Training Data Research Articles

A novel CNN-based image segmentation pipeline for individualized feline spinal cord stimulation modeling.

Spinal cord stimulation (SCS) is a well-established treatment for the management of certain chronic pain conditions. More recently, it has also garnered attention as a means of modulating neural activity with the goal of restoring lost autonomic or sensory-motor function. Personalized modeling and treatment planning are critical aspects of safe and effective SCS [46, 60]. However, the generation of spine models at the required level of detail and accuracy requires time and labor intensive manual image segmentation by human experts. Hence, there is a need for maximally automated segmentation routines capable of producing high-quality anatomical models that can be used even in cases where available data is limited. To this end, we developed an automated image segmentation and model generation pipeline based on a novel Convolutional Neural Network (CNN) architecture trained on feline spinal cord magnetic resonance imaging (MRI) data. The pipeline includes steps for image preprocessing, data augmentation, transfer learning and cleanup. To assess the relative importance of each step in the pipeline and of our choice of CNN architecture, we systematically dropped steps or substituted architectures, quantifying the downstream effects in terms of tissue segmentation quality (Jaccard index and Hausdorff distance) and predicted nerve recruitment (estimated axonal depolarization). This leaveone-out analysis demonstrated that each pipeline step contributed a small but measurable increment to mean segmentation quality. Surprisingly, minor differences in segmentation accuracy translated to significant deviations (ranging between 4% and 13% for each pipeline step) in predicted nerve recruitment, highlighting the importance of careful workflow design. To our knowledge, this is the first analysis to also assess the downstream impact of segmentation quality differences on neurostimulation predictions. Furthermore, transfer learning techniques enhanced segmentation metric consistency and allowed generalization to a completely different spine region with minimal additional training data. This work helps pave the way towards fully automated, personalized SCS treatment planning in clinical settings.

Deformable registration of magnetic resonance images using unsupervised deep learning in neuro-/radiation oncology

PurposeAccurate deformable registration of magnetic resonance imaging (MRI) scans containing pathologies is challenging due to changes in tissue appearance. In this paper, we developed a novel automated three-dimensional (3D) convolutional U-Net based deformable image registration (ConvUNet-DIR) method using unsupervised learning to establish correspondence between baseline pre-operative and follow-up MRI scans of patients with brain glioma.MethodsThis study involved multi-parametric brain MRI scans (T1, T1-contrast enhanced, T2, FLAIR) acquired at pre-operative and follow-up time for 160 patients diagnosed with glioma, representing the BraTS-Reg 2022 challenge dataset. ConvUNet-DIR, a deep learning-based deformable registration workflow using 3D U-Net style architecture as a core, was developed to establish correspondence between the MRI scans. The workflow consists of three components: (1) the U-Net learns features from pairs of MRI scans and estimates a mapping between them, (2) the grid generator computes the sampling grid based on the derived transformation parameters, and (3) the spatial transformation layer generates a warped image by applying the sampling operation using interpolation. A similarity measure was used as a loss function for the network with a regularization parameter limiting the deformation. The model was trained via unsupervised learning using pairs of MRI scans on a training data set (n = 102) and validated on a validation data set (n = 26) to assess its generalizability. Its performance was evaluated on a test set (n = 32) by computing the Dice score and structural similarity index (SSIM) quantitative metrics. The model’s performance also was compared with the baseline state-of-the-art VoxelMorph (VM1 and VM2) learning-based algorithms.ResultsThe ConvUNet-DIR model showed promising competency in performing accurate 3D deformable registration. It achieved a mean Dice score of 0.975 ± 0.003 and SSIM of 0.908 ± 0.011 on the test set (n = 32). Experimental results also demonstrated that ConvUNet-DIR outperformed the VoxelMorph algorithms concerning Dice (VM1: 0.969 ± 0.006 and VM2: 0.957 ± 0.008) and SSIM (VM1: 0.893 ± 0.012 and VM2: 0.857 ± 0.017) metrics. The time required to perform a registration for a pair of MRI scans is about 1 s on the CPU.ConclusionsThe developed deep learning-based model can perform an end-to-end deformable registration of a pair of 3D MRI scans for glioma patients without human intervention. The model could provide accurate, efficient, and robust deformable registration without needing pre-alignment and labeling. It outperformed the state-of-the-art VoxelMorph learning-based deformable registration algorithms and other supervised/unsupervised deep learning-based methods reported in the literature.

