COVID-19 is an air-borne viral infection, which infects the respiratory system in the human body, and it became a global pandemic in early March 2020. The damage caused by the COVID-19 disease in a human lung region can be identified using Computed Tomography (CT) scans. We present a novel approach in classifying COVID-19 infection and normal patients using a Random Forest (RF) model to train on a combination of Deep Learning (DL) features and Radiomic texture features extracted from CT scans of patient’s lungs. We developed and trained DL models using CNN architectures for extracting DL features. The Radiomic texture features are calculated using CT scans and its associated infection masks. In this work, we claim that the RFs classification using the DL features in conjunction with Radiomic texture features enhances prediction performance. The experiment results show that our proposed models achieve a higher True Positive rate with the average Area Under the Receiver Curve (AUC) of 0.9768, 95% Confidence Interval (CI) [0.9757, 0.9780].

Adversarial attacks attempt to disrupt the training, retraining, and utilizing of artificial intelligence (AI) and machine learning models in large-scale distributed machine learning systems. This causes security risks on its prediction outcome. For example, attackers attempt to poison the model by either presenting inaccurate misrepresentative data or altering the models' parameters. In addition, Byzantine faults including software, hardware, network issues occur in distributed systems which also lead to a negative impact on the prediction outcome. In this paper, we propose a novel distributed training algorithm, partial synchronous stochastic gradient descent (ParSGD), which defends adversarial attacks and/or tolerates Byzantine faults. We demonstrate the effectiveness of our algorithm under three common adversarial attacks again the ML models and a Byzantine fault during the training phase. Our results show that using ParSGD, ML models can still produce accurate predictions as if it is not being attacked nor having failures at all when almost half of the nodes are being compromised or failed. We will report the experimental evaluations of ParSGD in comparison with other algorithms.

Deep Neural Networks are used to solve the most challenging world problems. In spite of the numerous advancements in the field, most of the models are being tuned manually. Experienced Data Scientists have to manually optimize hyperparameters, such as dropout rate, learning rate or number of neurons for Big Data applications. We have implemented a flexible automatic real-time hyperparameter tuning methodology. It works for arbitrary models written in Python and Keras. We also utilized state of the art Cloud services such as trigger based serverless computing (Lambda), and advanced GPU instances to implement automation, reliability and scalability.The existing tuning libraries, such as hyperopt, Scikit-Optimize or SageMaker, require developers to provide a list of hyperparameters and the range of their values manually. Our novel approach detects potential hyperparameters automatically from the source code, updates the original model to tune the parameters, runs the evaluation in the Cloud on spot instances, finds the optimal hyperparameters, and saves the results in the No-SQL database. The methodology can be applied to numerous Big Data Machine Learning systems.

COVID-19 is a novel infectious disease responsible for over 1.2 million deaths worldwide as of November 2020. The need for rapid testing is a high priority and alternative testing strategies including x-ray image classification are a promising area of research. However, at present, public datasets for COVID-19 x-ray images have low data volumes, making it challenging to develop accurate image classifiers. Several recent papers have made use of Generative Adversarial Networks (GANs) in order to increase the training data volumes. But realistic synthetic COVID-19 x-rays remain challenging to generate. We present a novel Mean Teacher + Transfer GAN (MTT-GAN) that generates COVID-19 chest x-ray images of high quality. In order to create a more accurate GAN, we employ transfer learning from the Kaggle pneumonia x-ray dataset, a highly relevant data source orders of magnitude larger than public COVID-19 datasets. Furthermore, we employ the Mean Teacher algorithm as a constraint to improve stability of training. Our qualitative analysis shows that the MTT-GAN generates x-ray images that are greatly superior to a baseline GAN and visually comparable to real x-rays. Although board-certified radiologists can distinguish MTT-GAN fakes from real COVID-19 x-rays, quantitative analysis shows that MTT-GAN greatly improves the accuracy of both a binary COVID-19 classifier as well as a multi-class pneumonia classifier as compared to a baseline GAN. Our classification accuracy is favorable as compared to recently reported results in the literature for similar binary and multi-class COVID-19 screening tasks.

While adopting Blockchain technologies to automate their enterprise functionality, organizations are recognizing the challenges of scalability and manual configuration that the state of art present. Scalability of Hyperledger Fabric is an open challenge recognized by the research community. We have automated many of the configuration steps of installing Hyperledger Fabric Blockchain on AWS infrastructure and have benchmarked the scalability of that system. We have used the UCR (University of California Riverside) Time Series Archive with 128 timeseries datasets containing over 191,177 rows of data totaling 76,453,742 numbers. Using an automated Serverless approach, we have loaded this dataset, by chunks, into different AWS instances, triggering the load by SQS messaging. In this paper, we present the results of this benchmarking study and describe the approach we took to automate the Hyperledger Fabric processes using serverless Lambda functions and SQS triggering. We will also discuss what is needed to make the Blockchain technology more robust and scalable.

