Search

Hard landing is a typical flight safety incident, and interpretability plays an important role in flight safety research. However, existing studies failed to provide good interpretability of the reasons for hard landing incidents and suffer from low prediction accuracy. To address the above problems, in this paper we propose CurveCluster+, a curve clustering method based on quick access recorder (QAR) data for hard landing risk evaluation. Specifically, we first conduct an in-depth analysis on hard landing flights by comparing key QAR parameter curves with the group behavior, based on which we establish a two-level hierarchical classification of hard landing incidents according to the hard landing patterns. Then we extract curve-level features from key QAR parameters through interpolation and resampling. After that we turn the classic K-means clustering into a semi-supervised algorithm by incorporating some expert experience and apply it on the curve-level features to automatically recognize the hard landing patterns. Finally, we propose a risk evaluation model based on the clustering results to discover high-risk flights from normal ones. We evaluate our method on a QAR dataset of 37,943 Airbus 320 aircraft flights. The results show that compared with other state-of-the-art data-driven methods, CurveCluster+ provides strong interpretability of hard landing incidents and exhibits good performance in recognizing hard landing patterns (the overall accuracy of our method reaches up to 92.99%). Moreover, it only requires a handful of hard landing samples to discover high-risk flights from tremendous normal landing flights, which is critical for flight safety warnings.

Road traffic state estimation is an essential component of intelligent transportation systems (ITSs). However, road traffic state data collected by traffic detectors are often incomplete, which can cause problems across a variety of transportation applications, such as traffic state prediction and pattern recognition. We present GA-GAN (Graph Aggregate Generative Adversarial Network), consisting of graph sample and aggregate (GraphSAGE) and a generative adversarial network (GAN), to impute missing road traffic state data. Instead of using the original road network structure, which presents the spatial information to process a graph operation, we reconstruct the road network according to the correlation coefficients of road historical data. We utilize GraphSAGE to aggregate the temporal-spatial information from the neighbors of each road in the reconstructed road network. GAN is used to generate complete traffic state data from the extracted temporal-spatial information to achieve traffic state data imputation. To illustrate the efficient performance of the model, experiments are conducted on traffic data collected from California and Seattle, Washington, showing that the proposed model outperforms state-of-the-art methods.

Attribute reduction, also called feature selection, is one of the most important issues of rough set theory, which is regarded as a vital preprocessing step in pattern recognition, machine learning, and data mining. Nowadays, high-dimensional mixed and incomplete data sets are very common in real-world applications. Certainly, the selection of a promising feature subset from such data sets is a very interesting, but challenging problem. Almost all of the existing methods generated a cover on the space of objects to determine important features. However, some tolerance classes in the cover are useless for the computational process. Thus, this article introduces a new concept of stripped neighborhood covers to reduce unnecessary tolerance classes from the original cover. Based on the proposed stripped neighborhood cover, we define a new reduct in mixed and incomplete decision tables, and then design an efficient heuristic algorithm to find this reduct. For each loop in the main loop of the proposed algorithm, we use an error measure to select an optimal feature and put it into the selected feature subset. Besides, to deal more efficiently with high-dimensional data sets, we also determine redundant features after each loop and remove them from the candidate feature subset. For the purpose of verifying the performance of the proposed algorithm, we carry out experiments on data sets downloaded from public data sources to compare with existing state-of-the-art algorithms. Experimental results showed that our algorithm outperforms compared algorithms, especially in classification accuracy.

In vehicle operation, in order to maximize the fuel economy, the propulsion system control can easily adapt to pressure and temperature variations as these variations can be measured by sensors. However, it is challenging to detect driving cycles. With growing progress made in the artificial intelligence field, pattern recognition gains momentum in various applications. This study presents a study on driving cycle pattern recognition based on supervised learning. Training data 2-D visualization is achieved by t-SNE algorithm. Ten out of twelve supervised learning algorithms predict driving conditions with the accuracy at 88% or higher and Extra Tree algorithm leads the recognition accuracy at 90.26%. To improve recognition accuracy, two hierarchical frameworks are proposed by integrating multiple supervised learning methods using weighted average and vote methods. The two hierarchical frameworks boost the driving condition recognition accuracy from 90.26% to 90.43% (weighted average) and 91.76% (vote). The results are further validated in holdout test. Additionally, a plug-in hybrid electric vehicle simulation shows 3.88%-5.82% fuel economy improvement when compared to baseline method. Two hierarchical methods outperform the Extra Tree method by 2% fuel economy. In summary, supervised learning shows great potential to detect driving cycles for vehicle energy saving.

Fish classification (FC) is an expansively studied problem in the domains of image segmentation, pattern recognition, and information retrieval. It has been applied in a countless number of domains including target marketing. Meanwhile, governments are obliged to maintain the fish supply and balance between the ecosystem, commercial, agriculture field, marine scientists, and industrial arena of fish including the nutrition and canning factories. The various FC techniques performance is compared relying on the availability of preprocessing and feature extraction methods, the number of extracted features and classification accuracy, the number of fish families/species recognized. This survey also reviewed the use of Databases such as Fish4-Knowledge (F4K), knowledge database, and Global Information System (GIS) on Fishes and other FC databases. The study on preprocessing methods features extraction techniques and classifiers are gathered from recent works to enhance the understanding of the characteristics of preprocessing methods, features extraction techniques, and classifiers to guide future research directions and compensate for current research gaps.

In recent years, researchers have made a great success on the automatic classification and detection of malware utilizing machine learning methods. However, most machine learning based approaches over rely on the training samples such that a new malware family not belonging to the training set cannot be identified. To address such issue, we propose a soft relevance value (s-value), a new evaluating way of feature soft relevance that uses the mixed distance criterion to assess classified results. Specifically, we leverage the mixed distance criterion from pattern recognition to distinguish testing samples as a new family which is not labeled in training set. Finally, we evaluate how s-value can be used to distinguish and classify a new malware family with the malware datasets from the Research Prediction Competition of Microsoft Malware Classification Challenge and Windows (Kaggle). The experimental results show that, the train-ing time is approximately 12 hours, while the prediction time is only ∼0.5 second. Comparing against the Kaggle winner, our time costs for training and pprediction only occupy 16.7% and 3.8% of the winner.s time costs, respectively. The accuracy of classifying malware reaches 99.8%. Such results indicates that our proposed s-value achieves a balance in accuracy, training and prediction time, and outperforms the state-of-the-art machine learning based malware detection approaches. Besides, our method is able to identify new malware families that are not included in the training set.

