Subset Of Training Data Research Articles

1D CNN Based Detection and Localisation of Defective Droppers in Railway Catenary

Defective droppers pose a significant threat to the performance of the contact between the train pantograph and railway catenary. In this paper, the impact of damaged droppers on the performance of pantograph–catenary interaction behaviour is analysed, and the impact of varying degrees of damage to each dropper is labelled. To improve the classification accuracy when both the damage degree and position are considered, a model integrating multiple 1D CNNs is proposed. Approaches including randomly searching the optimal hyper-parameters and K-fold cross-validation are used to prevent overfitting and to ensure model performance regardless of the training data subset selected. Compared with a conventional 1D CNN, the classification performance of the integrated method is demonstrated using the metrics accuracy, F1-score, precision and recall. It is concluded that, through the use of the integrated 1D CNN, damaged droppers can be detected and localised based on the pantograph–catenary contact force. Hence, intelligent catenary inspection can be enhanced.

Spatial Prediction of Monthly Precipitation in Sulaimani Governorate using Artificial Neural Network Models

ANN modeling is used here to predict missing monthly precipitation data in one station of the eight weather stations network in Sulaimani Governorate. Eight models were developed, one for each station as for prediction. The accuracy of prediction obtain is excellent with correlation coefficients between the predicted and the measured values of monthly precipitation ranged from (90% to 97.2%). The eight ANN models are found after many trials for each station and those with the highest correlation coefficient were selected. All the ANN models are found to have a hyperbolic tangent and identity activation functions for the hidden and output layers respectively, with learning rate of (0.4) and momentum term of (0.9), but with different data set sub-division into training, testing and holdout data sub-sets, and different number of hidden nodes in the hidden layer. It is found that it is not necessary that the nearest station to the station under prediction has the highest effect; this may be attributed to the high differences in elevation between the stations. It can also found that the variance is not necessary has effect on the correlation coefficient obtained.

Thermal Bridges on Building Rooftops

Thermal Bridges on Building Rooftops (TBBR) is a multi-channel remote sensing dataset. It was recorded during six separate UAV fly-overs of the city center of Karlsruhe, Germany, and comprises a total of 926 high-resolution images with 6927 manually-provided thermal bridge annotations. Each image provides five channels: three color, one thermographic, and one computationally derived height map channel. The data is pre-split into training and test data subsets suitable for object detection and instance segmentation tasks. All data is organized and structured to comply with FAIR principles, i.e. being findable, accessible, interoperable, and reusable. It is publicly available and can be downloaded from the Zenodo data repository. This work provides a comprehensive data descriptor for the TBBR dataset to facilitate broad community uptake.

Continuous Renal Replacement Therapy during Extracorporeal Membrane Oxygenation: Circuit Haemodynamics and Circuit Failure

Introduction: Treatment with continuous renal replacement therapy (CRRT) is common during extracorporeal membrane oxygenation (ECMO). Such ECMO-CRRT has specific technical characteristics, which may affect circuit life. Accordingly, we studied CRRT haemodynamics and circuit life during ECMO. Methods: ECMO and non-ECMO-CRRT treatments in two adult intensive care units were compared using data collected over a 3-year period. A potential predictor of circuit survival identified in a 60% training data subset as a time-varying covariate within a Cox proportional hazard model was subsequently assessed in the complementary remaining data (40%). Results: Median [interquartile range] CRRT circuit life was greater when associated with ECMO (28.8 [14.0–65.2] vs. 20.2 [9.8–40.2] h, p < 0.0001). Access, return, prefilter, and effluent pressures were also greater during ECMO. Higher ECMO flows were associated with higher access and return pressures. Classification and regression tree analysis identified an association between high access pressures and accelerated circuit failure, while both first access pressures ≥190 mm Hg (HR 1.58 [1.09–2.30]) and patient weight (HR 1.85 [1.15–2.97] third tertile vs. first tertile) were independently associated with circuit failure in a multivariable Cox model. Access dysfunction was associated with a stepwise increase in transfilter pressure, suggesting a potential mechanism of membrane injury. Conclusion: CRRT circuits used in conjunction with ECMO have a longer circuit life than usual CRRT despite exposure to higher circuit pressures. Markedly elevated access pressures, however, may predict early CRRT circuit failure during ECMO, possibly via progressive membrane thrombosis as evidenced by increased transfilter pressure gradients.