Applying oversampling before cross-validation will lead to high bias in radiomics

Class imbalance is often unavoidable for radiomic data collected from clinical routine. It can create problems during classifier training since the majority class could dominate the minority class. Consequently, resampling methods like oversampling or undersampling are applied to the data to class-balance the data. However, the resampling must not be applied upfront to all data because it would lead to data leakage and, therefore, to erroneous results. This study aims to measure the extent of this bias. Five-fold cross-validation with 30 repeats was performed using a set of 15 radiomic datasets to train predictive models. The training involved two scenarios: first, the models were trained correctly by applying the resampling methods during the cross-validation. Second, the models were trained incorrectly by performing the resampling on all the data before cross-validation. The bias was defined empirically as the difference between the best-performing models in both scenarios in terms of area under the receiver operating characteristic curve (AUC), sensitivity, specificity, balanced accuracy, and the Brier score. In addition, a simulation study was performed on a randomly generated dataset for verification. The results demonstrated that incorrectly applying the oversampling methods to all data resulted in a large positive bias (up to 0.34 in AUC, 0.33 in sensitivity, 0.31 in specificity, and 0.37 in balanced accuracy). The bias depended on the data balance, and approximately an increase of 0.10 in the AUC was observed for each increase in imbalance. The models also showed a bias in calibration measured using the Brier score, which differed by up to −0.18 between the correctly and incorrectly trained models. The undersampling methods were not affected significantly by bias. These results emphasize that any resampling method should be applied correctly only to the training data to avoid data leakage and, subsequently, biased model performance and calibration.

Non-linear Kernel Optimisation of Support Vector Machine Algorithm for Online Marketplace Sentiment Analysis

Twitter is a social media platform that is very important in the digital world. Fast communication and interaction make Twitter a vital information center in sentiment analysis. The purpose of this research is to classify public opinion about the presence of marketplaces in Indonesia, both positive and negative sentiments, using a Non-linear SVM algorithm based on 1276 tweets. This research involves the stages of data pre-processing, labeling, feature extraction using TF-IDF, and data division into three scenarios: 80% training data and 20% test data, 50% training data and 50% test data scenario, and 20% training data and 80% test data scenario. The last process, GridSearchCV, combines cross-validation and non-linear SVM parameters for model evaluation using a confusion matrix. The best SVM model resulting from the scenario was 80% training and 20% test data, with hyperparameters Gamma = 100 and C = 0.01, achieving 89% accuracy. When tested on never-before-seen data, the accuracy increased to 90%, with an f1-score of 91%, precision of 88%, and recall of 95% on negative sentiments. In conclusion, evaluating the performance of non-linear SVM kernels with a combination of hyperparameter values can improve accuracy, especially on public response information about online marketplaces and public sentiment.

FlexDTI: flexible diffusion gradient encoding scheme-based highly efficient diffusion tensor imaging using deep learning

Objective. Most deep neural network-based diffusion tensor imaging methods require the diffusion gradients’ number and directions in the data to be reconstructed to match those in the training data. This work aims to develop and evaluate a novel dynamic-convolution-based method called FlexDTI for highly efficient diffusion tensor reconstruction with flexible diffusion encoding gradient scheme. Approach. FlexDTI was developed to achieve high-quality DTI parametric mapping with flexible number and directions of diffusion encoding gradients. The method used dynamic convolution kernels to embed diffusion gradient direction information into feature maps of the corresponding diffusion signal. Furthermore, it realized the generalization of a flexible number of diffusion gradient directions by setting the maximum number of input channels of the network. The network was trained and tested using datasets from the Human Connectome Project and local hospitals. Results from FlexDTI and other advanced tensor parameter estimation methods were compared. Main results. Compared to other methods, FlexDTI successfully achieves high-quality diffusion tensor-derived parameters even if the number and directions of diffusion encoding gradients change. It reduces normalized root mean squared error by about 50% on fractional anisotropy and 15% on mean diffusivity, compared with the state-of-the-art deep learning method with flexible diffusion encoding gradient scheme. Significance. FlexDTI can well learn diffusion gradient direction information to achieve generalized DTI reconstruction with flexible diffusion gradient scheme. Both flexibility and reconstruction quality can be taken into account in this network.