Feature selection is important in both machine learning and pattern recognition. Successfully selecting informative features can significantly increase learning accuracy and improve result comprehensibility. Various methods have been proposed to identify informative features from high-dimensional data by removing redundant and irrelevant features to improve classification accuracy. In this article, we systematically survey existing sparse learning models for feature selection from the perspectives of individual sparse feature selection and group sparse feature selection, and analyze the differences and connections among various sparse learning models. Promising research directions and topics on sparse learning models are analyzed.

Fishing is a crucial worldwide activity as it provides a source of food and economic income. A challenge in ecology and conservation is decreasing overfishing and illegal, unreported, and unregulated fishing (IUUF). One strategy to decrease those issues is to track vessels for detecting fishing behaviors through monitory systems. In this letter, we present an approach to classify fishing behaviors, specifically, for four fishing gear types (trawl, purse seine, fixed gear, and longline) using automatic identification systems (AISs) data from the Global Fishing Watch platform. Thus, our main contribution is how we propose data processing by including a supervised autoencoder dimensional reduction (SA-DR) processing data step. This step allows removing redundant features and noise, avoiding overfitting, decreasing data complexity, and preserving the differences between classes. Specifically, we propose to use IVIS and centroid encoder (CE) methods. The experimental results show how our approach applying SA-DR over the vessel trajectory feature representation reduces the variation results among different classifiers and achieves a high classification accuracy of up to 95%. This result could help prevent IUUF, overfishing, and improve fishery management strategies.

Detecting the evolution of large-area landscape patterns using long-term remote-sensing images is helpful in supporting research on the relationship between landscape patterns and ecological processes, as well as the development of ecological process simulations and spatiotemporal interaction models. However, detection methods have generally been developed as separate applications, each with a separate type of landscape pattern change; remote-sensing images are acquired at epochal timesteps. Consequently, in practical applications, many omission changes for some types of pattern changes and inaccurate evolution time are presented in the detected map. In this article, state-and-evolution detection models (SEDMs) are promoted to obtain complete information about the evolution of landscape patterns based on yearly land cover data. In the proposed framework, we first define the major categories of landscape pattern changes to comprehensively reveal the characteristics of landscape pattern changes associated with real change cases. Next, a morphological rule-based pattern recognition approach is proposed for quantitative discrimination among these categories. This approach is then applied in annual land cover data to continuously detect landscape pattern evolution processes and evolution time. Finally, the detected evolution time in different evolution processes is applied to measure the timestep between two disparate types. The performances of the SEDMs are presented by Landsat-derived land cover evolution in Shanxi, China. The detected results are indirectly verified by the land cover conversion matrix and connect index, indicating strong robustness and generalization ability of the SEDMs.

The Internet of Things (IoT) is becoming an indispensable part of everyday life, enabling a variety of emerging services and applications. However, the presence of rogue IoT devices has exposed the IoT to untold risks with severe consequences. The first step in securing the IoT is detecting rogue IoT devices and identifying legitimate ones. Conventional approaches use cryptographic mechanisms to authenticate and verify legitimate devices’ identities. However, cryptographic protocols are not available in many systems. Meanwhile, these methods are less effective when legitimate devices can be exploited or encryption keys are disclosed. Therefore, non-cryptographic IoT device identification and rogue device detection become efficient solutions to secure existing systems and will provide additional protection to systems with cryptographic protocols. Non-cryptographic approaches require more effort and are not yet adequately investigated. In this paper, we provide a comprehensive survey on machine learning technologies for the identification of IoT devices along with the detection of compromised or falsified ones from the viewpoint of passive surveillance agents or network operators. We classify the IoT device identification and detection into four categories: device-specific pattern recognition, Deep Learning enabled device identification, unsupervised device identification, and abnormal device detection. Meanwhile, we discuss various ML-related enabling technologies for this purpose. These enabling technologies include learning algorithms, feature engineering on network traffic traces and wireless signals, incremental learning, and abnormality detection.

Identifying the key risk factors of disease from a large amount of clinical data is a prerequisite for further scientific decision-making. In medical practice, the clinical symptom information of patients usually includes various types of data. Meanwhile, the occurrence and development of diseases are joint result of the mutual influence factors. Therefore, there is usually a correlation between attributes. In this paper, we discuss a kind of hybrid attribute feature selection problem considering the correlation between attributes. Firstly, we take the identification of disease pathogenic factors in medical decision as the background, and construct a hybrid attribute decision system. Secondly, by introducing kernel alignment, the uncertain relationship between attributes is defined. Based on this, a three-way clustering model in attribute space is established. Furthermore, a feature selection method for hybrid attribute data based on three-way clustering in attribute space is proposed. Finally, we applied the proposed model to identify the pathogenic factors of stroke and used 279 clinical random samples for simulation analysis. The results verified the applicability and validity of the model. The main contributions of this paper include two aspects. In terms of theory, by introducing kernel alignment, a three-way clustering algorithm in attribute space is established. Meanwhile, a hybrid attribute feature selection method based on three-way clustering is proposed. In terms of application, the proposed method is applied to identify risk factors of stroke.

The grading of gemstones is currently a manual procedure performed by gemologists. A popular approach uses reference stones, where those are visually inspected by specialists that decide which one of the available reference stone is the most similar to the inspected stone. This procedure is very subjective as different specialists may end up with different grading choices. This work proposes a complete framework that entails the image acquisition and goes up to the final stone categorization. The proposal is able to automate the entire process apart from including the stone in the created chamber for the image acquisition. It discards the subjective decisions made by specialists. This is the first work to propose a machine learning approach coupled with image processing techniques for emerald grading. The proposed framework achieves 98% of accuracy (correctly categorized stones), outperforming a deep learning approach. Furthermore, we also create and publish the used dataset that contains 192 images of emerald stones along with their extracted and pre-processed features.