Voting-Based Deep Convolutional Neural Networks (VB-DCNNs) for M-QAM and M-PSK Signals Classification

Automatic modulation classification (AMC) using convolutional neural networks (CNNs) is an active area of research that has the potential to improve the efficiency and reliability of wireless communication systems significantly. AMC is the approach used in a communication system to detect the type of modulation format at the receiver end. This paper proposes a voting-based deep convolutional neural network (VB-DCNN) for classifying M-QAM and M-PSK signals. M-QAM and M-PSK signal waveforms are generated and passed through the fading channel in the presence of additive white Gaussian noise (AWGN). The VB-DCNN extracts features from the input signal through convolutional layers, and classification is performed on these features. Multiple network instances are trained on different subsets of training data in the VB-DCNN. A network instance predicts the input signal during testing. Based on the votes, the final prediction is made. Different simulation experiments are carried out to analyze the performance of the trained network, and the DCNN is designed with the Deep Neural Network Toolbox in MATLAB. The generated frames are divided into training, validation, and test datasets. Lastly, the classification accuracy of the trained network is determined using test frames. The proposed model’s accuracy is near to 100% at lower SNRs. The simulation results show the superiority of the proposed VB-DCNN compared to existing state-of-the-art techniques.

A structure-protecting kernelized semi-supervised space adjustment for classification

We introduce a semi-supervised space adjustment framework in this paper. In the introduced framework, the dataset contains two subsets: (a) training data subset (space-one data (SOD)) and (b) testing data subset (space-two data (STD)). Our semi-supervised space adjustment framework learns under three assumptions: (I) it is assumed that all data points in the SOD are labeled, and only a minority of the data points in the STD are labeled (we call the labeled space-two data as LSTD), (II) the size of LSTD is very small comparing to the size of SOD, and (III) it is also assumed that the data of SOD and the data of STD have different distributions. We denote the unlabeled space-two data by ULSTD, which is equal to STD - LSTD. The aim is to map the training data, i.e., the data from the training labeled data subset and those from LSTD (note that all labeled data are considered to be training data, i.e., SOD ∪ LSTD) into a shared space (ShS). The mapped SOD, ULSTD, and LSTD into ShS are named MSOD, MULSTD, and MLSTD, respectively. The proposed method does the mentioned mapping in such a way that structures of the data points in SOD and MSOD, in STD and MSTD, in ULSTD and MULSTD, and in LSTD and MLSTD are the same. In the proposed method, the mapping is proposed to be done by a principal component analysis transformation on kernelized data. In the proposed method, it is tried to find a mapping that (a) can maintain the neighbors of data points after the mapping and (b) can take advantage of the class labels that are known in STD during transformation. After that, we represent and formulate the problem of finding the optimal mapping into a non-linear objective function. To solve it, we transform it into a semidefinite programming (SDP) problem. We solve the optimization problem with an SDP solver. The examinations indicate the superiority of the learners trained in the data mapped by the proposed approach to the learners trained in the data mapped by the state of the art methods.

Zero-Shot Machine Unlearning

Modern privacy regulations grant citizens the right to be forgotten by products, services and companies. In case of machine learning (ML) applications, this necessitates deletion of data not only from storage archives but also from ML models. Due to an increasing need for regulatory compliance required for ML applications, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">machine unlearning</i> is becoming an emerging research problem. The right to be forgotten requests come in the form of removal of a certain set or class of data from the already trained ML model. Practical considerations preclude retraining of the model from scratch after discarding the deleted data. The few existing studies use either the whole training data, or a subset of training data, or some metadata stored during training to update the model weights for unlearning. However, strict regulatory compliance requires time-bound deletion of data. Thus, in many cases, no data related to the training process or training samples may be accessible even for the unlearning purpose. We therefore ask the question: <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">is it possible to achieve unlearning with zero training samples?</i> In this paper, we introduce the novel problem of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">zero-shot machine unlearning</i> that caters for the extreme but practical scenario where zero original data samples are available for use. We then propose two novel solutions for <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">zero-shot machine unlearning</i> based on (a) error minimizing-maximizing noise and (b) gated knowledge transfer. These methods remove the information of the forget data from the model while maintaining the model efficacy on the retain data. The zero-shot approach offers good protection against the model inversion attacks and membership inference attacks. We introduce a new evaluation metric, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Anamnesis Index</i> (AIN) to effectively measure the quality of the unlearning method. The experiments show promising results for unlearning in deep learning models on benchmark vision data-sets. The source code is available here: https://github.com/ayu987/zero-shot-unlearning.