Saving nature using Big Data Analytics is a very noble goal. Using New York taxi rides data, we decided to learn how many rides could be consolidated. It was a journey we would like to share. First, we had to choose the platform for calculation between Amazon Athena, Serverless Microservices, SQL or NoSql databases, Hadoop and Spark. Then, we had to find an optimal solution for the platform using assorted tuning and optimization techniques. Although the problem seems to be straight forward, it turned out that the solution is quite challenging because of the input size, data quality, calculation complexities and numerous EMR/Spark tuning options. We have been using New York taxi data from 2009 to 2017 to quantify the rides that can be joined together. The taxi rides were consolidated based on pickup location, pickup time and drop-off location. We have been calculating the percentage of taxi rides that can be joined. The benchmark originally set was rides within five minutes with a pickup and drop-off locations within half a kilometer. Then we started experimenting with different times and locations. We have been using parquet format, parallel Scala collections, compression, filtering, new column introduction, tuning parameters, I/O overhead tuning, bucketing, timeouts and partitioning. Over 1.2 billion rides were processed using Amazon EMR with Spark. We have been optimizing calculation time and processing price. Spark has hundreds of parameters, EMR has over fifty instances to choose from. It was challenging to process our data within reasonable time. We were able to find the optimal Spark queries (plans), tested different types of joins and compared their performances. Also, we were able to compare I/O and in-memory operations during partitioning and large files manipulation (the input file sizes were hundreds of Gigabytes). The results were amazing - we could consolidate around thirty five percent of total rides, saving tons of gas and improving environment and traffic in New York City.

Machine Learning algorithms on large datasets can be executed in the Cloud. Amazon Web Services (AWS) provides over 60 different On-Demand EC2 instances [1]. The instance prices range from $0.0059 (t2.nano) to $14.4 (p2.16xlarge) per hour. We decided to build an automatic recommendation system to choose the best instance for a dataset and a machine learning algorithm to optimize time and money spent. After running multiple algorithms for different Big Data sets on assorted AWS instances and collecting the results in the NoSQL DynamoDB database, we have trained machine learning models to predict time and cost using assorted regression ML methods.

Machine learning and graph analytics typically process data in an iterative way, reading the same data multiple times and sharing intermediate results across the worker nodes in cluster. Hadoop MapReduce and Spark are two popular open source cluster compute frameworks for large scale data analytics. Apache Spark is currently the state-of-the-art in-memory computation model extending MapReduce by transforming data into RDDs stored in memory. One limitation of Spark, however, lies in the fact that data transformation and distribution is implicitly managed by HDFS. Data locality is not guaranteed for iterative machine learning algorithms which read the same data multiple times. For example, data needed for operations to one worker node might reside in RDDs stored in other worker nodes. The resulting data shuffling becomes a bottleneck when iteratively reading such RDDs. We propose YinMem, a parallel distributed indexed in-memory computation system, bridging the gap between Hadoop ecosystem and HPC by replacing MapReduce with MPI while obtaining the advantage of the distributed data storage. YinMem achieves fair load balancing prior to computation for large sparse matrix by scheduling and distributing indexed data from NoSQL database to the RAM of working nodes. YinMem explores Alluxio as the in-memory storage system and enables efficient data sharing of intermediate results. Preliminary results show that YinMem has achieved 3× speedup to Spark, for computing eigenvalue and eigenvectors of a 16-million scale sparse matrix.

We propose a novel iterative unified clustering algorithm for data with both continuous and categorical variables, in the big data environment. Clustering is a well-studied problem and finds several applications. However, none of the big data clustering works discuss the challenge of mixed attribute datasets, with both categorical and continuous attributes. We study an application in the health care domain namely Case Based Reasoning (CBR), which refers to solving new problems based on solutions to similar past problems. This is particularly useful when there is a large set of clinical records with several types of attributes, from which similar patients need to be identified. We go one step further and include the genomic components of patient records to enhance the CBR discovery. Thus, our contributions in this paper spans across the big data algorithmic research and a key contribution to the domain of heath care information technology research. First, our clustering algorithm deals with both continuous and categorical variables in the data; second, our clustering algorithm is iterative where it finds the clusters which are not well formed and iteratively drills down to form well defined clusters at the end of the process; third we provide a novel approach to CBR across clinical and genomic data. Our research has implications for clinical trials and facilitating precision diagnostics in large and heterogeneous patient records. We present extensive experimental results to show the efficacy or our approach.