Deep convolutional neural networks (DCNNs) have dominated as the best performers on almost all computer vision tasks over the past several years. However, it remains a major challenge to deploy these powerful DCNNs in resource-limited environments, such as embedded devices and smartphones. To this end, 1-bit CNNs have emerged as a feasible solution as they are much more resource-efficient. Unfortunately, they often suffer from a significant performance drop compared to their full-precision counterparts. In this paper, we propose a novel Bayesian Optimized compact 1-bit CNNs (BONNs) model, which has the advantage of Bayesian learning, to improve the performance of 1-bit CNNs significantly. BONNs incorporate the prior distributions of full-precision kernels, features, and filters into a Bayesian framework to construct 1-bit CNNs in a comprehensive end-to-end manner. The proposed Bayesian learning algorithms are well-founded and used to optimize the network simultaneously in different kernels, features, and filters, which largely improves the compactness and capacity of 1-bit CNNs. We further introduce a new Bayesian learning-based pruning method for 1-bit CNNs, which significantly increases the model efficiency with very competitive performance. This enables our method to be used in a variety of practical scenarios. Extensive experiments on the ImageNet, CIFAR, and LFW datasets show that BONNs achieve the best in classification performance compared to a variety of state-of-the-art 1-bit CNN models. In particular, BONN achieves a strong generalization performance on the object detection task.

How to represent and classify a testing sample for the representation-based classification (RBC) plays an important role in the filed of pattern recognition. As a typical kind of the representation-based classification with promising performance, collaborative representation-based classification (CRC) adopts all the training samples to collaboratively represent and then classify each testing sample with the reconstructive residuals among all the classes. However, most of the CRC methods fail to make full use of the localities and discrimination information of data in collaborative representation. To address this issue to further improve the classification performance, we design a novel supervised CRC method entitled locality-constrained weighted collaborative-competitive representation-based classification (LWCCRC). In the proposed method, the localities of data are taken into account by using the positive and negative nearest samples of each testing sample with their corresponding weighted constraints. Such devised locality-constrained weighted term can model the similarity and natural discrimination information contained in the neighborhood region for each testing sample to obtain the favorable representation. Moreover, a competitive constraint is introduced to enhance pattern discrimination among the categorical collaborative representations. To explore the effectiveness of our proposed LWCCRC, the extensive experiments are carried out on three different types of data sets. The experimental results demonstrate that the proposed LWCCRC significantly outperforms the recent state-of-the-art CRC methods.

In this paper, we propose to exploit the side-tuning framework for multimodal document classification. Side-tuning is a methodology for network adaptation recently introduced to solve some of the problems related to previous approaches. Thanks to this technique it is actually possible to overcome model rigidity and catastrophic forgetting of transfer learning by fine-tuning. The proposed solution uses off-the-shelf deep learning architectures leveraging the side-tuning framework to combine a base model with a tandem of two side networks. We show that side-tuning can be successfully employed also when different data sources are considered, e.g. text and images in document classification. The experimental results show that this approach pushes further the limit for document classification accuracy with respect to the state of the art., Comment: 2020 25th International Conference on Pattern Recognition (ICPR)

Most scientific contributions addressing cyber-security issues in water distribution networks present proposals of detection systems and very few propose location systems. A novel methodology for detection and location of cyber-attacks in water distribution networks (WDNs) is proposed in this paper. Structural analysis and autoencoder neural networks are effectively combined with a the control chart Adaptive Exponentially Weighted Moving Average (AEWMA). In the training phase, the proposed detection and location framework only requires data from normal operating conditions and knowledge about the behavioral model of the system which represents an advantage over previous works that demand for additional data of cyber-attacks. Among other advantages of the proposed methodology are the high performance in the effective, robust and early detection and the effectiveness of the location strategy. The proposal was evaluated with the known case study BATADAL.

Continual learning has attracted much attention in recent years, and many continual learning methods based on deep neural networks have been proposed. However, several important problems about these methods may lead to high decision cost and affect the practical application of continual learning networks. First, continual learning networks treat all categories equally, although the unbalance of misclassification cost happens in real-world cases. Second, there is a trade-off between learning new knowledge and keep old knowledge, which leads to the forgetting of old knowledge (i.e., the catastrophic forgetting). Third, even if low confidence of a sample, the continual learning methods based on the neural network will still give a clear classification result. We propose a sequential three-way decision model for continual learning to address these problems, named Incremental Sequential Three-Way Decision model (ISTWD). Introducing cost-sensitive sequential three-way decision to continual learning network, ISTWD reduces the decision cost of continual learning, which may alleviate the potentially high cost caused by the accuracy loss in continual learning. Besides, ISTWD includes a checkpoint procedure to judge whether the process of continual learning should stop. Experimental results on CIFAR-100 and Tiny-ImageNet verify the effectiveness of our method.

Heterogeneous graph representation learning is designed to learn meaningful representation vectors from heterogeneous networks in few dimensions to extract the structure and features of the attributes of these networks. The embedding vector is the basis of and crucial to complex network analysis, and can be used in such downstream tasks as classification, clustering, link prediction, and recommendation. Key issues in heterogeneous graph neural networks pertain to ways to define heterogeneous neighbors and ways to aggregate them. Although considerable research has been devoted to homogeneous and heterogeneous network representation, the effective combination of information on the network structure and the attributes of nodes, especially effective use of meta-paths containing specific semantic information, remains rare. Here a meta-path-based heterogeneous graph neural network model is proposed. The meta-path is applied to sample the heterogeneous neighbors of each node in the network, and aggregate features of the same types of nodes to form type-related embedding. A multi-head attention mechanism is then applied to aggregate information on neighbors of different types of nodes and the model is trained by reducing context loss. Experiments on classification, clustering, link prediction, and recommendation tasks verified the validity of this model, which significantly improved the results of baseline methods.

In the recommender system, the user’s historical behavior data is one of the most important sources of the system’s input data. According to the user’s feedback mechanism, behavior data can be divided into explicit feedback data and implicit feedback data. However, most recommendation algorithms focus separately on explicit feedback or implicit feedback. How to combine explicit and implicit feedback for recommendation tasks has always been a research problem. In recent years, deep learning technology has dominated the research on recommendation algorithms. But even the latest neural network-based recommendation algorithm cannot exceed classic methods (such as matrix factorization) in most cases. In this work, we propose a new collaborative filtering framework with neural network architecture. On the one hand, we use both explicit feedback data and implicit feedback data as input to learn multiple representations of users and items. On the other hand, we use multi-task learning to optimize our framework and use two relatively simple auxiliary tasks to enhance the generalization ability of our framework. Extensive experiments on five real-world datasets show significant improvements in our proposed framework over the state-of-the-art methods and vanilla matrix factorization.