Efficient design of peptide-binding polymers using active learning approaches.

Active learning (AL) has become a subject of active recent research both in industry and academia as an efficient approach for rapid design and discovery of novel chemicals, materials, and polymers. Herein, we have assessed the applicability of AL for the discovery of polymeric micelle formulations for poorly soluble drugs. We were motivated by the key advantages of this approach making it a desirable strategy for rational design of drug delivery systems due toto its ability to (i) employ relatively small datasets for model development, (ii) iterate between model development and model assessment using small external datasets that can be either generated in focused experimental studies or formed from subsets of the initial training data, and (iii) progressively evolve models towards increasingly more reliable predictions and the identification of novel chemicals with the desired properties. In this study, we compared various AL protocols for their effectiveness in finding biologically active molecules using synthetic datasets. We have investigated the dependency of AL performance on the size of the initial training set, the relative complexity of the task, and the choice of the initial training dataset. We found that AL techniques as applied to regression modeling offer no benefits over random search, while AL used for classification tasks performs better than models built for randomly selected training sets but still quite far from perfect. Using the best performing AL protocol,. Finally, the best performing AL approach was employed to discover and experimentally validate novel binding polymers for a case study of asialoglycoprotein receptor (ASGPR).

Hierarchical, multiclass prediction of Alzheimer’s clinical diagnosis using imputed, multimodal NACC data

AbstractBackgroundEarly detection and diagnosis are critical for effective treatment of Alzheimer’s disease (AD). Statistical machine learning holds promise for disease prediction using large AD repositories, such as the National Alzheimer’s Coordinating Center (NACC) dataset. Past work has extensively explored binary classifications of AD diagnosis. Here, we extended this to address two major, remaining challenges: data imputation and multiclass classification (e.g., AD vs Mild Cognitive Impairment, MCI vs. Healthy Cognition, HC).MethodFor imputation, we tested single‐value (baseline), k‐NN, and Bayesian imputation on a subset of complete training data with artificially removed values (completely at random). Per modality of interest, a systematic search was conducted for the best input feature set to impute the missing values. The mean absolute deviation and mismatch rate between the imputed and true values were used to evaluate the methods on a held‐out test set. For classification, we then implemented both “flat” multiclass and hierarchical classification, an extension of multiclass classification that organizes classes into a tree, to predict clinical diagnosis using multimodal inputs. Several pipelines with combinations of imputation methods, classifiers, and input modalities were tested on different hierarchical classification strategies, e.g., HC vs [AD vs MCI], compared with the flat, multiclass baseline (AD vs MCI vs HC).ResultCompared to the baseline methods for both imputation and classification, the proposed alternatives performed better, as measured by the predetermined imputation and classification metrics evaluated on the test set. For imputation, behavioral and neuropsychiatric features were more easily and accurately imputed while ApoE genotype was more difficult to impute. For classification, using a combination of classifiers and input features in the hierarchy outperformed the baseline method and support the notion that select biomarkers are better able to characterize AD subpopulations within the hierarchy.ConclusionEstablishing a systematic imputation method tailored to AD biomarker modalities fills a critical gap in achieving more robust disease prediction. Building a hierarchical classifier parallels working through a differential diagnosis for AD, which offers greater translational value than binary classification of AD. In combination, these results form an important step toward developing reliable, holistic classifiers of AD disease status.

RoS-KD: A Robust Stochastic Knowledge Distillation Approach for Noisy Medical Imaging.

AI-powered Medical Imaging has recently achieved enormous attention due to its ability to provide fast-paced healthcare diagnoses. However, it usually suffers from a lack of high-quality datasets due to high annotation cost, inter-observer variability, human annotator error, and errors in computer-generated labels. Deep learning models trained on noisy labelled datasets are sensitive to the noise type and lead to less generalization on the unseen samples. To address this challenge, we propose a Robust Stochastic Knowledge Distillation (RoS-KD) framework which mimics the notion of learning a topic from multiple sources to ensure deterrence in learning noisy information. More specifically, RoS-KD learns a smooth, well-informed, and robust student manifold by distilling knowledge from multiple teachers trained on overlapping subsets of training data. Our extensive experiments on popular medical imaging classification tasks (cardiopulmonary disease and lesion classification) using real-world datasets, show the performance benefit of RoS-KD, its ability to distill knowledge from many popular large networks (ResNet-50, DenseNet-121, MobileNet-V2) in a comparatively small network, and its robustness to adversarial attacks (PGD, FSGM). More specifically, RoS-KD achieves > 2% and > 4% improvement on F1-score for lesion classification and cardiopulmonary disease classification tasks, respectively, when the underlying student is ResNet-18 against recent competitive knowledge distillation baseline. Additionally, on cardiopulmonary disease classification task, RoS-KD outperforms most of the SOTA baselines by ~1% gain in AUC score.

