Deep learning (DL) systems are widely used in domains including aircraft collision avoidance systems, Alzheimer's disease diagnosis, and autonomous driving cars. Despite the requirement for high reliability, DL systems are difficult to test. Existing DL testing work focuses on testing the DL models, not the implementations (e.g., DL software libraries) of the models. One key challenge of testing DL libraries is the difficulty of knowing the expected output of DL libraries given an input instance. Fortunately, there are multiple implementations of the same DL algorithms in different DL libraries. Thus, we propose CRADLE, a new approach that focuses on finding and localizing bugs in DL software libraries. CRADLE (1) performs cross-implementation inconsistency checking to detect bugs in DL libraries, and (2) leverages anomaly propagation tracking and analysis to localize faulty functions in DL libraries that cause the bugs. We evaluate CRADLE on three libraries (TensorFlow, CNTK, and Theano), 11 datasets (including ImageNet, MNIST, and KGS Go game), and 30 pre-trained models. CRADLE detects 12 bugs and 104 unique inconsistencies, and highlights functions relevant to the causes of inconsistencies for all 104 unique inconsistencies.

CRADLE: Cr oss-backend v a lidation to D etect and L ocalize bugs in D e ep learning libraries

In modern-day medicine, medical imaging has undergone immense advancements and can capture several biomedical images from patients. In the wake of this, to assist medical specialists, these images can be used and trained in an intelligent system in order to aid the determination of the different diseases that can be identified from analyzing these images. Classification plays an important role in this regard; it enhances the grouping of these images into categories of diseases and optimizes the next step of a computer-aided diagnosis system. The concept of classification in machine learning deals with the problem of identifying to which set of categories a new population belongs. When category membership is known, the classification is done on the basis of a training set of data containing observations. The goal of this paper is to perform a survey of classification algorithms for biomedical images. The paper then describes how these algorithms can be applied to a big data architecture by using the Spark framework. This paper further proposes the classification workflow based on the observed optimal algorithms, Support Vector Machine and Deep Learning as drawn from the literature. The algorithm for the feature extraction step during the classification process is presented and can be customized in all other steps of the proposed classification workflow.

/pdf/biomedical-image-classification-in-a-big-data-architecture-224n8jqxur.pdf

Biomedical Image Classification in a Big Data Architecture Using Machine Learning Algorithms.

Landslides are considered to be among the most alarming natural hazards. Therefore, there is a growing demand for databases and inventories of these events worldwide, since they are a vital resource for landslide risk assessment applications. Given the recent advances in the field of image processing, the objective of this study is to evaluate the performance of a deep convolutional neural network architecture called U-Net for the mapping of landslide scars from satellite imagery. The question that drives the study is: can fully convolutional neural networks be successfully applied as the backbone of automatic frameworks for building landslide inventories, keeping or improving the identification accuracy and agility when compared to other methods? To seek for an answer to it, scenes from the Landsat-8 satellite of a region of Nepal were obtained and processed in order to compose a landslide image database that served as the basis for the training, validation and test of deep convolutional neural networks. The U-Net architecture was applied and the results indicate that it has the potential to identify landslide scars, improving over previously published research on the topic for the same study region. The validation process resulted in recall, precision and F1-score values of 0.74, 0.61 and 0.67, respectively, thus higher than those from previous studies using different methodologies. The results indicate the potential of the method to be applied in dynamic mapping systems for landslide scar identification, which paves the way to the composition and updating of landslide scar databases. These, in turn, can support a great deal of quantitative landslide susceptibility mapping methods that heavily rely on data to provide accurate results.

Convolutional neural networks applied to semantic segmentation of landslide scars

Classification of remotely sensed data requires the modelling of suitable image processing algorithms. The rise of machine learning systems upgraded the viability of satellite image applications. Using Convolutional Neural Networks (CNN), benchmark classification exactness can be accomplished for land cover grouping. Motivated by the concept of Normalized Difference Vegetation Index (NDVI), this paper utilizes only the red and near infrared (NIR) band information for classifying the publicly available SAT-4 and SAT-6 datasets. This is done, since NDVI computation requires only the two band (red and NIR) information and the classes involved in the dataset are types of vegetation. In this work, new deep learning architectures for three different networks (AlexNet, ConvNet, VGG) were proposed by hypertuning the network and the input as two band data. The modified architectures with the two band information along with reduced number of filters were trained and tested model manages to classify the images into different classes. The proposed architectures are compared against the existing architectures in terms of accuracy, precision and trainable parameters. The proposed architecture is found to perform equally efficient by retaining high accuracy with less number of trainable parameters, when compared against the the performance of benchmark deep learning architectures for satellite image classification.

Deep learning architectures for land cover classification using red and near-infrared satellite images

https://iopscience.iop.org/article/10.1088/1755-1315/577/1/012012/pdf

Automation of the process of selecting hyperparameters for artificial neural networks for processing retrospective text information

Deep Learning(DL) techniques are conquering over the prevailing traditional approaches of neural network, when it comes to the huge amount of dataset, applications requiring complex functions demanding increase accuracy with lower time complexities. Neurosciences has already exploited DL techniques, thus portrayed itself as an inspirational source for researchers exploring the domain of Machine learning. DL enthusiasts cover the areas of vision, speech recognition, motion planning and NLP as well, moving back and forth among fields. This concerns with building models that can successfully solve variety of tasks requiring intelligence and distributed representation. The accessibility to faster CPUs, introduction of GPUs-performing complex vector and matrix computations, supported agile connectivity to network. Enhanced software infrastructures for distributed computing worked in strengthening the thought that made researchers suffice DL methodologies. The paper emphases on the following DL procedures to traditional approaches which are performed manually for classifying medical images. The medical images are used for the study Diabetic Retinopathy(DR) and computed tomography (CT) emphysema data. Both DR and CT data diagnosis is difficult task for normal image classification methods. The initial work was carried out with basic image processing along with K-means clustering for identification of image severity levels. After determining image severity levels ANN has been applied on the data to get the basic classification result, then it is compared with the result of DNNs (Deep Neural Networks), which performed efficiently because of its multiple hidden layer features basically which increases accuracy factors, but the problem of vanishing gradient in DNNs made to consider Convolution Neural Networks (CNNs) as well for better results. The CNNs are found to be providing better outcomes when compared to other learning models aimed at classification of images. CNNs are favoured as they provide better visual processing models successfully classifying the noisy data as well. The work centres on the detection on Diabetic Retinopathy-loss in vision and recognition of computed tomography (CT) emphysema data measuring the severity levels for both cases. The paper discovers how various Machine Learning algorithms can be implemented ensuing a supervised approach, so as to get accurate results with less complexity possible.

https://iopscience.iop.org/article/10.1088/1757-899X/263/4/042097/pdf

Shalva Rai

Papers

A comparative study of deep learning models for medical image classification