Recent years have witnessed phenomenal developments worldwide in the field of NLP . But developments in Indian regional languages are very few compared to them. This work is a step towards the construction of a target word sense disambiguation system in Malayalam, which is the regional language of the state of Kerala, India. Word Sense Disambiguation/Determination refers to the task of correctly identifying the sense of an ambiguous word from its context. This is considered an AI-Complete problem in the field of Natural Language Processing . For this purpose, an exclusive corpus of 1,147 contexts of target ambiguous words has been created, which to the best of our knowledge is the first attempt in Malayalam. This work describes how the performance of an unsupervised LDA-based approach towards WSD could be improved using semantic features like synonyms and co-occurrence information.

With the continuous scaling down of process nodes, standard cells become much smaller and cell counts are dramatically increased. Pin accessibility becomes one of the major issues causing design rule violations (DRVs). To tackle this problem, many recent works apply machine-learning-based techniques to predict whether a local region has DRV or not by regarding global routing (GR) congestion and local pin density as the main features during the training process. Empirically, however, DRV occurrence is not necessary to be strongly correlated with the two features in advanced nodes. In this article, we propose the first work of deep-learning-based DRV prediction using pin pattern as our major feature to directly identify whether a DRV will exist or not due to bad pin accessibility of the given pin pattern. Unlike most of the existing models that can only be used for DRV prediction, the proposed models can be applied to guide detailed placement for pin accessibility optimization during physical design. Experimental results show that the proposed models are greatly superior than those of previous studies in terms of all quantitative metrics. Additionally, the numbers of DRVs can be dramatically reduced by applying the proposed model-guided detailed placement flow.

Background Echocardiography provides complex data on cardiac function that can be integrated into patterns of dysfunction related to the severity of cardiac disease. The aim of this study was to demonstrate the feasibility of applying machine learning (ML) to automate the integration of echocardiographic data from the whole cardiac cycle and to automatically recognize patterns in velocity profiles and deformation curves, allowing the identification of functional phenotypes. Methods Echocardiography was performed in 189 clinically managed patients with hypertension and 97 healthy individuals without hypertension. Speckle-tracking analysis of the left ventricle and atrium was performed, and deformation curves were extracted. Aortic and mitral blood pool pulsed-wave Doppler and mitral annular tissue pulsed-wave Doppler velocity profiles were obtained. These whole–cardiac cycle deformation and velocity curves were used as ML input. Unsupervised ML was used to create a representation of patients with hypertension in a virtual space in which patients are positioned on the basis of the similarity of their integrated whole–cardiac cycle echocardiography data. Regression methods were used to explore patterns of echocardiographic traces within this virtual ML-derived space, while clustering was used to define phenogroups. Results The algorithm captured different patterns in tissue and blood-pool velocity and deformation profiles and integrated the findings, yielding phenotypes related to normal cardiac function and others to advanced remodeling associated with pressure overload in hypertension. The addition of individuals without hypertension into the ML-derived space confirmed the interpretation of normal and remodeled phenotypes. Conclusions ML-based pattern recognition is feasible from echocardiographic data obtained during the whole cardiac cycle. Automated algorithms can consistently capture patterns in velocity and deformation data and, on the basis of these patterns, group patients into interpretable, clinically comprehensive phenogroups that describe structural and functional remodeling. Automated pattern recognition may potentially aid interpretation of imaging data and diagnostic accuracy.

Deep models have achieved prominent results in pattern recognition tasks, especially computer vision and natural language processing. However, the dataset bias caused by the distribution discrepancy between the training and testing data hinders the generalization ability of deep models. Though many domain adaptation approaches have been proposed to mitigate such negative effect, most of them improve the transferability of features by aligning global distributions of deep models. Few researchers pay attention to the versatility of deep features which can play a vital role in cross-domain recognition. In this paper, we propose to enrich the classic deep learning models by capturing high-low-frequency information and multi-scale features, which deal with the domain shift that cannot be easily addressed by merely feature-level alignment. The Hierarchical Transfer Network (HTN) leverages octave convolution, pyramid features, and self-attention mechanism for revamping the classic models, which can be further integrated with any domain alignment approaches by replacing the feature extractor with the proposed HTN. Extensive experiments have been conducted on three public domain adaptation benchmarks. The results show that the proposed HTN can effectively improve adversarial-based, statistics-based, and norm-based domain adaptation approaches, achieving competitive performance without involving model complexity.

Spatio-temporal information processing is fundamental in both brain functions and AI applications. Current strategies for spatio-temporal pattern recognition usually involve explicit feature extraction followed by feature aggregation, which requires a large amount of labeled data. In the present study, motivated by the subcortical visual pathway and early stages of the auditory pathway for motion and sound processing, we propose a novel brain-inspired computational model for generic spatio-temporal pattern recognition. The model consists of two modules, a reservoir module and a decision-making module. The former projects complex spatio-temporal patterns into spatially separated neural representations via its recurrent dynamics, the latter reads out neural representations via integrating information over time, and the two modules are linked together using known examples. Using synthetic data, we demonstrate that the model can extract the frequency and order information of temporal inputs. We apply the model to reproduce the looming pattern discrimination behavior as observed in experiments successfully. Furthermore, we apply the model to the gait recognition task, and demonstrate that our model accomplishes the recognition in an event-based manner and outperforms deep learning counterparts when training data is limited.

Cattle behavior monitoring is critical for understanding cattle welfare and health status. One of the most powerful and cost-effective monitoring methods is a neural network-based monitoring system that analyzes time series data from inertial sensors attached to cows. However, while deep learning has achieved many successes in pattern recognition, large-scale datasets are often required. When given a limited number of data, data augmentation is an extremely useful and low-cost preprocessing step for neural network-based systems. Data augmentation for inertial sensor data, however, has yet to be thoroughly investigated. This letter proposes several inertial sensor data augmentation methods in a manner that fits the characteristics of cattle behavioral data. The proposed approaches are applied to the task of cattle behavior classification with convolutional neural networks, which is challenging given limited data. The classification performance increases from 83.07 to 94.43% with appropriate augmentation steps. In conclusion, the data augmentation approaches presented here can help to improve deep learning performance regarding cattle behavior classification and decrease the overall system cost stemming from data acquisition and labeling.