Model Soups for Various Training and Validation Data

Model soups synthesize multiple models after fine-tuning them with different hyperparameters based on the accuracy of the validation data. They train different models on the same training and validation data sets. In this study, we maximized the model fine-tuning accuracy using the inference time and memory cost of a single model. We extended the model soups to create subsets of k training and validation data using a method similar to k-fold cross-validation and trained models on these subsets. First, we showed the correlation between the validation and test data when the models are synthesized, such that their training data contain validation data. Thereafter, we showed that synthesizing k of these models, after synthesizing models based on subsets of the same training and validation data, provides a single model with high test accuracy. This study provides a method for learning models with both high accuracy and reliability for small datasets such as medical images.

Search-based cost-sensitive hypergraph learning for anomaly detection

Detecting anomaly plays a vital role in ensuring the reliability and safety of many applications. However, this task is challenging due to the unclear correlation underneath the testing data and the class imbalanced issue between anomalous and normal data. Moreover, it is noted that the misclassification of anomalous and normal data is usually associated with different costs in real applications. To solve these issues, we propose a search-based cost-sensitive hypergraph learning method (SCSHL) for anomaly detection. More specifically, to deal with the imbalanced issue, we first construct a target-specific subset of training data according to the testing samples and conduct a cost-sensitive feature selection method to select the effective features. To explore the unclear high-order correlation underneath the data, we employ the hypergraph structure to formulate the complex relationship among the testing data and further combine the cost information into hypergraph learning and conduct label propagation on the cost-sensitive hypergraph structure. To evaluate the effectiveness of the proposed method, we have conducted experiments on industry anomaly detection datasets, software defect prediction (SDP) datasets and outlier detection datasets (ODDS). Experimental results and comparisons with state-of-the-art methods have shown the superiority of our method.

Training Data Subset Search With Ensemble Active Learning

Deep Neural Networks (DNNs) often rely on vast datasets for training. Given the large size of such datasets, it is conceivable that they contain specific samples that either do not contribute or negatively impact the DNN’s optimization. Modifying the training distribution to exclude such samples could provide an effective solution to improve performance and reduce training time. This paper proposes to scale up ensemble Active Learning (AL) methods to perform acquisition at a large scale (10k to 500k samples at a time). We do this with ensembles of hundreds of models, obtained at a minimal computational cost by reusing intermediate training checkpoints. This allows us to automatically and efficiently perform a training data subset search for large labeled datasets. We observe that our approach obtains favorable subsets of training data, which can be used to train more accurate DNNs than training with the entire dataset. We perform an extensive experimental study of this phenomenon on three image classification benchmarks (CIFAR-10, CIFAR-100, and ImageNet), as well as an internal object detection benchmark for prototyping perception models for autonomous driving. Unlike existing studies, our experiments on object detection are at the scale required for production-ready autonomous driving systems. We provide insights on the impact of different initialization schemes, acquisition functions, and ensemble configurations at this scale. Our results provide strong empirical evidence that optimizing the training data distribution can significantly benefit large-scale vision tasks.

Forest Fire Susceptibility Assessment and Mapping Using Support Vector Regression and Adaptive Neuro-Fuzzy Inference System-Based Evolutionary Algorithms

Support vector regression (SVR) and the adaptive neuro-fuzzy inference system (ANFIS) are two well-known and powerful artificial intelligence techniques which have been frequently used for hazard mapping. So far, a plethora of hybrid models have been developed using a combination of either the SVR or ANFIS and evolutionary algorithms, but there are only a handful of studies that compare the performance of these models when integrated with evolutionary algorithms, especially in forest fire susceptibility mapping (FFSM). The aim of this study was to compare performance of ANFIS-, and SVR-based evolutionary algorithms, namely, the genetic algorithm (GA) and the shuffled frog-leaping algorithm (SFLA) in FFSM in Ajloun Governorate in Jordan. Accordingly, four hybrid models, SVR-GA, SVR-SFLA, ANFIS-GA, and ANFIS-SFLA, were developed and compared. One hundred and one forest fire locations were used in this study to assess and model susceptibility of forests to fires. The forest fire inventory data were divided into a training data subset (70%) and a testing data subset (30%). Fourteen factors affecting incidence of forest fires were employed as conditioning factors. The area under the receiver operating characteristic (AUROC) curve was used to assess performance of the models in the validation phase. The results revealed that the SVR-based hybrid algorithms had better AUROC values than the ANFIS-based algorithms. Of the four integrated models, the SVR-GA model proved to be the model with the highest accuracy and best performance. It had AUROC values of 0.97 and 0.89 in the training and the testing phases, respectively.