PrivatEyes: Appearance-based Gaze Estimation Using Federated Secure Multi-Party Computation

Latest gaze estimation methods require large-scale training data but their collection and exchange pose significant privacy risks. We propose PrivatEyes - the first privacy-enhancing training approach for appearance-based gaze estimation based on federated learning (FL) and secure multi-party computation (MPC). PrivatEyes enables training gaze estimators on multiple local datasets across different users and server-based secure aggregation of the individual estimators' updates. PrivatEyes guarantees that individual gaze data remains private even if a majority of the aggregating servers is malicious. We also introduce a new data leakage attack DualView that shows that PrivatEyes limits the leakage of private training data more effectively than previous approaches. Evaluations on the MPIIGaze, MPIIFaceGaze, GazeCapture, and NVGaze datasets further show that the improved privacy does not lead to a lower gaze estimation accuracy or substantially higher computational costs - both of which are on par with its non-secure counterparts.

Limitations of Large Language Models in Propaganda Detection Task

Propaganda in the digital era is often associated with online news. In this study, we focused on the use of large language models and their detection of propaganda techniques in the electronic press to investigate whether it is a noteworthy replacement for human annotators. We prepared prompts for generative pre-trained transformer models to find spans in news articles where propaganda techniques appear and name them. Our study was divided into three experiments on different datasets—two based on an annotated SemEval2020 Task 11 corpora and one on an unannotated subset of the Polish Online News Corpus, which we claim to be an even bigger challenge as an example of an under-resourced language. Reproduction of the results of the first experiment resulted in a higher recall of 64.53% than the original run, and the highest precision of 81.82% was achieved for gpt-4-1106-preview CoT. None of our attempts outperformed the baseline F1 score. One of the attempts with gpt-4-0125-preview on original SemEval2020 Task 11 achieved an almost 20% F1 score, but it was below the baseline, which oscillated around 50%. Part of our work that was dedicated to Polish articles showed that gpt-4-0125-preview had a 74% accuracy in the binary detection of propaganda techniques and 69% in propaganda technique classification. The results for SemEval2020 show that the outputs of generative models tend to be unpredictable and are hardly reproducible for propaganda detection. For the time being, these are unreliable methods for this task, but we believe they can help to generate more training data.

SYNTHETIC DATA GENERATION FOR ANN MODELING OF THE HYDRODYNAMIC PROCESSES OF IN-SITU LEACHING

The work presents an approach to enhance the forecasting capabilities of In-Situ Leaching processes during both the production stage and early prognosis. ISL, a crucial method for resource extraction, demands rapid on-site forecasting to guide the deployment of new technological blocks. Traditional modeling techniques, though effective, are hindered by their computational demands and network throughput requirements, particularly when dealing with substantial datasets or remote computing needs. The integration of AI technologies, specifically neural networks, offers a promising opportunity for expedited calculations by leveraging the power of forward propagation through pretrained neural models. However, a critical challenge lies in transforming conventional numerical datasets into a format suitable for neural modeling. Furthermore, the scarcity of training data during the production phase, where vital parameters are concealed underground, poses an additional challenge in training AI models for In-Situ Leaching processes. This research addresses these challenges by proposing a methodology for generating training data tailored to the most resource-intensive Computational Fluid Dynamics problems encountered during modeling. Traditional numerical modeling techniques are harnessed to construct training datasets comprising input and corresponding expected output data, with a particular focus on varying well network patterns. Subsequent efforts are directed at the conversion of the acquired data into a format compatible with neural networks. The data is normalized to align with the data ranges stipulated by the activation functions employed within the neural network architecture. This preprocessing step ensures that the neural model can effectively learn from the generated data, facilitating accurate forecasting of In-Situ Leaching processes. An advantage of proposed technique lies in provision of large, reliable datasets to train neural network to predict hydrodynamic properties based on technological regimes currently active or expected on ISL site. A major implication of this approach lies in applicability of pre-trained AI technologies to forecast future or determine current hydrodynamic regime in the stratum circumventing cost deterministic simulations currently deployed at mining sites. Hence, innovative approach outlined in this paper holds promise for optimizing forecasting, allowing for quicker and more efficient decision-making in resource extraction operations while getting around the computational barriers associated with traditional methods.