The problem of developing universal classifiers of biomedical data, in particular those that characterize the presence of a large number of parameters, inaccuracies and uncertainty, is urgent. Many studies are aimed at developing methods for analyzing these data, among them there are methods based on a neural network (NN) in the form of a multilayer perceptron (MP) using GA. The question of the application of evolutionary algorithms (EA) for setting up and learning the neural network is considered. Theories of neural networks, genetic algorithms (GA) and decision trees intersect and penetrate each other, new developed neural networks and their applications constantly appear. An example of a problem that is solved using EA algorithms is considered. Its goal is to develop and research a classifier for the diagnosis of breast cancer, obtained by combining the capabilities of the multilayer perceptron using the genetic algorithm (GA) and the CART decision tree. The possibility of improving the classifiers of biomedical data in the form of NN based on GA by applying the process of appropriate preparation of biomedical data using the CART decision tree has been established. The obtained results of the study indicate that these classifiers show the highest efficiency on the set of learning and with the minimum reduction of Decision Trees; increasing the number of contractions usually degrades the simulation result. On two datasets on the test set, the simulation accuracy was »83–87%. The experiments carried out have confirmed the effectiveness of the proposed method for the synthesis of neural networks and make it possible to recommend it for practical use in processing data sets for further diagnostics, prediction, or pattern recognition

Pain management is gaining the attention of clinical practitioners to relieve patients from pain in an effective manner. Pain management is primarily dependent on pain measurement. Researchers have proposed various techniques to measure pain from facial expressions improving the accuracy and efficiency of the traditional pain measurement such as self-reporting and visual analog scale. Developments in the field of deep learning have further enhanced the pain assessment technique. Despite of the state of the art performance of deep learning algorithms, adaptation to new subjects is still a problem due to availability of a few samples of the same. Authors have addressed this issue by employing a model agnostic meta-learning algorithm for pain detection and fast adaptation of the trained algorithm for new subjects using only a few labeled images. The model is pre-trained with labeled images of subjects with five pain levels to acquire meta-knowledge in the presented work. This meta-knowledge is then used to adapt the model to a new learning task in the form of a new subject. The proposed model is evaluated on a benchmark dataset, i.e., UNBC McMaster pain archive database. Experimental results show that the model can be very easily adapted to new subjects with the accuracy of $$96\%$$ and $$98\%$$ for 1-shot and 5-shot learning respectively, proving the potential of the proposed algorithm for clinical use.

Even though most existing monocular 3D human pose estimation methods achieve very competitive performance, they are limited in estimating heterogeneous human body parts with the same decoder architecture. In this work, we present an approach to build a part-aware 3D human pose estimator to better deal with these heterogeneous human body parts. Our proposed method consists of two learning stages: (1) searching suitable decoder architectures for specific parts and (2) training the part-aware 3D human pose estimator built with these optimized neural architectures. Consequently, our searched model is very efficient and compact and can automatically select a suitable decoder architecture to estimate each human body part. In comparison with previous state-of-the-art models built with ResNet-50 network, our method can achieve better performance and reduce 64.4% parameters and 8.5% FLOPs (multiply-adds). We validate the robustness and stability of our searched models by conducting extensive and rigorous ablation experiments. Our method can advance state-of-the-art accuracy on both the single-person and multi-person 3D human pose estimation benchmarks with affordable computational cost.

Human activity recognition (HAR) based on wearable devices has found wide applications in fitness, health care, etc. Given the personalized wearing styles of such devices and distinctive motion patterns, the activities of daily living normally vary from person to person in terms of strength, amplitude, speed, category, etc. The specialization of a universal HAR model to a specific subject without experiencing catastrophic forgetting is a significant challenge. In this paper, we propose a novel incremental learning method, namely, an adaptive coefficient-based kernelized and regularized network (KeRNet-AC), for personalized activity recognition. During the adaptation stage of the model training process, KeRNet-AC consistently monitors the probable ill-conditioned degree of the generated solution, which we believe is strongly correlated with the catastrophic forgetting problem, and automatically makes the solution well conditioned. To reduce the computational complexity of KeRNet-AC, we also introduce an active data selection principle into KeRNet-AC. This variation is called A-KeRNet-AC. To evaluate the performance of KeRNet-AC and A-KeRNet-AC, we conduct extensive experiments on five public activity datasets. The experimental results demonstrate that KeRNet-AC outperforms related state-of-the-art methods in most cases and that A-KeRNet-AC can quickly perform model training and activity prediction. Moreover, the performance of the proposed methods steadily improves during the adaptation stage and ultimately converges without degradation, demonstrating the strong potential of KeRNet-AC and A-KeRNet-AC for personalized activity recognition.

Due to the rapid increase in exchange of text information via internet network, the security and the reliability of the digital content has become a major research issue. The main challenges faced by researchers are content authentication, integrity verification and tampering detection of digital contents. In this paper, these issues are addressed with great emphasis on text information, which is natural language dependent. Hence, a novel intelligent zero-watermarking approach is proposed for content authentication and tampering detection of Arabic text contents. In the proposed approach, both the embedding and extracting of the watermark are logically implemented, which causes no change on the digital text. This is achieved by using fourth-level-order and alphanumeric mechanism of Markov model as a soft computing technique for the analysis of Arabic text to obtain the features of the given text which is considered as the digital watermark. This digital watermark is later used for the detection of any tampering attack on the received Arabic text. An extensive set of experiments using four datasets of varying lengths proves the effectiveness of the proposed approach in terms of robustness, effectiveness and applicability under multiple random locations of insertion, reorder and deletion attacks. Compared with baseline approaches, the proposed approach has improved performance regarding watermark robustness and tampering detection accuracy.

Knowledge distillation has been widely used to produce portable and efficient neural networks which can be well applied on edge devices for computer vision tasks. However, almost all top-performing knowledge distillation methods need to access the original training data, which usually has a huge size and is often unavailable. To tackle this problem, we propose a novel data-free approach in this paper, named Dual Discriminator Adversarial Distillation (DDAD) to distill a neural network without the need of any training data or meta-data. To be specific, we use a generator to create samples through dual discriminator adversarial distillation, which mimics the original training data. The generator not only uses the pre-trained teacher’s intrinsic statistics in existing batch normalization layers but also obtains the maximum discrepancy from the student model. Then the generated samples are used to train the compact student network under the supervision of the teacher. The proposed method obtains an efficient student network which closely approximates its teacher network, without using the original training data. Extensive experiments are conducted to demonstrate the effectiveness of the proposed approach on CIFAR, Caltech101 and ImageNet datasets for classification tasks. Moreover, we extend our method to semantic segmentation tasks on several public datasets such as CamVid, NYUv2, Cityscapes and VOC 2012. To the best of our knowledge, this is the first work on generative model based data-free knowledge distillation on large-scale datasets such as ImageNet, Cityscapes and VOC 2012. Experiments show that our method outperforms all baselines for data-free knowledge distillation.

