We present a progressive meta-algorithm for single image super resolution of very large and seamless images. Deep single image super resolution backbone networks are capable of upsampling local image patches but are susceptible to severe tiling artifacts when attempting to integrate the patches together into a single large image. For large images such as 4K images, the individual low-resolution patches also comprise a narrow receptive field that lacks context regarding the pixel information from the surrounding patches. Our Progressive meta-algorithm is inspired by prior works in progressive GANs and is designed to resolve inter-patch tiling artifact across varying scales while further incorporating context by expanding the receptive field to include the surrounding patches through multiple rounds of progressive upsampling. Our meta-algorithm is compatible with standard super-resolving backbones and enables super-resolution of very large images such as 4K images while overcoming practical GPU memory limitations with commodity graphics cards. We evaluate our Meta-Algorithm with the challenging task of 16x super resolution for 4096×4096 images using the ESRGAN super-resolving backbone and demonstrate significantly improved image quality versus a baseline patch-based approach as evaluated using the Learned Perceptual Image Patch Similarity (LPIPS) metric.

Machine learning models have been widely used in security applications. However, it is well-known that adversaries can adapt their attacks to evade detection. There has been some work on making machine learning models more robust to such attacks. However, one simple but promising approach calledrandomizationis under-explored. In addition, most existing works focus on models with differentiable error functions while tree-based models do not have such error functions but are quite popular because they are easy to interpret. This paper proposes a novel randomization-based approach to improve robustness of tree-based models against evasion attacks. The proposed approach incorporates randomization into both model training time and model application time (meaning when the model is used to detect attacks). We also apply this approach to random forest, an existing ML method which already has incorporated randomness at training time but still often fails to generate robust models. We proposed a novel weighted-random-forest method to generate more robust models and a clustering method to add randomness at model application time. We also proposed a theoretical framework to provide a lower bound for adversaries’ effort. Experiments on intrusion detection and spam filtering data show that our approach further improves robustness of random-forest method.

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].

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.

Dust plumes originating from the Earth's major arid and semi-arid areas can significantly affect the climate system and human health. Many existing methods have been developed to identify dust from non-dust pixels from a remote sensing point of view. However, these methods use empirical rules and therefore have difficulty detecting dust above or below the detectable thresholds. Supervised machine learning methods have also been applied to detect dust from satellite imagery, but these methods are limited especially when applying to areas outside the training data due to the inadequate amount of ground truth data. In this work, we proposed an automatic dust segmentation framework using semi-supervised machine learning, based on a collocated dataset using Visible Infrared Imaging Radiometer Suite (VIIRS) and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO). The proposed method utilizes unsupervised machine learning for segmentation of VIIRS imagery, and leverages the guidance from the dust labels using the dust profile product of CALIPSO to determine the dust clusters as the final product. The dust clusters are determined based on the similarity of spectral signature from dust pixels along the CALIPSO tracks. Experiment results show that the accuracy of the proposed framework outperforms the traditional physical infrared method along CALIPSO tracks. In addition, the proposed method performs consistently over three different study areas, the North Atlantic Ocean, East Asia, and Northern Africa.

The risk adjusted mortality rate, which is also called standardized mortality ratio (SMR), is one widely used quality measure to evaluate healthcare provider performance. Logistic regression and decision tree are two traditional risk models for mortality rate calculation. Though some machine learning based approaches could achieve higher accuracy, they are hard to interpret and may have poor calibration scores. In this paper, we evaluated multiple machine learning approaches with different formats of longitudinal data, and proposed a hybrid approach based on long short-term memory (LSTM) model and decision tree. The new hybrid method provides a comparable area under the receiver operating characteristic curve (AUC) performance as LSTM with a better calibration score. Using a set of 3,473 patients with 10 months of data from historical, large scale ESRD patient data, the LSTM with long format data approach achieved AUC for prediction of mortality of 0.772 compared to 0.758 for logistic regression and 0.726 for a decision tree model. The hybrid approach could reach 0.783, a little higher than both LSTM and decision tree model. The hybrid approach has the best calibration performance based on the Hosmer Lemeshow test.

The recent years have seen a surge in the implementation of electronic health care records. These patient records contain valuable medical information including patient information, diagnosis, treatment methods, and eventual patient outcomes. It is important to analyze patterns within these records in order to more efficiently treat individuals. In this paper, we present a method for automatically discovering underlying themes and patterns within patient data. Our methodology includes the creation of the main themes or patterns in the data and linking the themes back to the corpus from which they were generated. In our research, we partitioned graphs from terms gathered from electronic health records. Modularity was used as the quality function, which strives to measure how well a given partition of a network compartmentalizes its communities. We have compared our method with probabilistic topic modeling algorithms, specifically LDA (Latent Dirichlet Allocation). Finally, recall and precision measures were used to evaluate the validity of our final results.

Data from sensors and sensor networks are being collected at astronomical rates. This results in a massive dataset that is increasingly difficult to navigate to find interesting time periods where the spatial pattern of a process changes. The ability to navigate to such areas can lead to new knowledge about the factors that contribute to a spatio-temporal process. This paper proposes a method to automatically characterize sensor datasets based on a measure of spatial change over time resulting in a set of multi-granular spatio-temporal intervals. The resulting intervals can be used to focus knowledge discovery tasks at multiple temporal granularities within the dataset. Furthermore, the intervals enable a drill-down-style analysis where events of varying magnitudes can be identified within each granularity. Experiments were performed on a real-world dataset measuring NEXRAD precipitation accumulation. The results show that the multi-granular spatio-temporal intervals identify interesting time periods in the dataset as evidenced by naturally occurring events.

