Deep Neural Network-based Classifier Research Articles

A prior knowledge-guided distributionally robust optimization-based adversarial training strategy for medical image classification

Medical image classification plays a vital role in computer vision applications within the field of healthcare and medicine, and deep neural network-based classifiers are continuously achieving new breakthroughs and demonstrating tremendous potential in medical imaging analysis. However, the lack of robustness in deep learning techniques makes it risky to apply these classifiers to the domain of healthcare. In addition, the existing adversarial training strategies and domain generation methods are difficult to generalize into the medical imaging field challenged by complex medical texture features. To address this issue, we propose a medical morphological knowledge-guided adversarial training strategy, by jointly considering the robustness against medical data distribution shifts and adversarial attacks from the view of distributionally robust optimization. First, we train a surrogate model with the augmented dataset by guided filtering for capturing model attention on medical morphological information. Next, we design a gradient normalization-based prior knowledge injection module to transfer the attention information learned by surrogate model to the main classifier. Finally, we design a distributionally robust a optimization-based training strategy to induce the main classifier to learn key diagnostic clues as well as enhance the robustness against adversarial attacks. To evaluate the effectiveness of the proposed methods, we perform experiments on two types of in-domain and out-of-domain medical image sets, which contain lung CT scan datasets and dermatoscopic image datasets. Comparative results show that the proposed training strategy achieves higher adversarial attack accuracy than all involved state-of-the-art adversarial training methods and domain generation methods. The code is available at https://github.com/sysu19351146/MMK-DRO.

A Test Statistic Estimation-Based Approach for Establishing Self-Interpretable CNN-Based Binary Classifiers.

Interpretability is highly desired for deep neural network-based classifiers, especially when addressing high-stake decisions in medical imaging. Commonly used post-hoc interpretability methods have the limitation that they can produce plausible but different interpretations of a given model, leading to ambiguity about which one to choose. To address this problem, a novel decision-theory-inspired approach is investigated to establish a self-interpretable model, given a pre-trained deep binary black-box medical image classifier. This approach involves utilizing a self-interpretable encoder-decoder model in conjunction with a single-layer fully connected network with unity weights. The model is trained to estimate the test statistic of the given trained black-box deep binary classifier to maintain a similar accuracy. The decoder output image, referred to as an equivalency map, is an image that represents a transformed version of the to-be-classified image that, when processed by the fixed fully connected layer, produces the same test statistic value as the original classifier. The equivalency map provides a visualization of the transformed image features that directly contribute to the test statistic value and, moreover, permits quantification of their relative contributions. Unlike the traditional post-hoc interpretability methods, the proposed method is self-interpretable, quantitative. Detailed quantitative and qualitative analyses have been performed with three different medical image binary classification tasks.

Enhancing adversarial robustness of quantum neural networks by adding noise layers

Abstract The rapid advancements in machine learning and quantum computing have given rise to a new research frontier: quantum machine learning. Quantum models designed for tackling classification problems possess the potential to deliver speed enhancements and superior predictive accuracy compared to their classical counterparts. However, recent research has revealed that quantum neural networks (QNNs), akin to their classical deep neural network-based classifier counterparts, are vulnerable to adversarial attacks. In these attacks, meticulously designed perturbations added to clean input data can result in QNNs producing incorrect predictions with high confidence. To mitigate this issue, we suggest enhancing the adversarial robustness of quantum machine learning systems by incorporating noise layers into QNNs. This is accomplished by solving a Min-Max optimization problem to control the magnitude of the noise, thereby increasing the QNN’s resilience against adversarial attacks. Extensive numerical experiments illustrate that our proposed method outperforms state-of-the-art defense techniques in terms of both clean and robust accuracy.

A Privacy-Protecting Step-Level Walking Direction Detection Algorithm Based on Floor Vibration

We present an algorithm and measurement system to detect the walking direction of persons based on ground vibrations. The approach is privacy-preserving because it solely relies on piezoelectric sensors built into the floor. Therefore, our system can be used in areas where cameras are not allowed or cannot capture the entire area. We present and compare our two innovative methods to analyze the ground vibrations caused by footsteps: the multi-peaks average algorithm (MPAA) and the multi-peaks averaged feature with a deep neural network-based classifier (MPAF-DNNC). MPAA judges the walking direction of pedestrians by analyzing the time-space relationship of at least two consecutive footstep vibration signals from multiple sensors. MPAF-DNNC receives multi-peaks averaged feature as input and uses a deep neural network-based classifier to judge walking direction. Our experiments and evaluation show that our system can correctly determine the walking direction based on only 3 input step events and provides an average F1 score of 0.97. When more than 5 step events are inputted, the proposed system can correctly determine the walking direction with an average F1 score of 1.00.