This paper develops a weak-label-based global and local multi-view multi-label learning with three-way clustering (WL-GLMVML-ATC) to solve multi-view multi-label data sets and exploit more authentic global and local label correlations of both the whole data set and each view simultaneously. Different from the traditional learning methods, WL-GLMVML-ATC pays more attention to the solutions of weak-label cases and uncertain relationships of clusters with the usage of Universum and active three-way clustering. According to Universum notion, even though the size of labeled instances is much more smaller than the unlabeled ones, the useful sample information can still be enhanced. Through the active three-way clustering strategy, the belongingness of instances to a cluster depend on the probabilities of uncertain instances belonging to core regions. This strategy brings a more authentic local label correlation since many traditional methods suppose that instances and the corresponding clusters always exhibit certain relationships such as belong-to definitely and not belong-to definitely. This hypothesis is not ubiquitous in real-world applications. According to the experiments, we can see WL-GLMVML-ATC (1) achieves a better performance, be superior to the classical multi-view learning methods and multi-label learning methods in statistical, advances the development of these learning methods in final; (2) won’t add too much running time; (3) has a good convergence and ability to process multi-view multi-label data sets.

Kirsten Ras (KRAS) mutation identification has great clinical significance to formulate the rectal cancer treatment scheme. Recently, the development of deep learning does much help to improve the computer-aided diagnosis technology. However, deep learning models are usually designed for only one task, ignoring the potential benefits in jointly performing both tasks. In this paper, we proposed a joint network named segmentation-based multi-scale attention model (SMSAM) to predict the mutation status of KRAS gene in rectal cancer. More specifically, the network performs segmentation and prediction tasks at the same time. The two tasks mutually transfer knowledge between each other by sharing the same encoder. Meanwhile, two universal multi-scale attention blocks are introduced to ensure that the network more focuses on the region of interest. Besides, we also proposed an entropy branch to provide more discriminative features for the model. Finally, the method is evaluated on internal and external datasets. The results show that the comprehensive performance of SMSAM is better than the existing methods. The code and model have been publicly available.

Representation learning for video is increasingly gaining attention in the field of computer vision. For instance, video prediction models enable activity and scene forecasting or vision-based planning and control. In this article, we investigate the combination of differentiable physics and spatial transformers in a deep action conditional video representation network. By this combination our model learns a physically interpretable latent representation and can identify physical parameters. We propose supervised and self-supervised learning methods for our architecture. In experiments, we consider simulated scenarios with pushing, sliding and colliding objects, for which we also analyze the observability of the physical properties. We demonstrate that our network can learn to encode images and identify physical properties like mass and friction from videos and action sequences. We evaluate the accuracy of our training methods, and demonstrate the ability of our method to predict future video frames from input images and actions.

With the growing demands of data products, the selection of satellite image data products becomes a challenging decision issue for customers. The objective of this study is to propose a practically sound decision-making approach for solving the satellite image data products selection problems. First, the influencing factors of selecting satellite image data products are identified. Then, hybrid evaluation information is recommended to represent these criteria. That is, numerical and interval-valued quantification is used for quantitative criteria, and picture fuzzy numbers (PFNs) are considered to express qualitative criteria. To reflect decision makers’ preferences, a non-linear optimization is implemented to treat criteria weights with constraints. Thereafter, some penalty functions are defined and the artificial fish swarm algorithm (AFSA) is improved to calculate weight values. Furthermore, six main parameters of AFSA are analyzed. Compared with other commonly used algorithms (such as genetic algorithm (GA) and particle swarm optimization (PSO)), the largest advantage of AFSA is its high robustness of parameters and initial values. Finally, the traditional multi-attributive border approximation area comparison (MABAC) is modified with likelihood measures to obtain the best data product in hybrid evaluation environments. Furthermore, the feasibility and effectiveness of the proposed approach is validated by comparing with existing methods in some representative literature. The results demonstrate that the proposed method is feasible and can provide useful guidelines for the selection and pricing of satellite image data products.

The systematic processing of unstructured communication data as well as the milestone of pattern recognition in order to determine communication groups in negotiations bears many challenges in Machine Learning. In particular, the so-called curse of dimensionality makes the pattern recognition process demanding and requires further research in the negotiation environment. In this paper, various selected renowned clustering approaches are evaluated with regard to their pattern recognition potential based on high-dimensional negotiation communication data. A research approach is presented to evaluate the application potential of selected methods via a holistic framework including three main evaluation milestones: the determination of optimal number of clusters, the main clustering application, and the performance evaluation. Hence, quantified Term Document Matrices are initially pre-processed and afterwards used as underlying databases to investigate the pattern recognition potential of clustering techniques by considering the information regarding the optimal number of clusters and by measuring the respective internal as well as external performances. The overall research results show that certain cluster separations are recommended by internal and external performance measures by means of a holistic evaluation approach, whereas three of the clustering separations are eliminated based on the evaluation results.

The ambulatory ECG (AECG) is an important diagnostic tool for many heart electrophysiology-related cases. AECG covers a wide spectrum of devices and applications. At the core of these devices and applications are the algorithms responsible for signal conditioning, ECG beat detection and classification, and event detections. Over the years, there has been huge progress for algorithm development and implementation thanks to great efforts by researchers, engineers, and physicians, alongside the rapid development of electronics and signal processing, especially machine learning (ML). The current efforts and progress in machine learning fields are unprecedented, and many of these ML algorithms have also been successfully applied to AECG applications. This review covers some key AECG applications of ML algorithms. However, instead of doing a general review of ML algorithms, we are focusing on the central tasks of AECG and discussing what ML can bring to solve the key challenges AECG is facing. The center tasks of AECG signal processing listed in the review include signal preprocessing, beat detection and classification, event detection, and event prediction. Each AECG device/system might have different portions and forms of those signal components depending on its application and the target, but these are the topics most relevant and of greatest concern to the people working in this area.