Optical Remote Sensing Image Understanding With Weak Supervision: Concepts, methods, and perspectives

In recent years, supervised learning has been widely used in various tasks of optical remote sensing image understanding, including remote sensing image classification, pixel-wise segmentation, change detection, and object detection. The methods based on supervised learning need a large amount of high-quality training data and their performance highly depends on the quality of the labels. However, in practical remote sensing applications, it is often expensive and time-consuming to obtain large-scale data sets with high-quality labels, which leads to a lack of sufficient supervised information. In some cases, only coarse-grained labels can be obtained, resulting in the lack of exact supervision. In addition, the supervised information obtained manually may be wrong, resulting in a lack of accurate supervision. Therefore, remote sensing image understanding often faces the problems of incomplete, inexact, and inaccurate supervised information, which will affect the breadth and depth of remote sensing applications. In order to solve the above-mentioned problems, researchers have explored various tasks in remote sensing image understanding under weak supervision. This paper summarizes the research progress of weakly supervised learning in the field of remote sensing, including three typical weakly supervised paradigms: 1) Incomplete supervision, where only a subset of training data is labeled; 2) Inexact supervision, where only coarse-grained labels of training data are given; 3) Inaccurate supervision, where the labels given are not always true on the ground.

Do Machine Learning Approaches Offer Skill Improvement for Short-Term Forecasting of Wind Gust Occurrence and Magnitude?

Abstract Wind gusts, and in particular intense gusts, are societally relevant but extremely challenging to forecast. This study systematically assesses the skill enhancement that can be achieved using artificial neural networks (ANNs) for forecasting of wind gust occurrence and magnitude. Geophysical predictors from the ERA5 reanalysis are used in conjunction with an autoregressive term in regression and ANN models with different predictors, and varying model complexity. Models are derived and assessed for the warm (April–September) and cold (October–March) seasons for three high passenger volume airports in the United States. Model uncertainty is assessed by deriving models for 1000 different randomly selected training (70%) and testing (30%) subsets. Gust prediction fidelity in independent test samples is critically dependent on inclusion of an autoregressive term. Gust occurrence probabilities derived using five-layer ANNs exhibit consistently higher fidelity than those from regression models and shallower ANNs. Inclusion of the autoregressive term and increasing the number of hidden layers in ANNs from 1 to 5 also improve the model performance for gust magnitudes (lower RMSE, increased correlation, and model standard deviations that more closely approximate observed values). Deeper ANNs (e.g., 20 hidden layers) exhibit higher skill in forecasting strong (17–25.7 m s−1) and damaging (≥25.7 m s−1) wind gusts. However, such deep networks exhibit evidence of overfitting and still substantially underestimate (by 50%) the frequency of strong and damaging wind gusts at the three airports considered herein. Significance Statement Improved short-term forecasting of wind gusts will enhance aviation safety and logistics and may offer other societal benefits. Here we present a rigorous investigation of the relative skill of models of wind gust occurrence and magnitude that employ different statistical methods. It is shown that artificial neural networks (ANNs) offer considerable skill enhancement over regression methods, particularly for strong and damaging wind gusts. For wind gust magnitudes in particular, application of deeper learning networks (e.g., five or more hidden layers) offers tangible improvements in forecast accuracy. However, deeper networks are vulnerable to overfitting and exhibit substantial variability with the specific training and testing data subset used. Also, even deep ANNs reproduce only half of strong and damaging wind gusts. These results indicate the need for future work to elucidate the dynamical mechanisms of intense wind gusts and advance solutions to their prediction.