Improving Stroke Detection with Hybrid Sampling and Cascade Generalization

The prevalence of stroke in Indonesia has increased. One survey in Indonesia that contains information about the health conditions of the Indonesian people is the Indonesian Family Life Survey (IFLS). The proportion of respondents who had a stroke and non-stroke in IFLS5 showed an imbalance with an extreme level of imbalance; hence, this research aims to overcome this problem with SMOTE, SMOTE-Tomek Link, and SMOTE-ENN; then, the balanced dataset is classified using the ensemble and cascade approaches to improve the detection of stroke risk and to identify the important variables. However, the stroke respondents were still challenging to classify after imbalance class handling, presumably because of the large amount of data before and after balancing. The solution is to balance the training data with various percentages. The results showed the best percentage is applied to 5% of the training data, balanced by the SMOTE-ENN, and the ensemble method with the cascade approach increases the sensitivity and balanced accuracy values. Random forest and logistic regression combine models that produce the best performance, with a classification tree as the final model. The important variables obtained from this combination are the addition of probability from random forest, logistic regression, history of hypertension, age, and physical activity.

A Projection Approach to Local Regression with Variable-Dimension Covariates

Incomplete covariate vectors are known to be problematic for estimation and inferences on model parameters, but their impact on prediction performance is less understood. We develop an imputation-free method that builds on a random partition model admitting variable-dimension covariates. Cluster-specific response models further incorporate covariates via linear predictors, facilitating estimation of smooth prediction surfaces with relatively few clusters. We exploit marginalization techniques of Gaussian kernels to analytically project response distributions according to any pattern of missing covariates, yielding a local regression with internally consistent uncertainty propagation that utilizes only one set of coefficients per cluster. Aggressive shrinkage of these coefficients regulates uncertainty due to missing covariates. The method allows in- and out-of-sample prediction for any missingness pattern, even if the pattern in a new subject’s incomplete covariate vector was not seen in the training data. We develop an MCMC algorithm for posterior sampling that improves a computationally expensive update for latent cluster allocation. Finally, we demonstrate the model’s effectiveness for nonlinear point and density prediction under various circumstances by comparing with other recent methods for regression of variable dimensions on synthetic and real data. Supplemental materials are available online.

Bagging Algorithm Based on Possibilistic Data Interpolation for Brain-Computer Interface

Recently, brain-computer interfaces (BCIs) and brain-machine interfaces have garnered the attention of researchers. Based on connections with external devices, external computers and machines can be controlled by brain signals measured via near-infrared spectroscopy (NIRS) or electroencephalograph devices. Herein, we propose a novel bagging algorithm that generates interpolation data around misclassified data using a possibilistic function, to be applied to BCIs. In contrast to AdaBoost, which is a conventional ensemble learning method that increases the weight of misclassified data to incorporate them with high probability to the next datasets, we generate interpolation data using a membership function centered on misclassified data and incorporate them into the next datasets simultaneously. The interpolated data are known as virtual data herein. By adding the virtual data to the training data, the volume of the training data becomes sufficient for adjusting the discriminate boundary more accurately. Because the membership function is defined as a possibility distribution, this method is named the bagging algorithm based on the possibility distribution. Herein, we formulate a bagging-type ensemble learning based on the possibility distribution and discuss the usefulness of the proposed method for solving simple calculations using NIRS data.