Segmentation is a significant stage for the recognition of old newspapers. Text-line extraction in the documents like newspaper pages which have very complex layouts poses a significant challenge. Old newspaper documents printed in Gurumukhi script present several forms of hurdles in segmentation due to noise, degradation, bleed-through of ink, multiple font styles and sizes, little space between neighboring text lines, overlapping of lines, etc. Because of the low quality and the complexity of these documents, automatic text line segmentation remains an open research field. Very few researches are available in the literature to segment news articles in Gurumukhi script. This is one of the first few attempts to recognize Gurumukhi newspaper text. The goal of this paper is to present a new methodology for text-line extraction by integrating median calculation and strip height calculation techniques. Non-suitability of existing techniques to segment newspaper text lines have also been discussed with results in the article. The efficiency of the proposed algorithm is demonstrated by experimentation directed on two diverse own made datasets: (a) on the data set of single-column documents with headlines block (b) on the dataset of multi-column documents with headlines block.

Context. The problem of structural modification of pre-synthesized models based on artificial neural networks to ensure the property of interpretation when working with big data is considered. The object of the study is the process of structural modification of artificial neural networks using adaptive mechanisms. Objective of the work is to develop a method for structural modification of neural networks to increase their speed and reduce resource consumption when processing big data. Method. A method of structural adjustment of neural networks based on adaptive mechanisms borrowed from neuroevolutionary synthesis methods is proposed. At the beginning, the method uses a system of indicators to evaluate the existing structure of an artificial neural network. The assessment is based on the structural features of neuromodels. Then the obtained indicator estimates are compared with the criteria values for choosing the type of structural changes. Variants of mutational changes from the group of methods of neuroevolutionary modification of the topology and weights of the neural network are used as variants of structural change. The method allows to reduce the resource intensity during the operation of neuromodels, by accelerating the processing of big data, which expands the field of practical application of artificial neural networks. Results. The developed method is implemented and investigated by the example of using a recurrent artificial network of the long short-term memory type when solving the classification problem. The use of the developed method allowed speed up of the neuromodel with a test sample by 25.05%, depending on the computing resources used. Conclusions. The conducted experiments confirmed the operability of the proposed mathematical software and allow us to recommend it for use in practice in the structural adjustment of pre-synthesized neuromodels for further solving problems of diagnosis, forecasting, evaluation and pattern recognition using big data. The prospects for further research may consist in a more fine-tuning of the indicator system to determine the connections encoding noisy data in order to further improve the accuracy of models based on neural networks.

In this work, we propose a 3D fully convolutional architecture for video saliency prediction that employs hierarchical supervision on intermediate maps (referred to as conspicuity maps) generated using features extracted at different abstraction levels. We provide the base hierarchical learning mechanism with two techniques for domain adaptation and domain-specific learning. For the former, we encourage the model to unsupervisedly learn hierarchical general features using gradient reversal at multiple scales, to enhance generalization capabilities on datasets for which no annotations are provided during training. As for domain specialization, we employ domain-specific operations (namely, priors, smoothing and batch normalization) by specializing the learned features on individual datasets in order to maximize performance. The results of our experiments show that the proposed model yields state-of-the-art accuracy on supervised saliency prediction. When the base hierarchical model is empowered with domain-specific modules, performance improves, outperforming state-of-the-art models on three out of five metrics on the DHF1K benchmark and reaching the second-best results on the other two. When, instead, we test it in an unsupervised domain adaptation setting, by enabling hierarchical gradient reversal layers, we obtain performance comparable to supervised state-of-the-art. Source code, trained models and example outputs are publicly available at https://github.com/perceivelab/hd2s.

The phenomenon of Adversarial Examples has become one of the most intriguing topics associated to deep learning. The so-called adversarial attacks have the ability to fool deep neural networks with inappreciable perturbations. While the effect is striking, it has been suggested that such carefully selected injected noise does not necessarily appear in real-world scenarios. In contrast to this, some authors have looked for ways to generate adversarial noise in physical scenarios (traffic signs, shirts, etc.), thus showing that attackers can indeed fool the networks. In this paper we go beyond that and show that adversarial examples also appear in the real-world without any attacker or maliciously selected noise involved. We show this by using images from tasks related to microscopy and also general object recognition with the well-known ImageNet dataset. A comparison between these natural and the artificially generated adversarial examples is performed using distance metrics and image quality metrics. We also show that the natural adversarial examples are in fact at a higher distance from the originals that in the case of artificially generated adversarial examples.

Muscle imaging has progressively gained popularity in the neuromuscular field. Together with detailed clinical examination and muscle biopsy, it has become one of the main tools for deep phenotyping and orientation of etiological diagnosis. Even in the current era of powerful new generation sequencing, muscle MRI has arisen as a tool for prioritization of certain genetic entities, supporting the pathogenicity of variants of unknown significance and facilitating diagnosis in cases with an initially inconclusive genetic study. Although the utility of muscle imaging is increasingly clear, it has not reached its full potential in clinical practice. Pattern recognition is known for a number of diseases and will certainly be enhanced by the use of machine learning approaches. For instance, MRI heatmap representations might be confronted with molecular results by obtaining a probabilistic diagnosis based in each disease "MRI fingerprints". Muscle ultrasound as a screening tool and quantified techniques such as Dixon MRI seem still underdeveloped. In this paper, we aim to appraise the advances in recent years in pediatric muscle imaging and try to define areas of uncertainty and potential advances that might become standardized to be widely used in the future.

In this work, we propose two deep learning-based architectures tailored for gas identification and quantification, which automatically tune hyper-parameters of the network for optimal performance. The immense success of deep learning in the field of computer vision and natural language processing inspired us to design deep learning-based gas identification and quantification network. The first architecture is proposed for gas quantification, which is based on 1D-CNN. It makes use of raw time-series gas sensor array data and provides the concentration of each gas in a mixture of gases. The second architecture is presented for gas quantification, which is based on a deep belief network combined with drift-aware feature adaptation strategy. The proposed models identify and quantify the gases with improved accuracy despite the presence of sensor drift. Additionally, hyper-parameters of both the networks are automatically tuned for optimal performance. Although several pattern recognition methods related to machine learning, fuzzy logic and hybrid models have been used to identify gas and quantify the gases in the mixture, the performances of these techniques enormously depend on the feature engineering and selection of hyper-parameters. Experimental results show that the proposed methods are an effective technique for identifying gases and quantifying the mixture of gases for e-nose data. We also present that the proposed methods outperforms various other methods and can provide higher identification and quantification accuracy in the pres-ence of sensor drift.