Temporal data, which is a sequence of data tuples measured at successive time instances, is typically very large. Hence instead of mining the entire data, we are interested in dividing the huge data into several smaller intervals of interest which we call temporal neighborhoods. In this paper we propose an approach to generate temporal neighborhoods through unequal depth discretization. We describe two novel algorithms (a) similarity based merging (SMerg) and, (b) stationary distribution based merging (StMerg). These algorithms are based on the robust framework of Markov models and the Markov stationary distribution respectively. We identify temporal neighborhoods with distinct demarcations based on unequal depth discretization of the data. We discuss detailed experimental results in both synthetic and real world data. Specifically we show (i) the efficacy of our approach through precision and recall of labeled bins, (ii) the ground truth validation in real world datasets and, (iii) knowledge discovery in the temporal neighborhoods such as global anomalies. Our results indicate that we are able to identify valuable knowledge based on our ground truth validation from real world traffic data.

The focus of this paper is the discovery of anomalous spatio-temporal windows. We propose a Discretized Spatio-Temporal Scan Window approach to address the question of how we can treat Space and Time together without compromising on the properties of each and their impact on each other. In doing so we discover anomalous Spatio-Temporal windows, identify at what point in time the window changes, identify the spatial patterns of change over time and identify a spatial extent in time which is completely deviant with respect to the rest of the anomalous spatio-temporal windows. None of the current approaches address all these issues in combination. Subsequently we perform experiments on several real world datasets to validate our approach while comparing with the established approach of discovering a cylindrical spatio-temporal Scan window.

Environmental research and knowledge discovery both require extensive use of data stored in various sources and created in different ways for diverse purposes. We describe a new metadata approach to elicit semantic information from environmental data and implement semantic-based techniques to assist users in integrating, navigating, and mining multiple environmental data sources. Our system contains specifications of various environmental data sources and the relationships that are formed among them. User requests are augmented with semantically related data sources and automatically presented as a visual semantic network. In addition, we present a methodology for data navigation and pattern discovery using multi-resolution browsing and data mining. The data semantics are captured and utilized in terms of their patterns and trends at multiple levels of resolution. We present the efficacy of our methodology through experimental results.

In this paper we study the behavior and performance of bilingual users in using an electronic catalog. The purpose of this research is to further the knowledge required for building electronic commerce systems that operate in multiple languages in global settings. We describe a bilingual electronic catalog that can be used by online retailers for selling products and/or services to customers interacting in either English or Chinese. We investigate into the nature of user interactions in multilingual electronic catalogs. We have defined three different groups of users: only Chinese speaking, only English speaking, and bilingual. We are specifically interested in investigating into the language preferences of the third group of users. In order to test language preferences, we have selected two types of products: office supplies and ethnic food. We hypothesize that bilingual users will exhibit differential language preferences for the type of products and the tasks performed in using the electronic catalog. Furthermore, learning curves and interaction effects are also tested. Three different task categories have been designed: browsing, directed search, and exact matches. In the first case, the user is a general browser who is looking for what is available in the catalog. In the second case, the user is looking for a class of products but is unsure of the exact item. In the third case the user knows exactly what item he/she is looking for. We propose to test the efficiency of usage by measuring the time as well as studying the path followed by the user in retrieving product information. This research will shed light on the important issue of designing multilingual electronic catalogs for both local and global applications.

The paper presents a methodology for developing a user interface that combines fourth generation interface tools (SQL forms) with a natural language processor for a database management system. The natural language processor consists of an index, a lexicon and a parser. The index is used to uniquely identify each form in the system through a conceptual representation of its purpose. The form fields specify database or nondatabase fields whose values are either entered by the user (user-defined) or are derived by the form (system-defined) in response to user input. A set of grammar rules are associated with each form. The lexicon consists of all words recognized by the system, their grammatical categories, roots, their associations (if any) with database objects and forms. The parser scans, a natural language query to identify a form in a bottom-up fashion. The information requested in the user query is determined in a top-down manner by parsing, through the grammar rules associated with the identified form. Extragrammatical inputs with limited deviations from the grammar rules are supported. Combining a natural language processor with SQL forms allows processing data modification tasks without violating any database integrity constraint, having duplicate records, or entering invalid data. A prototype natural language interface is described as a front-end to an ORACLE database for a computer integrated manufacturing system.

A structured methodology for linking data modeling with functional modeling in a computer integrated manufacturing system is presented. The target application, the functional and data models, and a method for developing the data model starting from the functional model are described. This approach ensures that the data model is complete and non-redundant with respect to the functional model. A scheme that enables various functions in the functional model to be linked with the data elements of the data model is also presented. Such a linkage makes it possible to determine the impact of a change in the functional model on the data model and vice versa.>

This paper deals with query processing using semantic knowledge in relational databases. The Select-Project-Join (SPJ) conjunctive class of queries are dealt with in this paper. We propose to optimize highly repetitive queries by using semantic transformations in addition to syntactic transformations. Thus, we generate a set of pre-optimized queries. This set contains queries that are semantically equivalent to, syntactically different from, and more efficient to process than the user queries that we started with. The issues we address in this paper are: how to map a user query to a query that is in the set of pre-optimized and already optimized queries, how to search efficiently through the set of pre-optimized queries and set of semantic rules, and how to incorporate new queries to the set of pre-optimized queries, so that the number of queries that can be optimized using this method increases with the passage of time. Furthermore, we suggest some ideas of handling queries that do not have any semantically equivalent counterpart in the set of pre-optimized queries. We have tested the performance of the proposed method. An algorithm for mapping is implemented in Prolog. A database schema is implemented in the INGRES database management system. We have adopted a database schema that is widely used for measuring performance in the semantic query optimization literature.