Is Federated Learning Still Alive in the Foundation Model Era?

Federated learning (FL) has arisen as an alternative to collecting large amounts of data in a central place to train a machine learning (ML) model. FL is privacy-friendly, allowing multiple parties to collaboratively train an ML model without exchanging or transmitting their training data. For this purpose, an aggregator iteratively coordinates the training process among parties, and parties simply share with the aggregator model updates, which contain information pertinent to the model such as neural network weights. Besides privacy, generalization has been another key driver for FL: parties who do not have enough data to train a good performing model by themselves can now engage in FL to obtain an ML model suitable for their tasks. Products and real applications in the industry and consumer space have demonstrated the power of this learning paradigm. Recently, foundation models have taken the AI community by storm, promising to solve the shortage of labeled data. A foundation model is a powerful model that can be recycled for a variety of use cases by applying techniques such as zero-shot learning and full or parameter-efficient fine tuning. The premise is that the amount of data required to fine tune a foundation model for a new task is much smaller than fully training a traditional model from scratch. The reason why this is the case is that a good foundation model has already learned relevant general representations, and thus, adapting it to a new task only requires a minimal number of additional samples. This raises the question: Is FL still alive in the era of foundation models? In this talk, I will address this question. I will present some use cases where FL is very much alive. In these use cases, finding a foundation model with a desired representation is difficult if not impossible. With this pragmatic point of view, I hope to shed some light into a real use case where disparate private data is available in isolation at different parties and where labels may be located at a single party that doesn’t have any other information, making it impossible for a single party to train a model on its own. Furthermore, in some vertically-partitioned scenarios, cleaning data is not an option due to privacy-related reasons and it is not clear how to apply foundation models. Finally, I will also go over a few other requirements that are often overlooked, such as unlearning of data and its implications for the lifecycle management of FL and systems based on foundation models.

Synthetic aperture radar automatic target recognition based on cost‐sensitive awareness generative adversarial network for imbalanced data

AbstractIn military contexts, synthetic aperture radar (SAR) automatic target recognition (ATR) models frequently encounter the challenge of imbalanced data, resulting in a noticeable degradation in the recognition performance. Therefore, the authors propose a cost‐sensitive awareness generative adversarial network (CAGAN) model, aiming to improve the robustness of ATR models under imbalanced data. Firstly, the authors introduce a convolutional neural network (DCNN) to extract features. Then, the synthetic minority over‐sampling technique (SMOTE) is applied to achieve feature‐level balancing for the minority category. Finally, a CAGAN model is designed to perform the final classification task. In this process, the GAN‐based adversarial training mechanism enriches the diversity of training samples, making the ATR model more comprehensive in understanding different categories. In addition, the cost matrix increases the penalty for misclassification results and further improves the classification accuracy. Simultaneously, the cost‐sensitive awareness can accurately adjust the cost matrix through training data, thus reducing dependence on expert knowledge and improving the generalisation performance of the ATR model. This model is an end‐to‐end ATR scheme, which has been experimentally validated on the MSTAR and OpenSARship datasets. Compared to other methods, the proposed method exhibits strong robustness in dealing with various imbalanced scenarios and significant generalisation capability across different datasets.

Diversity, Equity, and Inclusion, and the Deployment of Artificial Intelligence Within the Department of Defense

