing the results of experiments conducted are showing our proposed method is working better as compared to the existing methods.

An Evolution Approach for Pre-trained Neural Network Pruning without Original Training Dataset

An Evolution Approach for Pre-trained Neural Network Pruning without Original Training Dataset

An update to the custom-made MS Excel workbook performing the log-rank test with extended functionality and a new original COVID-19 training data set

An update to the custom-made MS Excel workbook performing the log-rank test with extended functionality and a new original COVID-19 training data set

Improving the Classification Rate of Insulation Defects in Power Cables by Automatic Features Extraction and Sample Augmentation of Partial Discharge Signal

Conventional pattern recognition methods employed for differentiating the types of insulation defects in power cables usually rely on the manual extraction of partial discharge features, which is inefficient and easily affected by subjective uncertainty. This work addresses this problem by proposing a new framework based on the automatic features extraction of partial discharge signal. The method first applies a sliding time window to convert partial discharge signals in the time domain into two-dimensional images that serve directly as the input to the convolutional neural networks (CNNs). Then a nonlinear encoder is employed to automatically extract the features of the partial discharge image data as the input of CNNs for classification. In addition, we address the overfitting problem associated with the few-shot by applying a deep convolutional generative adversarial network (DCGAN) to augment the original training dataset. Experimental results demonstrate the validity of the proposed algorithm; it increases the classification accuracy by 4.18% relative to that achieved with manually extracted features; the overall accuracy of the proposed algorithm training with the augmented dataset is 3.175% higher than that with the original experimental dataset.

Development of a fully automated RULA assessment system based on computer vision

The purpose of this study was to develop an automated, RULA-based posture assessment system using a deep learning algorithm to estimate RULA scores, including scores for wrist posture, based on images of workplace postures. The proposed posture estimation system reported a mean absolute error (MAE) of 2.86 on the validation dataset obtained by randomly splitting 20% of the original training dataset before data augmentation. The results of the proposed system were compared with those of two experts’ manual evaluation by computing the intraclass correlation coefficient (ICC), which yielded index values greater than 0.75, thereby confirming good agreement between manual raters and the proposed system. This system will reduce the time required for postural evaluation while producing highly reliable RULA scores that are consistent with those generated by manual approach. Thus, we expect that this study will aid ergonomic experts in conducting RULA-based surveys of occupational postures in workplace conditions.

A data-driven machine learning-based approach for urban land cover change modeling: A case of Khulna City Corporation area

Land use and land cover (LULC) changes have significant consequences on habitat and the environment. Past studies developed several LULC change models to identify the factors behind the changes and to simulate future LULC scenarios. However, the accuracy of these models remained contentious and a matter of ongoing debate and research. Most of these studies used a training dataset to train the model and a validation dataset to validate the prediction accuracy, both of which are a part of the original training dataset. However, to evaluate the model's actual predictive capability in terms of spatial data modeling, it is necessary to test the model's performance on the real-world dataset. In this study, we presented an XGBoost model aiming at improving the prediction accuracy while used a separate test dataset to test the model's actual predictive capacity. We applied the method to predict the land cover change of the Khulna City Corporation (KCC) area in Bangladesh. The study reveals that the KCC area experienced rapid urban development during the 2002–2018 period while the agricultural and vacant land declined at a similar rate. The major factors contributing to this substantial change of the city's land covers are-proximity to existing built-up areas and proximity to major roads. Our study indicates that agricultural areas and wetlands closer to the major roads and existing urban areas have a greater probability of converting into built-up areas. Our experiment demonstrates that the XGBoost model can predict the city's land cover change with greater accuracy and outperforms the benchmark models such as the LR-CA and ANN-CA. The finding assures the reliability of the XGBoost model while predicting future land-cover scenarios.

Improved YOLO Based Detection Algorithm for Floating Debris in Waterway.

Various floating debris in the waterway can be used as one kind of visual index to measure the water quality. The traditional image processing method is difficult to meet the requirements of real-time monitoring of floating debris in the waterway due to the complexity of the environment, such as reflection of sunlight, obstacles of water plants, a large difference between the near and far target scale, and so on. To address these issues, an improved YOLOv5s (FMA-YOLOv5s) algorithm by adding a feature map attention (FMA) layer at the end of the backbone is proposed. The mosaic data augmentation is applied to enhance the detection effect of small targets in training. A data expansion method is introduced to expand the training dataset from 1920 to 4800, which fuses the labeled target objects extracted from the original training dataset and the background images of the clean river surface in the actual scene. The comparisons of accuracy and rapidity of six models of this algorithm are completed. The experiment proves that it meets the standards of real-time object detection.