In spite of significant progress in the area of data management and integration, heterogeneous nature of clinical data makes it challenging to develop a unified view of clinical data. Therefore, a central question we are trying to address is how we can utilize data analytics to discover insightful knowledge from the scattered & large amount of clinical data to simplify clinical decision making. We propose a Unified Framework for Clinical Data Analytics (U-CDA) for mining large amounts of heterogeneous data to build enhanced clinical data analytics system. The proposed framework (U-CDA) in this paper integrates relevant clinical data from structured and unstructured data sources such as electronic health records, legacy health information system databases, clinical notes, public registries, and genomic datasets after applying necessary cleansing and transformations. It further uses intelligent, versatile data analytics engine to analyze clinical data.

Big Data deals with enormous volumes of complex and exponentially growing data sets from multiple sources. With rapid growth in technology, we are now able to generate immense amount of data in almost any field imaginable including physical, biological and biomedical sciences. With the diversity and amount of data in health care industry there is an increasing need to evaluate the components in big data frameworks and gauge their adaptability to analytics techniques. However, a key step in adapting big data tools is the portability of relational databases to big data environment. Since SQL is considered to be the de-facto language for interactive queries, in this paper, we evaluate the performance of SQL-like big data solutions for the portability of existing relational databases. Our work focuses on benchmarking multiple SQL-like big data technologies over Hadoop based distributed file system (HDFS) for Study Data Tabulation Model (SDTM) used in clinical trial databases for improving the efficiency of research in clinical trials. We use publically available clinical trial data (from National Institute on Drug Abuse (NIDA)), which follows SDTM, as a test bed to measure key parameters like usability, adaptability, modularity, robustness and efficiency of these solutions. With the intention to demonstrate how current clinical trial functionality can be replicated on a big data backend with high SQL-like functionality, we evaluate several types of ad-hoc SQL queries.

NoSQL distributed databases have been devised to tackle the challenges resulting from volume, velocity and variety of big data. Graph representation of datasets requires efficient distributed linear algebra operations for large sparse matrix constructed from big data. Storing the transformed matrix into the database not only speeds up the big data analysis process but also facilitates the computation because of indexing. The Hadoop based approach does not natively support iterative algorithms due to data shuffling during each iteration. This paper presents a novel database-based distributed computation architecture bridging the gap between Hadoop and HPC. The novelty results from exploring the indexing capability of D4M (Dynamic Distributed Dimensional Data Model) to support linear algebra operations in a distributed computation environment. The idea is to store input data and intermediate results in associative array format inside Accumulo table to facilitate the data sharing among working nodes. pMatlab is deployed as the parallel computation engine. Our proposed architecture is proved to be lighter, easier and faster than MapReduce based approach. One example application is calculating top k eigenvalues and eigenvectors for large sparse matrix. Experiments on Graph500 benchmark datasets demonstrate 2X speedup of our architecture as compared to HEIGEN (An eigensolver for billion-scale matrices using MapReduce).

Pointwise mutual information (PMI) is a widely used word similarity measure, but it lacks a clear explanation of how it works. We explore how PMI differs from distributional similarity, and we introduce a novel metric, PMImax, that augments PMI with information about a word's number of senses. The coefficients of PMImaxare determined empirically by maximizing a utility function based on the performance of automatic thesaurus generation. We show that it outperforms traditional PMI in the application of automatic thesaurus generation and in two word similarity benchmark tasks: human similarity ratings and TOEFL synonym questions. PMImaxachieves a correlation coefficient comparable to the best knowledge-based approaches on the Miller-Charles similarity rating data set.

This paper presents a framework to address new data management challenges introduced by data-intensive, pervasive computing environments. These challenges include a spatio-temporal variation of data and data source availability, lack of a global catalog and schema, and no guarantee of reconnection among peers due to the serendipitous nature of the environment. An important aspect of our solution is to treat devices as semi-autonomous peers guided in their interactions by profiles and context. The profiles are grounded in a semantically rich language and represent information about users, devices and data described in terms of “beliefs”, “desires”, and “intentions”. We present a prototype implementation of this framework over combined Bluetooth and Ad-Hoc 802.11 networks, and present experimental and simulation results that validate our approach and measure system performance.