Speech emotion recognition using overlapping sliding window and Shapley additive explainable deep neural network

ABSTRACT Speech emotion recognition (SER) has several applications, such as e-learning, human-computer interaction, customer service, and healthcare systems. Although researchers have investigated lots of techniques to improve the accuracy of SER, it has been challenging with feature extraction, classifier schemes, and computational costs. To address the aforementioned problems, we propose a new set of 1D features extracted by using an overlapping sliding window (OSW) technique for SER in this study. In addition, a deep neural network-based classifier scheme called the deep Pattern Recognition Network (PRN) is designed to categorize emotional states from the new set of 1D features. We evaluate the proposed method on the Emo-DB and the AESSD datasets that contain several different emotional states. The experimental results show that the proposed method achieves an accuracy of 98.5% and 87.1% on the Emo-DB and AESSD datasets, respectively. It is also more comparable with accuracy to and better than the state-of-the-art and current approaches that use 1D features on the same datasets for SER. Furthermore, the SHAP (SHapley Additive exPlanations) analysis is employed for interpreting the prediction model to assist system developers in selecting the optimal features to integrate into the desired system.

WePos

On-demand delivery is an emerging business in recent years where accurate indoor locations of Gig couriers play an important role in the order dispatch and delivery process. To cater to this need, WiFi-based indoor positioning methods have become an alternative method for on-demand delivery thanks to extensive WiFi deployment in the indoor environment. Existing WiFi-based indoor localization and positioning methods are not suitable for large-scale on-demand delivery scenarios due to high costs (e.g., high labor cost to collect fingerprints) and limited coverage due to limited labeled data. In this work, we explore (i) massive crowdsourced WiFi data collecting from wearable or mobile devices of couriers with little extra effort and (ii) natural manual reports data in the delivery process as two opportunities to perform merchant-level indoor positioning in a weak-supervised manner. Specifically, we proposed WePos, an end-to-end weak-supervised-based merchant-level positioning framework, which consists of the following three parts: (i) a Bidirectional Encoder Representations from Transformers (BERT) based pre-training module to learn latent embeddings of WiFi access points, (ii) a contrastive label self-generate module to produce pseudos for WiFi scanning lists by matching similarity embedding clustering results and couriers' reporting behaviors. (iii) a deep neural network-based classifier to fine-tune the whole training process and conduct online merchant-level position inference. To evaluate the performance of our system, we conduct extensive experiments in both a large-scale public crowdsourcing dataset with over 50 GB of WiFi signal records and a real-world WiFi crowdsourced dataset collected from Eleme, (i.e., one of the largest on-demand delivery platforms in China) in four multi-floor malls in Shanghai. Experimental results show that WePos outperforms state-of-the-art baselines in the merchant-level positioning performance, offer up to 91.4% in positioning accuracy.

Multiclass classification by Min–Max ECOC with Hamming distance optimization

Two questions often arise in the field of the ensemble in multiclass classification problems, (i) how to combine base classifiers and (ii) how to design possible binary classifiers. Error-correcting output codes (ECOC) methods answer these questions, but they focused on only the general goodness of the classifier. The main purpose of our research was to strengthen the bottleneck of the ensemble method, i.e., to minimize the largest values of two types of error ratios in the deep neural network-based classifier. The research was theoretical and experimental, the proposed Min–Max ECOC method suggests a theoretically proven optimal solution, which was verified by experiments on image datasets. The optimal solution was based on the maximization of the lowest value in the Hamming matrix coming from the ECOC matrix. The largest ECOC matrix, the so-called full matrix is always a Min–Max ECOC matrix, but smaller matrices generally do not reach the optimal Hamming distance value, and a recursive construction algorithm was proposed to get closer to it. It is not easy to calculate optimal values for large ECOC matrices, but an interval with upper and lower limits was constructed by two theorems, and they were proved. Convolutional Neural Networks with Min–Max ECOC matrix were tested on four real datasets and compared with OVA (one versus all) and variants of ECOC methods in terms of known and two new indicators. The experimental results show that the suggested method surpasses the others, thus our method is promising in the ensemble learning literature.