The Implementation of Multiple Linear Regression for Swimming Pool Facilities: Case Study at Jøa, Norway

This paper presents a statistical model for predicting the time-averaged total power consumption of an indoor swimming facility. The model can be a powerful tool for continuous supervision of the facility’s energy performance that can quickly disclose possible operational disruptions/irregularities and thus minimize annual energy use. Multiple linear regression analysis is used to analyze data collected in a swimming facility in Norway. The resolution of the original training dataset was in 1 min time steps and during the investigation was transposed both by time-averaging the data, and by treating part of the dataset exclusively. The statistically significant independent variables were found to be the outdoor dry-bulb temperature and the relative pool usage factor. The model accurately predicted the power consumption in the validation process, and also succeeded in disclosing all the critical operational disruptions in the validation dataset correctly. The model can therefore be applied as a dynamic energy benchmark for fault detection in swimming facilities. The final energy prediction model is relatively simple and can be deployed either in a spreadsheet or in the building automation reporting system, thus the method can contribute instantly to keep the operation of any swimming facility within the optimal individual energy performance range.

Applications of object detection in modular construction based on a comparative evaluation of deep learning algorithms

Purpose The practice of artificial intelligence (AI) is increasingly being promoted by technology developers. However, its adoption rate is still reported as low in the construction industry due to a lack of expertise and the limited reliable applications for AI technology. Hence, this paper aims to present the detailed outcome of experimentations evaluating the applicability and the performance of AI object detection algorithms for construction modular object detection. Design/methodology/approach This paper provides a thorough evaluation of two deep learning algorithms for object detection, including the faster region-based convolutional neural network (faster RCNN) and single shot multi-box detector (SSD). Two types of metrics are also presented; first, the average recall and mean average precision by image pixels; second, the recall and precision by counting. To conduct the experiments using the selected algorithms, four infrastructure and building construction sites are chosen to collect the required data, including a total of 990 images of three different but common modular objects, including modular panels, safety barricades and site fences. Findings The results of the comprehensive evaluation of the algorithms show that the performance of faster RCNN and SSD depends on the context that detection occurs. Indeed, surrounding objects and the backgrounds of the objects affect the level of accuracy obtained from the AI analysis and may particularly effect precision and recall. The analysis of loss lines shows that the loss lines for selected objects depend on both their geometry and the image background. The results on selected objects show that faster RCNN offers higher accuracy than SSD for detection of selected objects. Research limitations/implications The results show that modular object detection is crucial in construction for the achievement of the required information for project quality and safety objectives. The detection process can significantly improve monitoring object installation progress in an accurate and machine-based manner avoiding human errors. The results of this paper are limited to three construction sites, but future investigations can cover more tasks or objects from different construction sites in a fully automated manner. Originality/value This paper’s originality lies in offering new AI applications in modular construction, using a large first-hand data set collected from three construction sites. Furthermore, the paper presents the scientific evaluation results of implementing recent object detection algorithms across a set of extended metrics using the original training and validation data sets to improve the generalisability of the experimentation. This paper also provides the practitioners and scholars with a workflow on AI applications in the modular context and the first-hand referencing data.

Data-Free Knowledge Distillation with Soft Targeted Transfer Set Synthesis

Knowledge distillation (KD) has proved to be an effective approach for deep neural network compression, which learns a compact network (student) by transferring the knowledge from a pre-trained, over-parameterized network (teacher). In traditional KD, the transferred knowledge is usually obtained by feeding training samples to the teacher network to obtain the class probabilities. However, the original training dataset is not always available due to storage costs or privacy issues. In this study, we propose a novel data-free KD approach by modeling the intermediate feature space of the teacher with a multivariate normal distribution and leveraging the soft targeted labels generated by the distribution to synthesize pseudo samples as the transfer set. Several student networks trained with these synthesized transfer sets present competitive performance compared to the networks trained with the original training set and other data-free KD approaches.