Electronic commerce is emerging as one of the major Websupported applications requiring database support. We introduce and study high-level declarative specifications of business models, using an approach in the spirit of active databases. More precisely, business models are specified as relational transducers that map sequences of input relations into sequences of output relations. The semantically meaningful trace of an input-output exchange is kept as a sequence of log relations. We consider problems motivated by electronic commerce applications, such as log validation, verifying temporal properties of transducers, and comparing two relational transducers. Positive results are obtained for a restricted class of relational transducers called Spocus transducers (for semi-positive outputs and cumulative state). We argue that despite the restrictions, these capture a wide range of practically significant business models. 1 Introduction Electronic commerce is emerging as a major Web-s...

As we embark on the information age the use of electronic information is spreading through all sectors of society, both nationally and internationally. As a result, commercial organizations, educational institutions and government agencies are finding it essential to be linked by world wide networks, and commercial Internet usage is growing at an accelerating pace.

One of the features that distinguishes digital libraries from traditional databases is new cost models for client access to intellectual property. Clients will pay for accessing data items in digital libraries, and we believe that optimizing these costs will be as important as optimizing performance in traditional databases. In this article we discuss cost models and protocols for accessing digital libraries, with the objective of determining the minimum cost protocol for each model. We expect that in the future information appliances will come equipped with a cost optimizer, in the same way that computers today come with a built-in operating system. This article makes the initial steps towards a thery and practice of intellectual property cost management.

Many complex and dynamic database applications such as product modeling and negotiation monitoring require a number of features that have been adopted in semantic models and databases such as active rules, constraints, inheritance, etc. Unfortunately, each feature has largely been considered in isolation. Furthermore, in a commercial negotiation, participants staking their financial well beings will never accept a system they cannot gain a precise behavioral understanding of. We attack these problems with a rich and extensible database model, evolving databases, with a clear and precise semantics based on evolving algebras (E. Borger, 1994). We also briefly describe a prototype implementation of the model (B. Fordham et al.).

The computer science community is increasingly focusing its efforts on tasks related to the National Information Infrastructure. Hardware and software advances are being sought to make wide-area networks and technological resources in general more useful to mainstream society. With this transition comes a set of unavoidable political and social issues that have never been satisfactorily dealt with in the past. Grand challenges face the computer science community on both the technical and the socio-political sides of this “major upgrade.” This paper discusses various aspects of the enormous coordination problem that faces all of us and considers what the database community can do to help.

In a large information system, the amount of cache space available to store read-only replicas of data may be limited. Since acquiring these data from their sources may be an expensive and time-consuming operation, it is essential to make efficient use of the available cache space. This cache space may be unevenly distributed over a large number of loosely coupled sites. An intelligent caching strategy is needed to insure that replicas are created often enough that they can be inexpensively reached when necessary, but not so often that important data are forced out to make room. We present such a strategy, which we have developed for use in ALIBI, a networked resource discovery and information retrieval system. The TCF Strategy, as it is called, allows individual sites to adjust their level of cache turnover to provide better overall performance. This novel approach could no doubt be beneficially applied in other distributed systems which use caching. We include discussion and simulation results supporting the efficiency of the TCF Strategy.

Considers a distributed database with partial replication of data objects located on a ring network. Certain placements of replicated objects are shown to optimize the probability of read-only success and the probability of write-only success. We also obtain optimal placements for k-terminal reliability and expected minimal path length for read-only and write-only operations.>

Article Properties of networked information retrieval with ALIBI Share on Authors: David W. Flater University of Maryland, Baltimore County, Baltimore, MD University of Maryland, Baltimore County, Baltimore, MDView Profile , Yelena Yesha University of Maryland, Baltimore County, Baltimore, MD University of Maryland, Baltimore County, Baltimore, MDView Profile Authors Info & Claims CIKM '93: Proceedings of the second international conference on Information and knowledge managementDecember 1993 Pages 31–38https://doi.org/10.1145/170088.170098Online:01 December 1993Publication History 0citation199DownloadsMetricsTotal Citations0Total Downloads199Last 12 Months1Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

When dealing with massive amounts of primarily read-only data, significant improvements can be made over distributed DBMS for making these data available to a large network. This paper outlines some methods for heuristic query routing and cooperative caching which manage read-only replicas of data in a fully distributed manner on a network of arbitrary topology. These methods insure that query throughout increases steadily as nodes are added to the network while maintaining good response time. The resulting system is capable of providing automatic resource discovery and information retrieval over a wide area network without relying on resource directories.