Real-Time Driving Distraction Recognition Through a Wrist-Mounted Accelerometer.

We propose a method for recognizing driver distraction in real time using a wrist-worn inertial measurement unit (IMU). Distracted driving results in thousands of fatal vehicle accidents every year. Recognizing distraction using body-worn sensors may help mitigate driver distraction and consequently improve road safety. Twenty participants performed common behaviors associated with distracted driving while operating a driving simulator. Acceleration data collected from an IMU secured to each driver's right wrist were used to detect potential manual distractions based on 2-s long streaming data. Three deep neural network-based classifiers were compared for their ability to recognize the type of distractive behavior using F1-scores, a measure of accuracy considering both recall and precision. The results indicated that a convolutional long short-term memory (ConvLSTM) deep neural network outperformed a convolutional neural network (CNN) and recursive neural network with long short-term memory (LSTM) for recognizing distracted driving behaviors. The within-participant F1-scores for the ConvLSTM, CNN, and LSTM were 0.87, 0.82, and 0.82, respectively. The between-participant F1-scores for the ConvLSTM, CNN, and LSTM were 0.87, 0.76, and 0.85, respectively. The results of this pilot study indicate that the proposed driving distraction mitigation system that uses a wrist-worn IMU and ConvLSTM deep neural network classifier may have potential for improving transportation safety.

Adversarial attack and defense strategies for deep speaker recognition systems

Robust speaker recognition, including in the presence of malicious attacks, is becoming increasingly important and essential, especially due to the proliferation of smart speakers and personal agents that interact with an individual’s voice commands to perform diverse and even sensitive tasks. Adversarial attack is a recently revived domain which is shown to be effective in breaking deep neural network-based classifiers, specifically, by forcing them to change their posterior distribution by only perturbing the input samples by a very small amount. Although, significant progress in this realm has been made in the computer vision domain, advances within speaker recognition is still limited. We present an expository paper that considers several adversarial attacks to a deep speaker recognition system, employs strong defense methods as countermeasures, and reports a comprehensive set of ablation studies to better understand the problem. The experiments show that the speaker recognition systems are vulnerable to adversarial attacks, and the strongest attacks can reduce the accuracy of the system from 94% to even 0%. The study also compares the performances of the employed defense methods in detail, and finds adversarial training based on Projected Gradient Descent (PGD) to be the best defense method in our setting. We hope that the experiments presented in this paper provide baselines that can be useful for the research community interested in further studying adversarial robustness of speaker recognition systems.

OctNET: A Lightweight CNN for Retinal Disease Classification from Optical Coherence Tomography Images

Background and ObjectiveRetinal diseases are becoming a major health problem in recent years. Their early detection and ensuing treatment are essential to prevent visual damage, as the number of people affected by diabetes is expected to grow exponentially. Retinal diseases progress slowly, without any discernible symptoms. Optical Coherence Tomography (OCT) is a diagnostic tool capable of analyzing and identifying the quantitative discrimination in the disease affected retinal layers with high resolution. This paper proposes a deep neural network-based classifier for the computer-aided classification of Diabetic Macular Edema (DME), drusen, Choroidal NeoVascularization (CNV) from normal OCT images of the retina. MethodsIn the proposed method, we demonstrate the feasibility of classifying and detecting severe retinal pathologies from OCT images using a deep convolutional neural network having six convolutional blocks. The classification results are explained using a gradient-based class activation mapping algorithm. ResultsTraining and validation of the model are performed on a public dataset of 83,484 images with expert-level disease grading of CNV, DME, and drusen, in addition to normal retinal image. We achieved a precision of 99.69%, recall of 99.69%, and accuracy of 99.69% with only three misclassifications out of 968 test cases. ConclusionIn the proposed work, downsampling and weight sharing were introduced to improve the training efficiency and were found to reduce the trainable parameters significantly. The class activation mapping was also performed, and the output image was similar to the retina’s actual color OCT images. The proposed network used only 6.9% of learnable parameters compared to the existing ResNet-50 model and yet outperformed it in classification. The proposed work can be potentially employed in real-time applications due to reduced complexity and fewer learnable parameters over other models.