Surrogate-assisted evolutionary optimization of expensive many-objective irregular problems

Surrogate-assisted evolutionary algorithms are one effective approach to handling expensive problems and have attracted increasing attention over the past decades. However, existing surrogate-assisted evolutionary algorithms pay little attention to expensive many-objective problems with irregular Pareto fronts, also called irregular problems. In this study, we propose a surrogate-assisted evolutionary algorithm for dealing with expensive irregular problems, where only a small number of expensive fitness evaluations is allowed. In the proposed algorithm, the reference vectors are adapted based on both the individuals in the current population and the non-dominated solutions that have been evaluated using the real objective functions. A surrogate management strategy is then designed to balance convergence and diversity according to the adaptive reference vectors as well as the non-dominated solutions that have been evaluated using the expensive objective functions so that the irregularity of the Pareto front can be taken into account. To reduce the computational cost for updating the Gaussian process based surrogates, a subset of training data near the adaptive reference vectors are prioritized. Experimental results on the DTLZ, WFG, DPF and MaF test suites demonstrate that the proposed algorithm is able to solve expensive many-objective optimization problems with both irregular and regular Pareto fronts. The proposed algorithm is also tested on a real-world application example to further confirm its effectiveness and competitiveness.

A Novel Selective Ensemble Learning Method for Smartphone Sensor-Based Human Activity Recognition Based on Hybrid Diversity Enhancement and Improved Binary Glowworm Swarm Optimization

Human activity recognition (HAR) is gaining interest with many important applications including ubiquitous computing, health-care services and detection of diseases. Smartphone sensors have high acceptance and adherence in daily life and they provide an alternative and economic way for activity recognition. To improve the performance of smartphone sensor-based HAR, a novel smartphone sensor-based HAR method (hybrid diversity enhancement with selective ensemble learning, HDESEN) that utilizes selective ensemble learning with differentiated extreme learning machines (ELMs) is proposed, where hybrid diversity enhancement is proposed to boost the diversity of base models and an improved binary glowworm swarm optimization (IBGSO) is employed to effectively enhance the learning process by choosing a superior subset for ensemble instead of all. Firstly, statistical features in the time domain and frequency domain are extracted and integrated from smartphone sensors and then three filter-based feature selection methods are utilized for desirable base models. Secondly, to enhance the diversity of the base models, three types of diversities are introduced to construct different base models, respectively. Among them, Bootstrap is introduced to design distinctive training data subsets for differential base models, random subspace and optimized subspace are proposed to obtain different feature spaces for constructing base models. Thirdly, a pruning method based on glowworm swarm optimization (GSO) is proposed to find the optimal sub-ensemble from the pool of models from all diverse types to implement selective ensemble learning. The experimental results on tow publicly available datasets (UCI-HAR and WISDM) demonstrate the proposed HDESEN can reliably improve the performance of HAR and outperforms the relevant state-of-the-art approaches.

Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging

Medical imaging is a central part of clinical diagnosis and treatment guidance. Machine learning has increasingly gained relevance because it captures features of disease and treatment response that are relevant for therapeutic decision-making. In clinical practice, the continuous progress of image acquisition technology or diagnostic procedures, the diversity of scanners, and evolving imaging protocols hamper the utility of machine learning, as prediction accuracy on new data deteriorates, or models become outdated due to these domain shifts. We propose a continual learning approach to deal with such domain shifts occurring at unknown time points. We adapt models to emerging variations in a continuous data stream while counteracting catastrophic forgetting. A dynamic memory enables rehearsal on a subset of diverse training data to mitigate forgetting while enabling models to expand to new domains. The technique balances memory by detecting pseudo-domains, representing different style clusters within the data stream. Evaluation of two different tasks, cardiac segmentation in magnetic resonance imaging and lung nodule detection in computed tomography, demonstrate a consistent advantage of the method.

Sparse data-based image super-resolution with ANFIS interpolation

Image processing is a very broad field containing various areas, including image super-resolution (ISR) which re-represents a low-resolution image as a high-resolution one through a certain means of image transformation. The problem with most of the existing ISR methods is that they are devised for the condition in which sufficient training data is expected to be available. This article proposes a new approach for sparse data-based (rather than sufficient training data-based) ISR, by the use of an ANFIS (Adaptive Network-based Fuzzy Inference System) interpolation technique. Particularly, a set of given image training data is split into various subsets of sufficient and sparse training data subsets. Typical ANFIS training process is applied for those subsets involving sufficient data, and ANFIS interpolation is employed for the rest that contains sparse data only. Inadequate work is available in the current literature for the sparse data-based ISR. Consequently, the implementations of the proposed sparse data-based approach, for both training and testing processes, are compared with the state-of-the-art sufficient data-based ISR methods. This is of course very challenging, but the results of experimental evaluation demonstrate positively about the efficacy of the work presented herein.