Artificial Intelligence (AI) adoption has seen substantial growth across industries. This paper explores the escalating use of AI within the United States Department of Defense (DoD) and the implications that diversity, equity, and inclusion (DEI) have on Service members and Civilians across the Department. More specifically, this paper explores the DEI considerations within AI technologies on individual, team, and Department readiness. The DoD's AI usage spans various strategic and operational capabilities, however this paper explores two critical domains: healthcare and recruitment. In healthcare, AI offers the promise of early disease detection, enhanced diagnostic capabilities, and streamlined administrative processes. However, potential biases stemming from homogenous training data threaten the accuracy and reliability of these systems, jeopardizing Service member health and eroding trust in AI-assisted medical decision-making and potentially the DoD at large. In recruitment, while AI promises efficiency in identifying ideal candidates, its deployment can perpetuate biases, especially when the training data used is not representative of all demographics. Despite efforts to design "unbiased" systems by excluding demographic data, such strategies may inadvertently overlook the unique challenges faced by marginalized communities, further entrenching existing disparities. Both case studies underscore the importance of considering DEI in the development and deployment of AI systems. As the DoD continues to integrate AI into its operations, this paper’s recommendations stress the necessity of continuous DEI assessment to ensure that AI serves as an asset rather than a liability. The authors recommend the following: 1. Data diversity & review 2. Continuous monitoring and calibration 3. Stakeholder engagement 4. Adoption of DEI requirements within Ethical AI Frameworks 5. Further research

Continuous and Unconstrained Tremor Monitoring in Parkinson's Disease Using Supervised Machine Learning and Wearable Sensors.

Accurately assessing the severity and frequency of fluctuating motor symptoms is important at all stages of Parkinson's disease management. Contrarily to time-consuming clinical testing or patient self-reporting with uncertain reliability, recordings with wearable sensors show promise as a tool for continuously and objectively assessing PD symptoms. While wearables-based clinical assessments during standardised and scripted tasks have been successfully implemented, assessments during unconstrained activity remain a challenge. We developed and implemented a supervised machine learning algorithm, trained and tested on tremor scores. We evaluated the algorithm on a 67-hour database comprising sensor data and clinical tremor scores for 24 Parkinson patients at four extremities for periods of about 3 hours. A random 25% subset of the labelled samples was used as test data, the remainder as training data. Based on features extracted from the sensor data, a Support Vector Machine was trained to predict tremor severity. Due to the inherent imbalance in tremor scores, we applied dataset rebalancing techniques. Our classifier demonstrated robust performance in detecting tremor events with a sensitivity of 0.90 on the test-portion of the resampled dataset. The overall classification accuracy was high at 0.88. We implemented an accurate classifier for tremor monitoring in free-living environments that can be trained even with modestly sized and imbalanced datasets. This advancement offers significant clinical value in continuously monitoring Parkinson's disease symptoms beyond the hospital setting, paving the way for personalized management of PD, timely therapeutic adjustments, and improved patient quality of life.

Training a Dataset Simulated Using RGB Images for an End-to-End Event-Based DoLP Recovery Network

Event cameras are bio-inspired neuromorphic sensors that have emerged in recent years, with advantages such as high temporal resolutions, high dynamic ranges, low latency, and low power consumption. Event cameras can be used to build event-based imaging polarimeters, overcoming the limited frame rates and low dynamic ranges of existing systems. Since events cannot provide absolute brightness intensity in different angles of polarization (AoPs), degree of linear polarization (DoLP) recovery in non-division-of-time (non-DoT) event-based imaging polarimeters is an ill-posed problem. Thus, we need a data-driven deep learning approach. Deep learning requires large amounts of data for training, and constructing a dataset for event-based non-DoT imaging polarimeters requires significant resources, scenarios, and time. We propose a method for generating datasets using simulated polarization distributions from existing red–green–blue images. Combined with event simulator V2E, the proposed method can easily construct large datasets for network training. We also propose an end-to-end event-based DoLP recovery network to solve the problem of DoLP recovery using event-based non-DoT imaging polarimeters. Finally, we construct a division-of-time event-based imaging polarimeter simulating an event-based four-channel non-DoT imaging polarimeter. Using real-world polarization events and DoLP ground truths, we demonstrate the effectiveness of the proposed simulation method and network.