Semantic analysis-based relevant data retrieval model using feature selection, summarization and CNN

Semantic analysis is playing a major role and task in text mining process caused by the presence of huge number of relevant and irrelevant data in Internet and other resources. Here, the semantic-based text summarization must be incorporated for the successful relevant data extraction by using data classification. The accurate classification process is done by using deep learning techniques recently. However, no existing model is achieved reasonable relevancy accuracy. For overcoming the drawbacks, we propose an effective semantic analysis-based relevant data retrieval model for retrieving the relevant data from local repository or web applications in Internet. This new model consists of (i) semantic similarity-based feature selection and (ii) enrichment technique, (iii) data summarization technique and iv) text relationship-based deep neural network classifier. Here, we propose a new semantic analysis-based feature selection algorithm to select the similarity indexed relevant data from local repositories or web applications. In addition, a new semantic-based data summarization technique is also introduced for summarizing the text that is available in the online resources. Finally, a new semantic similarity-based deep neural network-based classifier is also introduced for categorizing the data according to the semantic relation. The proposed model is proved the effectiveness of the data retrieval process by conducting various experiments based on the relevant data extraction from Internet resources, and it also tested with the recognized datasets.

Effect of Data Representation for Time Series Classification—A Comparative Study and a New Proposal

Time series classification (TSC) is becoming very important in the area of pattern recognition with the increased availability of time series data in various natural and real life phenomena. TSC is a challenging problem because, due to the attributes being ordered, traditional machine learning algorithms for static data are not quite suitable for processing temporal data. Due to the gradual increase of computing power, a large number of TSC algorithms have been developed recently. In addition to traditional feature-based, model-based or distance-based algorithms, ensemble and deep networks have recently become popular for time series classification. Time series are essentially huge, and classifying raw data is computationally expensive in terms of both processing and storage. Representation techniques for data reduction and ease of visualization are needed for accurate classification. In this work a recurrence plot-based data representation is proposed and time series classification in conjunction with a deep neural network-based classifier has been studied. A simulation experiment with 85 benchmark data sets from UCR repository has been undertaken with several state of the art algorithms for time series classification in addition to our proposed scheme of classification for comparative study. It was found that, among non-ensemble algorithms, the proposed algorithm produces the highest classification accuracy for most of the data sets.

Towards scaling Twitter for digital epidemiology of birth defects

Social media has recently been used to identify and study a small cohort of Twitter users whose pregnancies with birth defect outcomes—the leading cause of infant mortality—could be observed via their publicly available tweets. In this study, we exploit social media on a larger scale by developing natural language processing (NLP) methods to automatically detect, among thousands of users, a cohort of mothers reporting that their child has a birth defect. We used 22,999 annotated tweets to train and evaluate supervised machine learning algorithms—feature-engineered and deep learning-based classifiers—that automatically distinguish tweets referring to the user’s pregnancy outcome from tweets that merely mention birth defects. Because 90% of the tweets merely mention birth defects, we experimented with under-sampling and over-sampling approaches to address this class imbalance. An SVM classifier achieved the best performance for the two positive classes: an F1-score of 0.65 for the “defect” class and 0.51 for the “possible defect” class. We deployed the classifier on 20,457 unlabeled tweets that mention birth defects, which helped identify 542 additional users for potential inclusion in our cohort. Contributions of this study include (1) NLP methods for automatically detecting tweets by users reporting their birth defect outcomes, (2) findings that an SVM classifier can outperform a deep neural network-based classifier for highly imbalanced social media data, (3) evidence that automatic classification can be used to identify additional users for potential inclusion in our cohort, and (4) a publicly available corpus for training and evaluating supervised machine learning algorithms.

Dual deep neural network-based classifiers to detect experimental seizures.

Manually reviewing electroencephalograms (EEGs) is labor-intensive and demands automated seizure detection systems. To construct an efficient and robust event detector for experimental seizures from continuous EEG monitoring, we combined spectral analysis and deep neural networks. A deep neural network was trained to discriminate periodograms of 5-sec EEG segments from annotated convulsive seizures and the pre- and post-EEG segments. To use the entire EEG for training, a second network was trained with non-seizure EEGs that were misclassified as seizures by the first network. By sequentially applying the dual deep neural networks and simple pre- and post-processing, our autodetector identified all seizure events in 4,272 h of test EEG traces, with only 6 false positive events, corresponding to 100% sensitivity and 98% positive predictive value. Moreover, with pre-processing to reduce the computational burden, scanning and classifying 8,977 h of training and test EEG datasets took only 2.28 h with a personal computer. These results demonstrate that combining a basic feature extractor with dual deep neural networks and rule-based pre- and post-processing can detect convulsive seizures with great accuracy and low computational burden, highlighting the feasibility of our automated seizure detection algorithm.