Today’s deep learning systems deliver high performance based on end-to-end training but are notoriously hard to inspect. We argue that there are at least two reasons making inspectability challenging: (i) representations are distributed across hundreds of channels and (ii) a unifying metric quantifying inspectability is lacking. In this paper, we address both issues by proposing Semantic Bottlenecks (SB), which can be integrated into pretrained networks, to align channel outputs with individual visual concepts and introduce the model agnostic Area Under inspectability Curve (AUiC) metric to measure the alignment. We present a case study on semantic segmentation to demonstrate that SBs improve the AUiC up to six-fold over regular network outputs. We explore two types of SB-layers in this work. First, concept-supervised SB-layers (SSB), which offer inspectability w.r.t. predefined concepts that the model is demanded to rely on. And second, unsupervised SBs (USB), which offer equally strong AUiC improvements by restricting distributedness of representations across channels. Importantly, for both SB types, we can recover state of the art segmentation performance across two different models despite a drastic dimensionality reduction from 1000s of non aligned channels to 10s of semantics-aligned channels that all downstream results are based on.

This work presents a gesture-based approach to estimate task understanding and performance during remote ultrasound tasks. Our approach is comprised of two main components. The first component uses the Multi-Agent Gestural Instruction Comparer (MAGIC) framework to represent and compare the gestures performed by collaborators. Through MAGIC, gestures can be compared based in their morphology, semantics, and pragmatics. The second component computes the Physical Instructions Assimilation (PIA) metric, a score representing how well are gestures being used to communicate and execute physical instructions. To evaluate our hypothesis, 20 participants performed a remote ultrasound task consisting of three subtasks: vessel detection, blood extraction, and foreign body detection. MAGIC’s gesture comparison approaches were compared against two other approaches based on how well they replicated human-annotated gestures matchings. Our approach outperformed the others, agreeing with the human baseline over 76% of the times. Subsequently, a correlation analysis was performed to compare PIA’s task understanding insights with those of three other metrics: error rate, idle time rate, and task completion percentage. Significant correlations ( $$p\,\le \,0.04$$ ) were found between PIA and all the other metrics, positioning PIA as an effective metric for task understanding estimation. Finally, post-experiment questionnaires were used to subjectively evaluate the participants’ perceived understanding. The PIA score was found to be significantly correlated with the participants’ overall task understanding ( $$p\le 0.05$$ ), hinting to the relation between the assimilation of physical instructions and self-perceived understanding. These results demonstrate that gestures an be used to estimate task understanding in remote ultrasound tasks, which can improve how these tasks are performed and assessed.

Nowadays, the world is witnessing a significant rise in the cases of both reported and unnoticed violations. As an answer to this rising menace, video surveillance can fill the gap of covering untapped actions which lead to violence, while also ensuring a secure life. In our everyday life, surveillance can be accomplished efficiently by activity classification from drone videos. The prominent fields that have employed this technology are police work, video categorization, biometrics, and human–computer interaction. So far, no public dataset is available for violent activity classification using drone surveillance. Hence, this work aims to look into the domain of machine-driven recognition and classification of human actions from drone videos. In this study, the dataset is created using drones from different heights for an unconstrained environment. The study begins by performing key-point extraction and generate 2D skeletons for the persons in the frame. These extracted key points are given as features in the classification module to recognize the actions. The classification models used in the proposed method are SVM (support vector machine) and Random Forest. Experimental results show that the SVM model with RBF (radial basis function) kernel for activity classification is more efficient when compared to the prior proposed approaches and other experimented models. The research work has also analyzed the run time performance of the proposed system and achieve its real-time performance.

Thanks to the significant developments in healthcare industries, various types of medical data are generated. Analysing such valuable resources aid healthcare experts to understand the illnesses more precisely and provide better clinical services. Machine learning as one of the capable tools could assist healthcare experts in achieving expressive interpretation and making proper decisions. As annotation of medical data is a costly and sensitive task that can be performed just by healthcare professionals, label-free methods could be significantly promising. Interpretability and evidence-based decision are other concerns in medicine. These needs were our motivators to propose a novel clustering method based on hierarchical Dirichlet process mixtures of multivariate Beta distributions. To learn it, we applied batch and online variational methods for finding the proper number of clusters as well as estimating model parameters at the same time. The effectiveness of the proposed models is evaluated on three medical real applications, namely oropharyngeal carcinoma diagnosis, osteosarcoma analysis, and white blood cell counting.

Sign Language is the linguistic system adopted by the Deaf to communicate. The lack of fully-fledged Automatic Sign Language (ASLR) technologies contributes to the numerous difficulties that deaf individuals face in the absence of an interpreter, such as in private health appointments or in emergency situations. A challenging problem in the development of reliable ASLR systems is that sign languages do not rely only on manual gestures but also on facial expressions and other non-manual markers. This paper proposes to adopt Facial Action Coding System to encode sign language facial expressions. However, the state-of-the-art of Action Unit (AU) recognition models is mostly targeted to classify two dozen of AUs, typically related to the expression of emotions. We adopted Brazilian Sign Language (Libras) as our case study and we identified more than one hundred of AUs (with a great intersection with other sign languages). We then implemented and evaluated a novel AU recognition model architecture that combines SqueezeNet and geometric-based features. Our model obtained 88% of accuracy for 119 classes. Combined with the state-of-the-art of gesture recognition, our model is ready to improve sign disambiguation and to advance ASLR.

Due to the low computational complexity, the two-phase test sample representation method shows outstanding advantages in the field of real-time face recognition. However, its first phase does not fully consider the imbalance of the determined K-nearest training samples. This defect is not well reflected and will seriously affect the recognition accuracy of the second phase. In this paper, we explore the above issues and propose to incorporate the unselected training samples into the modeling process. The proposed method not only allows the unselected samples to perform the final representation but also makes the selected nearest training samples play a more significant role than the unselected ones in classification. It is efficiently solved with an analytical solution. The rationales and probability interpretation of the proposed method are presented to further guarantee its rationality and effectiveness. Extensive experiments on diverse face databases are conducted to demonstrate the superior recognition performance in comparison with other competing classifiers.