Construction and optimization of non-parametric analysis model for meter coefficients via back propagation neural network

This study addresses the drawbacks of traditional methods used in meter coefficient analysis, which are low accuracy and long processing time. A new method based on non-parametric analysis using the Back Propagation (BP) neural network is proposed to overcome these limitations. The study explores the classification and pattern recognition capabilities of the BP neural network by analyzing its non-parametric model and optimization methods. For model construction, the study uses the United Kingdom Domestic Appliance-Level Electricity dataset’s meter readings and related data for training and testing the proposed model. The non-parametric analysis model is used for data pre-processing, feature extraction, and normalization to obtain the training and testing datasets. Experimental tests compare the proposed non-parametric analysis model based on the BP neural network with the traditional Least Squares Method (LSM). The results demonstrate that the proposed model significantly improves the accuracy indicators such as mean absolute error (MAE) and mean relative error (MRE) when compared with the LSM method. The proposed model achieves an MAE of 0.025 and an MRE of 1.32% in the testing dataset, while the LSM method has an MAE of 0.043 and an MRE of 2.56% in the same dataset. Therefore, the proposed non-parametric analysis model based on the BP neural network can achieve higher accuracy in meter coefficient analysis when compared with the traditional LSM method. This study provides a novel non-parametric analysis method with practical reference value for the electricity industry in energy metering and load forecasting.

A Deep Learning-Based Crop Disease Diagnosis Method Using Multimodal Mixup Augmentation

With the widespread adoption of smart farms and continuous advancements in IoT (Internet of Things) technology, acquiring diverse additional data has become increasingly convenient. Consequently, studies relevant to deep learning models that leverage multimodal data for crop disease diagnosis and associated data augmentation methods are significantly growing. We propose a comprehensive deep learning model that predicts crop type, detects disease presence, and assesses disease severity at the same time. We utilize multimodal data comprising crop images and environmental variables such as temperature, humidity, and dew points. We confirmed that the results of diagnosing crop diseases using multimodal data improved 2.58%p performance compared to using crop images only. We also propose a multimodal-based mixup augmentation method capable of utilizing both image and environmental data. In this study, multimodal data refer to data from multiple sources, and multimodal mixup is a data augmentation technique that combines multimodal data for training. This expands the conventional mixup technique that was originally applied solely to image data. Our multimodal mixup augmentation method showcases a performance improvement of 1.33%p compared to the original mixup method.

A Three-Zone Identification Method for Coal Mine Area Based on DS Evidence Theory

As coal ore and other resources are continuously mined, a three-zone structure is formed underground consisting of a sagging zone, fault zone, and caving zone. The use of well-logging data to identify the three zones is important for production safety and environmental management. Owing to the scarcity of data that can reflect three zones in normal coal mining, conventional identification and prediction methods face challenges when extracting data features, incurring a degree of uncertainty within prediction results. Accordingly, the accurate identification of the three zones has become a critical objective in daily production. To address this issue, we developed a method called a method called backpropagation neural networks with Dempster–Shafer (DS) evidence theory. Initially, we preprocessed the training data and deployed two backpropagation neural networks (BPNNs) to predict the three zones according to two parameters. According to these prediction results, the local and global credibility of each prediction is calculated and used to obtain the basic probability assignment function required for the DS evidence theory. Finally, the DS evidence theory is used to fuse the two BPNNs prediction results, thereby producing the final prediction results. The proposed method was demonstrated to improve prediction accuracy by 6.4% compared to a conventional neural network.

Sampling the user controls in neural modeling of audio devices

This work studies neural modeling of nonlinear parametric audio circuits, focusing on how the diversity of settings of the target device user controls seen during training affects network generalization. To study the problem, a large corpus of training datasets is synthetically generated using SPICE simulations of two distinct devices, an analog equalizer and an analog distortion pedal. A proven recurrent neural network architecture is trained using each dataset. The difference in the datasets is in the sampling resolution of the device user controls and in their overall size. Based on objective and subjective evaluation of the trained models, a sampling resolution of five for the device parameters is found to be sufficient to capture the behavior of the target systems for the types of devices considered during the study. This result is desirable, since a dense sampling grid can be impractical to realize in the general case when no automated way of setting the device parameters is available, while collecting large amounts of data using a sparse grid only incurs small additional costs. Thus, the result provides guidance for efficient collection of training data for neural modeling of other similar audio devices.