Colonic MicroRNA Profiles, Identified by a Deep Learning Algorithm, That Predict Responses to Therapy of Patients With Acute Severe Ulcerative Colitis

Acute severe ulcerative colitis (ASUC) is a life-threatening condition managed with intravenous steroids followed by infliximab, cyclosporine, or colectomy (for patients with steroid resistance). There are no biomarkers to identify patients most likely to respond to therapy; ineffective medical treatment can delay colectomy and increase morbidity and mortality. We aimed to identify biomarkers of response to medical therapy for patients with ASUC. We performed a retrospective analysis of 47 patients with ASUC, well characterized for their responses to steroids, cyclosporine, or infliximab, therapy at 2 centers in France. Fixed colonic biopsies, collected before or within the first 3 days of treatment, were used for microarray analysis of microRNA expression profiles. Deep neural network-based classifiers were used to derive candidate biomarkers for discriminating responders from non-responders to each treatment and to predict which patients would require colectomy. Levels of identified microRNAs were then measured by quantitative PCR analysis in a validation cohort of 29 independent patients-the effectiveness of the classification algorithm was tested on this cohort. A deep neural network-based classifier identified 9 microRNAs plus 5 clinical factors, routinely recorded at time of hospital admission, that associated with responses of patients to treatment. This panel discriminated responders to steroids from non-responders with 93% accuracy (area under the curve, 0.91). We identified 3 algorithms, based on microRNA levels, that identified responders to infliximab vs non-responders (84% accuracy, AUC = 0.82) and responders to cyclosporine vs non-responders (80% accuracy, AUC = 0.79). We developed an algorithm that identifies patients with ASUC who respond vs do not respond to first- and second-line treatments, based on microRNA expression profiles in colon tissues.

Opening up the blackbox: an interpretable deep neural network-based classifier for cell-type specific enhancer predictions

BackgroundGene expression is mediated by specialized cis-regulatory modules (CRMs), the most prominent of which are called enhancers. Early experiments indicated that enhancers located far from the gene promoters are often responsible for mediating gene transcription. Knowing their properties, regulatory activity, and genomic targets is crucial to the functional understanding of cellular events, ranging from cellular homeostasis to differentiation. Recent genome-wide investigation of epigenomic marks has indicated that enhancer elements could be enriched for certain epigenomic marks, such as, combinatorial patterns of histone modifications.MethodsOur efforts in this paper are motivated by these recent advances in epigenomic profiling methods, which have uncovered enhancer-associated chromatin features in different cell types and organisms. Specifically, in this paper, we use recent state-of-the-art Deep Learning methods and develop a deep neural network (DNN)-based architecture, called EP-DNN, to predict the presence and types of enhancers in the human genome. It uses as features, the expression levels of the histone modifications at the peaks of the functional sites as well as in its adjacent regions. We apply EP-DNN to four different cell types: H1, IMR90, HepG2, and HeLa S3. We train EP-DNN using p300 binding sites as enhancers, and TSS and random non-DHS sites as non-enhancers. We perform EP-DNN predictions to quantify the validation rate for different levels of confidence in the predictions and also perform comparisons against two state-of-the-art computational models for enhancer predictions, DEEP-ENCODE and RFECS.ResultsWe find that EP-DNN has superior accuracy and takes less time to make predictions. Next, we develop methods to make EP-DNN interpretable by computing the importance of each input feature in the classification task. This analysis indicates that the important histone modifications were distinct for different cell types, with some overlaps, e.g., H3K27ac was important in cell type H1 but less so in HeLa S3, while H3K4me1 was relatively important in all four cell types. We finally use the feature importance analysis to reduce the number of input features needed to train the DNN, thus reducing training time, which is often the computational bottleneck in the use of a DNN.ConclusionsIn this paper, we developed EP-DNN, which has high accuracy of prediction, with validation rates above 90 % for the operational region of enhancer prediction for all four cell lines that we studied, outperforming DEEP-ENCODE and RFECS. Then, we developed a method to analyze a trained DNN and determine which histone modifications are important, and within that, which features proximal or distal to the enhancer site, are important.